Skip-gram word2vec

In this notebook, I'll lead you through using TensorFlow to implement the word2vec algorithm using the skip-gram architecture. By implementing this, you'll learn about embedding words for use in natural language processing. This will come in handy when dealing with things like machine translation.

Readings

Here are the resources I used to build this notebook. I suggest reading these either beforehand or while you're working on this material.

A really good conceptual overview of word2vec from Chris McCormick
First word2vec paper from Mikolov et al.
NIPS paper with improvements for word2vec also from Mikolov et al.
An implementation of word2vec from Thushan Ganegedara
TensorFlow word2vec tutorial

Word embeddings

When you're dealing with words in text, you end up with tens of thousands of classes to predict, one for each word. Trying to one-hot encode these words is massively inefficient, you'll have one element set to 1 and the other 50,000 set to 0. The matrix multiplication going into the first hidden layer will have almost all of the resulting values be zero. This a huge waste of computation.

To solve this problem and greatly increase the efficiency of our networks, we use what are called embeddings. Embeddings are just a fully connected layer like you've seen before. We call this layer the embedding layer and the weights are embedding weights. We skip the multiplication into the embedding layer by instead directly grabbing the hidden layer values from the weight matrix. We can do this because the multiplication of a one-hot encoded vector with a matrix returns the row of the matrix corresponding the index of the "on" input unit.

Instead of doing the matrix multiplication, we use the weight matrix as a lookup table. We encode the words as integers, for example "heart" is encoded as 958, "mind" as 18094. Then to get hidden layer values for "heart", you just take the 958th row of the embedding matrix. This process is called an embedding lookup and the number of hidden units is the embedding dimension.

There is nothing magical going on here. The embedding lookup table is just a weight matrix. The embedding layer is just a hidden layer. The lookup is just a shortcut for the matrix multiplication. The lookup table is trained just like any weight matrix as well.

Embeddings aren't only used for words of course. You can use them for any model where you have a massive number of classes. A particular type of model called Word2Vec uses the embedding layer to find vector representations of words that contain semantic meaning.

Word2Vec

The word2vec algorithm finds much more efficient representations by finding vectors that represent the words. These vectors also contain semantic information about the words. Words that show up in similar contexts, such as "black", "white", and "red" will have vectors near each other. There are two architectures for implementing word2vec, CBOW (Continuous Bag-Of-Words) and Skip-gram.

In this implementation, we'll be using the skip-gram architecture because it performs better than CBOW. Here, we pass in a word and try to predict the words surrounding it in the text. In this way, we can train the network to learn representations for words that show up in similar contexts.

First up, importing packages.



In [1]:

    
import time

import numpy as np
import tensorflow as tf

import utils

Load the text8 dataset, a file of cleaned up Wikipedia articles from Matt Mahoney. The next cell will download the data set to the data folder. Then you can extract it and delete the archive file to save storage space.



In [2]:

    
from urllib.request import urlretrieve
from os.path import isfile, isdir
from tqdm import tqdm
import zipfile

dataset_folder_path = 'data'
dataset_filename = 'text8.zip'
dataset_name = 'Text8 Dataset'

class DLProgress(tqdm):
    last_block = 0

    def hook(self, block_num=1, block_size=1, total_size=None):
        self.total = total_size
        self.update((block_num - self.last_block) * block_size)
        self.last_block = block_num

if not isfile(dataset_filename):
    with DLProgress(unit='B', unit_scale=True, miniters=1, desc=dataset_name) as pbar:
        urlretrieve(
            'http://mattmahoney.net/dc/text8.zip',
            dataset_filename,
            pbar.hook)

if not isdir(dataset_folder_path):
    with zipfile.ZipFile(dataset_filename) as zip_ref:
        zip_ref.extractall(dataset_folder_path)
        
with open('data/text8') as f:
    text = f.read()









    



Text8 Dataset: 31.4MB [01:08, 458KB/s]

Preprocessing

Here I'm fixing up the text to make training easier. This comes from the utils module I wrote. The preprocess function coverts any punctuation into tokens, so a period is changed to <PERIOD>. In this data set, there aren't any periods, but it will help in other NLP problems. I'm also removing all words that show up five or fewer times in the dataset. This will greatly reduce issues due to noise in the data and improve the quality of the vector representations. If you want to write your own functions for this stuff, go for it.



In [3]:

    
words = utils.preprocess(text)
print(words[:30])









    



['anarchism', 'originated', 'as', 'a', 'term', 'of', 'abuse', 'first', 'used', 'against', 'early', 'working', 'class', 'radicals', 'including', 'the', 'diggers', 'of', 'the', 'english', 'revolution', 'and', 'the', 'sans', 'culottes', 'of', 'the', 'french', 'revolution', 'whilst']



In [4]:

    
print("Total words: {}".format(len(words)))
print("Unique words: {}".format(len(set(words))))









    



Total words: 16680599
Unique words: 63641

And here I'm creating dictionaries to covert words to integers and backwards, integers to words. The integers are assigned in descending frequency order, so the most frequent word ("the") is given the integer 0 and the next most frequent is 1 and so on. The words are converted to integers and stored in the list int_words.



In [6]:

    
vocab_to_int, int_to_vocab = utils.create_lookup_tables(words)
int_words = [vocab_to_int[word] for word in words]

Subsampling

Words that show up often such as "the", "of", and "for" don't provide much context to the nearby words. If we discard some of them, we can remove some of the noise from our data and in return get faster training and better representations. This process is called subsampling by Mikolov. For each word $w_i$ in the training set, we'll discard it with probability given by

$$ P(w_i) = 1 - \sqrt{\frac{t}{f(w_i)}} $$

where $t$ is a threshold parameter and $f(w_i)$ is the frequency of word $w_i$ in the total dataset.

I'm going to leave this up to you as an exercise. This is more of a programming challenge, than about deep learning specifically. But, being able to prepare your data for your network is an important skill to have. Check out my solution to see how I did it.

Exercise: Implement subsampling for the words in int_words. That is, go through int_words and discard each word given the probablility $P(w_i)$ shown above. Note that $P(w_i)$ is the probability that a word is discarded. Assign the subsampled data to train_words.



In [7]:

    
from collections import Counter
import random

threshold = 1e-5
word_counts = Counter(int_words)
total_count = len(int_words)
freqs = {word: count/total_count for word, count in word_counts.items()}
p_drop = {word: 1 - np.sqrt(threshold/freqs[word]) for word in word_counts}
train_words = [word for word in int_words if p_drop[word] < random.random()]

Making batches

Now that our data is in good shape, we need to get it into the proper form to pass it into our network. With the skip-gram architecture, for each word in the text, we want to grab all the words in a window around that word, with size $C$.

From Mikolov et al.:

"Since the more distant words are usually less related to the current word than those close to it, we give less weight to the distant words by sampling less from those words in our training examples... If we choose $C = 5$, for each training word we will select randomly a number $R$ in range $< 1; C >$, and then use $R$ words from history and $R$ words from the future of the current word as correct labels."

Exercise: Implement a function get_target that receives a list of words, an index, and a window size, then returns a list of words in the window around the index. Make sure to use the algorithm described above, where you choose a random number of words from the window.



In [9]:

    
def get_target(words, idx, window_size=5):
    ''' Get a list of words in a window around an index. '''
    
    R = np.random.randint(1, window_size+1)
    start = idx - R if (idx - R) > 0 else 0
    stop = idx + R
    target_words = set(words[start:idx] + words[idx+1:stop+1])
    
    return list(target_words)

Here's a function that returns batches for our network. The idea is that it grabs batch_size words from a words list. Then for each of those words, it gets the target words in the window. I haven't found a way to pass in a random number of target words and get it to work with the architecture, so I make one row per input-target pair. This is a generator function by the way, helps save memory.



In [10]:

    
def get_batches(words, batch_size, window_size=5):
    ''' Create a generator of word batches as a tuple (inputs, targets) '''
    
    n_batches = len(words)//batch_size
    
    # only full batches
    words = words[:n_batches*batch_size]
    
    for idx in range(0, len(words), batch_size):
        x, y = [], []
        batch = words[idx:idx+batch_size]
        for ii in range(len(batch)):
            batch_x = batch[ii]
            batch_y = get_target(batch, ii, window_size)
            y.extend(batch_y)
            x.extend([batch_x]*len(batch_y))
        yield x, y

Building the graph

From Chris McCormick's blog, we can see the general structure of our network.

The input words are passed in as integers. This will go into a hidden layer of linear units, then into a softmax layer. We'll use the softmax layer to make a prediction like normal.

The idea here is to train the hidden layer weight matrix to find efficient representations for our words. We can discard the softmax layer becuase we don't really care about making predictions with this network. We just want the embedding matrix so we can use it in other networks we build from the dataset.

I'm going to have you build the graph in stages now. First off, creating the inputs and labels placeholders like normal.

Exercise: Assign inputs and labels using tf.placeholder. We're going to be passing in integers, so set the data types to tf.int32. The batches we're passing in will have varying sizes, so set the batch sizes to [None]. To make things work later, you'll need to set the second dimension of labels to None or 1.



In [11]:

    
train_graph = tf.Graph()
with train_graph.as_default():
    inputs = tf.placeholder(tf.int32, [None], name='inputs')
    labels = tf.placeholder(tf.int32, [None, None], name='labels')

Embedding

The embedding matrix has a size of the number of words by the number of units in the hidden layer. So, if you have 10,000 words and 300 hidden units, the matrix will have size $10,000 \times 300$. Remember that we're using tokenized data for our inputs, usually as integers, where the number of tokens is the number of words in our vocabulary.

Exercise: Tensorflow provides a convenient function tf.nn.embedding_lookup that does this lookup for us. You pass in the embedding matrix and a tensor of integers, then it returns rows in the matrix corresponding to those integers. Below, set the number of embedding features you'll use (200 is a good start), create the embedding matrix variable, and use tf.nn.embedding_lookup to get the embedding tensors. For the embedding matrix, I suggest you initialize it with a uniform random numbers between -1 and 1 using tf.random_uniform.



In [12]:

    
n_vocab = len(int_to_vocab)
n_embedding =  200
with train_graph.as_default():
    embedding = tf.Variable(tf.random_uniform((n_vocab, n_embedding), -1, 1))
    embed = tf.nn.embedding_lookup(embedding, inputs)

Negative sampling

For every example we give the network, we train it using the output from the softmax layer. That means for each input, we're making very small changes to millions of weights even though we only have one true example. This makes training the network very inefficient. We can approximate the loss from the softmax layer by only updating a small subset of all the weights at once. We'll update the weights for the correct label, but only a small number of incorrect labels. This is called "negative sampling". Tensorflow has a convenient function to do this, tf.nn.sampled_softmax_loss.

Exercise: Below, create weights and biases for the softmax layer. Then, use tf.nn.sampled_softmax_loss to calculate the loss. Be sure to read the documentation to figure out how it works.



In [13]:

    
# Number of negative labels to sample
n_sampled = 100
with train_graph.as_default():
    softmax_w = tf.Variable(tf.truncated_normal((n_vocab, n_embedding), stddev=0.1))
    softmax_b = tf.Variable(tf.zeros(n_vocab))
    
    # Calculate the loss using negative sampling
    loss = tf.nn.sampled_softmax_loss(softmax_w, softmax_b, 
                                      labels, embed,
                                      n_sampled, n_vocab)
    
    cost = tf.reduce_mean(loss)
    optimizer = tf.train.AdamOptimizer().minimize(cost)

Validation

This code is from Thushan Ganegedara's implementation. Here we're going to choose a few common words and few uncommon words. Then, we'll print out the closest words to them. It's a nice way to check that our embedding table is grouping together words with similar semantic meanings.



In [14]:

    
with train_graph.as_default():
    ## From Thushan Ganegedara's implementation
    valid_size = 16 # Random set of words to evaluate similarity on.
    valid_window = 100
    # pick 8 samples from (0,100) and (1000,1100) each ranges. lower id implies more frequent 
    valid_examples = np.array(random.sample(range(valid_window), valid_size//2))
    valid_examples = np.append(valid_examples, 
                               random.sample(range(1000,1000+valid_window), valid_size//2))

    valid_dataset = tf.constant(valid_examples, dtype=tf.int32)
    
    # We use the cosine distance:
    norm = tf.sqrt(tf.reduce_sum(tf.square(embedding), 1, keep_dims=True))
    normalized_embedding = embedding / norm
    valid_embedding = tf.nn.embedding_lookup(normalized_embedding, valid_dataset)
    similarity = tf.matmul(valid_embedding, tf.transpose(normalized_embedding))



In [15]:

    
# If the checkpoints directory doesn't exist:
!mkdir checkpoints

Training

Below is the code to train the network. Every 100 batches it reports the training loss. Every 1000 batches, it'll print out the validation words.



In [16]:

    
epochs = 10
batch_size = 1000
window_size = 10

with train_graph.as_default():
    saver = tf.train.Saver()

with tf.Session(graph=train_graph) as sess:
    iteration = 1
    loss = 0
    sess.run(tf.global_variables_initializer())

    for e in range(1, epochs+1):
        batches = get_batches(train_words, batch_size, window_size)
        start = time.time()
        for x, y in batches:
            
            feed = {inputs: x,
                    labels: np.array(y)[:, None]}
            train_loss, _ = sess.run([cost, optimizer], feed_dict=feed)
            
            loss += train_loss
            
            if iteration % 100 == 0: 
                end = time.time()
                print("Epoch {}/{}".format(e, epochs),
                      "Iteration: {}".format(iteration),
                      "Avg. Training loss: {:.4f}".format(loss/100),
                      "{:.4f} sec/batch".format((end-start)/100))
                loss = 0
                start = time.time()
            
            if iteration % 1000 == 0:
                ## From Thushan Ganegedara's implementation
                # note that this is expensive (~20% slowdown if computed every 500 steps)
                sim = similarity.eval()
                for i in range(valid_size):
                    valid_word = int_to_vocab[valid_examples[i]]
                    top_k = 8 # number of nearest neighbors
                    nearest = (-sim[i, :]).argsort()[1:top_k+1]
                    log = 'Nearest to %s:' % valid_word
                    for k in range(top_k):
                        close_word = int_to_vocab[nearest[k]]
                        log = '%s %s,' % (log, close_word)
                    print(log)
            
            iteration += 1
    save_path = saver.save(sess, "checkpoints/text8.ckpt")
    embed_mat = sess.run(normalized_embedding)









    



Epoch 1/10 Iteration: 100 Avg. Training loss: 5.6297 0.4538 sec/batch
Epoch 1/10 Iteration: 200 Avg. Training loss: 5.6205 0.4370 sec/batch
Epoch 1/10 Iteration: 300 Avg. Training loss: 5.5030 0.4630 sec/batch
Epoch 1/10 Iteration: 400 Avg. Training loss: 5.6015 0.4343 sec/batch
Epoch 1/10 Iteration: 500 Avg. Training loss: 5.5062 0.4305 sec/batch
Epoch 1/10 Iteration: 600 Avg. Training loss: 5.5432 0.4315 sec/batch
Epoch 1/10 Iteration: 700 Avg. Training loss: 5.5746 0.4284 sec/batch
Epoch 1/10 Iteration: 800 Avg. Training loss: 5.5290 0.4319 sec/batch
Epoch 1/10 Iteration: 900 Avg. Training loss: 5.4559 0.4308 sec/batch
Epoch 1/10 Iteration: 1000 Avg. Training loss: 5.3904 0.4435 sec/batch
Nearest to all: prozac, persuade, caribou, instructors, shams, mandarin, universitat, samadhi,
Nearest to six: limburg, class, cimon, couscous, sing, hitters, nikolai, throop,
Nearest to their: bounty, handsel, drake, lethal, corsair, muhi, fleshed, perfumes,
Nearest to also: creed, pharisee, transports, akkad, weddings, ring, rescheduling, steinitz,
Nearest to into: shared, panavia, nitz, repairs, hewn, applicability, faxing, pd,
Nearest to some: pty, joys, meroitic, ananias, sip, enemies, kitten, combine,
Nearest to called: emporis, gta, arts, hubs, mercenary, interdependence, poet, antidote,
Nearest to in: nanobots, sirhan, xerox, originate, supplied, gonorrhea, vanguard, almost,
Nearest to consists: fascination, censorial, ligase, he, scrolling, siculus, rising, parliaments,
Nearest to numerous: haggard, coelacanth, sexuality, injects, calculi, disheveled, bun, reinforcing,
Nearest to know: breslau, markets, ionization, tlatelolco, amine, surya, demeaning, bandung,
Nearest to versions: audit, rajendra, pushing, franconian, tattooed, obscene, jellies, unintentional,
Nearest to shows: hapgood, gingerich, encompasses, rigid, integrative, macrinus, prs, montague,
Nearest to orthodox: regress, transept, stack, counsels, indulging, hydrochloric, thema, informant,
Nearest to scale: casing, forges, definable, simulators, antiquities, azar, officers, auroral,
Nearest to joseph: bangor, psalter, subversive, octavian, moyen, armorial, suriname, diversions,
Epoch 1/10 Iteration: 1100 Avg. Training loss: 5.4799 0.4523 sec/batch
Epoch 1/10 Iteration: 1200 Avg. Training loss: 5.3893 0.4416 sec/batch
Epoch 1/10 Iteration: 1300 Avg. Training loss: 5.3453 0.4547 sec/batch
Epoch 1/10 Iteration: 1400 Avg. Training loss: 5.2340 0.4417 sec/batch
Epoch 1/10 Iteration: 1500 Avg. Training loss: 5.2030 0.4534 sec/batch
Epoch 1/10 Iteration: 1600 Avg. Training loss: 5.1452 0.4920 sec/batch
Epoch 1/10 Iteration: 1700 Avg. Training loss: 5.1058 0.4597 sec/batch
Epoch 1/10 Iteration: 1800 Avg. Training loss: 5.0645 0.4226 sec/batch
Epoch 1/10 Iteration: 1900 Avg. Training loss: 4.9773 0.4286 sec/batch
Epoch 1/10 Iteration: 2000 Avg. Training loss: 4.9889 0.4311 sec/batch
Nearest to all: mandarin, prozac, persuade, hague, immediately, sentenced, carney, shek,
Nearest to six: limburg, class, couscous, cimon, sing, other, explosion, august,
Nearest to their: either, bounty, drake, longitude, concept, lethal, handsel, management,
Nearest to also: result, creed, are, ring, standing, idea, some, frequently,
Nearest to into: shared, repairs, accordingly, nitz, applicability, panavia, bore, pd,
Nearest to some: etc, enemies, pty, meroitic, intersection, victories, also, combine,
Nearest to called: arts, gta, enacted, empires, mercenary, nevertheless, abandoned, exclusive,
Nearest to in: originate, xerox, nanobots, supplied, almost, sirhan, structural, d,
Nearest to consists: he, backward, fascination, fathers, rising, scrolling, censorial, ligase,
Nearest to numerous: statutes, haggard, strand, defense, sports, injects, reinforcing, tips,
Nearest to know: breslau, markets, bounded, developments, perform, ionization, amine, tlatelolco,
Nearest to versions: pushing, expect, man, franconian, disturbed, audit, burden, paley,
Nearest to shows: hapgood, encompasses, rigid, diomedes, bitter, integrative, bermudez, bracks,
Nearest to orthodox: transept, regress, stack, counsels, informant, seasonal, bones, decision,
Nearest to scale: casing, definable, officers, forges, conform, antiquities, simulators, armies,
Nearest to joseph: bangor, psalter, subversive, lucrative, texcoco, suriname, psychology, moyen,
Epoch 1/10 Iteration: 2100 Avg. Training loss: 4.9356 0.4438 sec/batch
Epoch 1/10 Iteration: 2200 Avg. Training loss: 4.9138 0.4479 sec/batch
Epoch 1/10 Iteration: 2300 Avg. Training loss: 4.8847 0.4177 sec/batch
Epoch 1/10 Iteration: 2400 Avg. Training loss: 4.8636 0.4164 sec/batch
Epoch 1/10 Iteration: 2500 Avg. Training loss: 4.8308 0.4117 sec/batch
Epoch 1/10 Iteration: 2600 Avg. Training loss: 4.8251 0.4236 sec/batch
Epoch 1/10 Iteration: 2700 Avg. Training loss: 4.7906 0.4130 sec/batch
Epoch 1/10 Iteration: 2800 Avg. Training loss: 4.7937 0.4120 sec/batch
Epoch 1/10 Iteration: 2900 Avg. Training loss: 4.7696 0.4164 sec/batch
Epoch 1/10 Iteration: 3000 Avg. Training loss: 4.7959 0.4158 sec/batch
Nearest to all: prozac, sentenced, persuade, mandarin, hague, hewlett, roc, instructors,
Nearest to six: limburg, class, sing, couscous, coronation, cimon, explosion, max,
Nearest to their: either, lethal, longitude, bounty, privacy, drake, handsel, modernist,
Nearest to also: creed, transports, are, weddings, vii, excluding, ring, ministries,
Nearest to into: shared, accordingly, bore, repairs, nitz, applicability, panavia, pd,
Nearest to some: pty, enemies, etc, combine, joys, ip, meroitic, chanting,
Nearest to called: arts, gta, hubs, mercenary, calendar, enacted, abandoned, poet,
Nearest to in: xerox, originate, nanobots, supplied, sirhan, vanguard, almost, structural,
Nearest to consists: backward, censorial, fascination, fathers, he, rising, siculus, ligase,
Nearest to numerous: haggard, statutes, sexuality, flick, reinforcing, tips, defense, ballistics,
Nearest to know: markets, breslau, ionization, developments, bounded, cpi, tlatelolco, blockbuster,
Nearest to versions: audit, pushing, disturbed, expect, burden, franconian, man, paley,
Nearest to shows: hapgood, encompasses, rigid, diomedes, bitter, bermudez, bracks, loyalists,
Nearest to orthodox: transept, stack, regress, codepoint, counsels, departmental, informant, decision,
Nearest to scale: casing, forges, simulators, definable, conform, armies, antiquities, officers,
Nearest to joseph: bangor, psalter, subversive, lucrative, suriname, deities, moyen, stunt,
Epoch 1/10 Iteration: 3100 Avg. Training loss: 4.7672 0.4221 sec/batch
Epoch 1/10 Iteration: 3200 Avg. Training loss: 4.7704 0.4138 sec/batch
Epoch 1/10 Iteration: 3300 Avg. Training loss: 4.7158 0.4164 sec/batch
Epoch 1/10 Iteration: 3400 Avg. Training loss: 4.6965 0.4142 sec/batch
Epoch 1/10 Iteration: 3500 Avg. Training loss: 4.7360 0.4185 sec/batch
Epoch 1/10 Iteration: 3600 Avg. Training loss: 4.6803 0.4154 sec/batch
Epoch 1/10 Iteration: 3700 Avg. Training loss: 4.7054 0.4068 sec/batch
Epoch 1/10 Iteration: 3800 Avg. Training loss: 4.7252 0.4090 sec/batch
Epoch 1/10 Iteration: 3900 Avg. Training loss: 4.6706 0.4142 sec/batch
Epoch 1/10 Iteration: 4000 Avg. Training loss: 4.6483 0.4123 sec/batch
Nearest to all: prozac, caribou, shams, canceled, instructors, sentenced, persuade, hewlett,
Nearest to six: limburg, class, coronation, max, august, cimon, five, hyde,
Nearest to their: bounty, lethal, handsel, either, longitude, privacy, muhi, maghrib,
Nearest to also: transports, weddings, encompasses, creed, akkad, ring, excluding, pharisee,
Nearest to into: shared, nitz, repairs, bore, faxing, accordingly, pd, hewn,
Nearest to some: pty, ip, chanting, joys, mansfield, meroitic, etc, enemies,
Nearest to called: emporis, interdependence, gta, arts, hubs, antidote, mercenary, interface,
Nearest to in: xerox, originate, vanguard, opold, sirhan, ranks, luminaries, participating,
Nearest to consists: backward, censorial, inconsistently, fascination, dravidians, fathers, stat, behavior,
Nearest to numerous: haggard, bun, reinforcing, flick, calculi, injects, sexuality, ballistics,
Nearest to know: breslau, markets, ionization, cpi, amine, bounded, tlatelolco, word,
Nearest to versions: disturbed, audit, rajendra, pushing, man, expect, franconian, forgiveness,
Nearest to shows: hapgood, encompasses, rigid, gingerich, integrative, bermudez, diomedes, loyalists,
Nearest to orthodox: transept, codepoint, regress, stack, indulging, departmental, yucatan, playlists,
Nearest to scale: casing, simulators, wrongdoing, forges, definable, rawalpindi, azar, mainframes,
Nearest to joseph: bangor, psalter, subversive, sabbath, moyen, assisi, lucrative, deities,
Epoch 1/10 Iteration: 4100 Avg. Training loss: 4.6752 0.4169 sec/batch
Epoch 1/10 Iteration: 4200 Avg. Training loss: 4.6880 0.4133 sec/batch
Epoch 1/10 Iteration: 4300 Avg. Training loss: 4.6477 0.4145 sec/batch
Epoch 1/10 Iteration: 4400 Avg. Training loss: 4.6091 0.4142 sec/batch
Epoch 1/10 Iteration: 4500 Avg. Training loss: 4.6274 0.4129 sec/batch
Epoch 1/10 Iteration: 4600 Avg. Training loss: 4.6583 0.4120 sec/batch
Epoch 2/10 Iteration: 4700 Avg. Training loss: 4.6017 0.3104 sec/batch
Epoch 2/10 Iteration: 4800 Avg. Training loss: 4.5593 0.4142 sec/batch
Epoch 2/10 Iteration: 4900 Avg. Training loss: 4.5252 0.4119 sec/batch
Epoch 2/10 Iteration: 5000 Avg. Training loss: 4.5311 0.4374 sec/batch
Nearest to all: prozac, caribou, shams, instructors, canceled, mandarin, sentenced, roc,
Nearest to six: limburg, five, statesman, max, hockey, coronation, cimon, hyde,
Nearest to their: lethal, handsel, longitude, either, bounty, maghrib, privacy, muhi,
Nearest to also: transports, encompasses, weddings, are, frequently, heritability, akkad, ministries,
Nearest to into: shared, bore, repairs, faxing, hewn, nitz, codebase, applicability,
Nearest to some: pty, etc, ip, meroitic, joys, chanting, misled, audible,
Nearest to called: interdependence, emporis, gta, interface, arts, presocratic, antidote, hubs,
Nearest to in: xerox, originate, opold, nanobots, participating, sirhan, luminaries, ranks,
Nearest to consists: censorial, backward, inconsistently, dravidians, stat, behavior, siculus, expended,
Nearest to numerous: calculi, haggard, reinforcing, flick, defense, injects, bun, arrangements,
Nearest to know: breslau, markets, ionization, tlatelolco, word, demeaning, cpi, grandees,
Nearest to versions: audit, disturbed, rajendra, expect, forgiveness, franconian, bus, paley,
Nearest to shows: hapgood, encompasses, rigid, gingerich, integrative, diomedes, aqsa, bermudez,
Nearest to orthodox: transept, codepoint, regress, yucatan, indulging, quarterbacks, extremity, mcpherson,
Nearest to scale: casing, simulators, forges, azar, rawalpindi, constans, wrongdoing, yourselves,
Nearest to joseph: bangor, psalter, sabbath, assisi, moyen, subversive, naturalist, stunt,
Epoch 2/10 Iteration: 5100 Avg. Training loss: 4.5101 0.4772 sec/batch
Epoch 2/10 Iteration: 5200 Avg. Training loss: 4.4926 0.5361 sec/batch
Epoch 2/10 Iteration: 5300 Avg. Training loss: 4.4697 0.5471 sec/batch
Epoch 2/10 Iteration: 5400 Avg. Training loss: 4.5162 0.5454 sec/batch
Epoch 2/10 Iteration: 5500 Avg. Training loss: 4.4861 0.5427 sec/batch
Epoch 2/10 Iteration: 5600 Avg. Training loss: 4.4768 0.5444 sec/batch
Epoch 2/10 Iteration: 5700 Avg. Training loss: 4.4339 0.5455 sec/batch
Epoch 2/10 Iteration: 5800 Avg. Training loss: 4.3864 0.5464 sec/batch
Epoch 2/10 Iteration: 5900 Avg. Training loss: 4.4259 0.5489 sec/batch
Epoch 2/10 Iteration: 6000 Avg. Training loss: 4.4437 0.5439 sec/batch
Nearest to all: prozac, shams, caribou, instructors, unfavorable, roc, mandarin, canceled,
Nearest to six: limburg, max, five, hockey, cimon, class, dod, milestones,
Nearest to their: either, lethal, handsel, longitude, bounty, maghrib, privacy, muhi,
Nearest to also: encompasses, transports, are, weddings, frequently, heritability, arrhythmias, excluding,
Nearest to into: shared, bore, faxing, hewn, accordingly, repairs, ore, codebase,
Nearest to some: pty, etc, ip, meroitic, chanting, joys, have, sip,
Nearest to called: emporis, interdependence, gta, interface, calvinistic, hubs, valued, steed,
Nearest to in: xerox, originate, sirhan, opold, participating, restated, nanobots, luminaries,
Nearest to consists: censorial, backward, dravidians, inconsistently, behavior, stat, rising, fathers,
Nearest to numerous: haggard, reinforcing, calculi, defense, flick, arrangements, bun, coelacanth,
Nearest to know: breslau, tlatelolco, ionization, word, markets, make, demeaning, cpi,
Nearest to versions: audit, disturbed, rajendra, bus, expect, forgiveness, hammurabi, hoe,
Nearest to shows: hapgood, encompasses, rigid, integrative, gingerich, diomedes, aqsa, bermudez,
Nearest to orthodox: transept, apostolic, codepoint, regress, extremity, sigismund, indulging, byzantium,
Nearest to scale: casing, forges, simulators, velocity, rawalpindi, azar, constans, cooling,
Nearest to joseph: bangor, assisi, psalter, sabbath, moyen, octavian, subversive, naturalist,
Epoch 2/10 Iteration: 6100 Avg. Training loss: 4.4230 0.5496 sec/batch
Epoch 2/10 Iteration: 6200 Avg. Training loss: 4.4345 0.4834 sec/batch
Epoch 2/10 Iteration: 6300 Avg. Training loss: 4.4480 0.4152 sec/batch
Epoch 2/10 Iteration: 6400 Avg. Training loss: 4.3603 0.4153 sec/batch
Epoch 2/10 Iteration: 6500 Avg. Training loss: 4.4398 0.4131 sec/batch
Epoch 2/10 Iteration: 6600 Avg. Training loss: 4.4669 0.4141 sec/batch
Epoch 2/10 Iteration: 6700 Avg. Training loss: 4.3897 0.4139 sec/batch
Epoch 2/10 Iteration: 6800 Avg. Training loss: 4.4010 0.4133 sec/batch
Epoch 2/10 Iteration: 6900 Avg. Training loss: 4.4147 0.4125 sec/batch
Epoch 2/10 Iteration: 7000 Avg. Training loss: 4.3883 0.4127 sec/batch
Nearest to all: prozac, shams, caribou, canceled, unfavorable, voluminous, instructors, roc,
Nearest to six: limburg, five, max, cimon, hockey, capita, median, three,
Nearest to their: either, lethal, privacy, handsel, epistemological, longitude, bounty, predominate,
Nearest to also: are, frequently, encompasses, transports, weddings, some, heritability, deranged,
Nearest to into: shared, faxing, repairs, accordingly, hewn, codebase, bore, ore,
Nearest to some: etc, pty, ip, meroitic, chanting, audible, joys, have,
Nearest to called: emporis, interdependence, gta, interface, known, hubs, hangul, valued,
Nearest to in: tok, opold, originate, participating, sirhan, xerox, luminaries, nanobots,
Nearest to consists: censorial, backward, dravidians, inconsistently, stat, civilize, expended, turkestan,
Nearest to numerous: haggard, flick, arrangements, coelacanth, reinforcing, calculi, bun, injects,
Nearest to know: breslau, word, ionization, tlatelolco, make, demeaning, grandees, cpi,
Nearest to versions: disturbed, audit, expect, paley, digital, bus, leprosy, rajendra,
Nearest to shows: hapgood, encompasses, rigid, aqsa, integrative, gingerich, instantaneously, diomedes,
Nearest to orthodox: apostolic, transept, eucharist, extremity, codepoint, catholic, sigismund, regress,
Nearest to scale: casing, velocity, rawalpindi, forges, azar, constans, yourselves, cooling,
Nearest to joseph: bangor, moyen, psalter, sabbath, assisi, subversive, naturalist, winnings,
Epoch 2/10 Iteration: 7100 Avg. Training loss: 4.3541 0.4192 sec/batch
Epoch 2/10 Iteration: 7200 Avg. Training loss: 4.3907 0.4140 sec/batch
Epoch 2/10 Iteration: 7300 Avg. Training loss: 4.3418 0.4181 sec/batch
Epoch 2/10 Iteration: 7400 Avg. Training loss: 4.3448 0.4210 sec/batch
Epoch 2/10 Iteration: 7500 Avg. Training loss: 4.3926 0.4304 sec/batch
Epoch 2/10 Iteration: 7600 Avg. Training loss: 4.3533 0.4334 sec/batch
Epoch 2/10 Iteration: 7700 Avg. Training loss: 4.3953 0.4497 sec/batch
Epoch 2/10 Iteration: 7800 Avg. Training loss: 4.3904 0.4474 sec/batch
Epoch 2/10 Iteration: 7900 Avg. Training loss: 4.3407 0.4482 sec/batch
Epoch 2/10 Iteration: 8000 Avg. Training loss: 4.3367 0.4426 sec/batch
Nearest to all: prozac, shams, canceled, roc, caribou, instructors, sentenced, unfavorable,
Nearest to six: limburg, five, capita, cimon, median, max, three, dod,
Nearest to their: either, lethal, privacy, handsel, predominate, epistemological, however, bounty,
Nearest to also: are, encompasses, weddings, frequently, transports, some, select, deranged,
Nearest to into: accordingly, hewn, shared, repairs, faxing, bore, codebase, refilled,
Nearest to some: pty, etc, ip, sip, chanting, mansfield, meroitic, strictest,
Nearest to called: emporis, interdependence, gta, hangul, known, ker, existance, calvinistic,
Nearest to in: opold, originate, border, participating, sirhan, nanobots, chrominance, tok,
Nearest to consists: censorial, dravidians, grt, turkestan, rising, inconsistently, expended, backward,
Nearest to numerous: haggard, coelacanth, flick, arrangements, injects, pathetic, bun, alewife,
Nearest to know: breslau, word, tlatelolco, make, ionization, grandees, demeaning, we,
Nearest to versions: disturbed, bus, paley, audit, digital, dvb, expect, leprosy,
Nearest to shows: hapgood, rigid, encompasses, gingerich, instantaneously, integrative, bermudez, protagonists,
Nearest to orthodox: apostolic, eucharist, catholic, transept, byzantium, christians, codepoint, catholics,
Nearest to scale: casing, forges, azar, rawalpindi, velocity, cooling, declination, constans,
Nearest to joseph: bangor, psalter, moyen, assisi, subversive, sabbath, naturalist, winnings,
Epoch 2/10 Iteration: 8100 Avg. Training loss: 4.3417 0.4507 sec/batch
Epoch 2/10 Iteration: 8200 Avg. Training loss: 4.2642 0.4457 sec/batch
Epoch 2/10 Iteration: 8300 Avg. Training loss: 4.3774 0.4406 sec/batch
Epoch 2/10 Iteration: 8400 Avg. Training loss: 4.4014 0.4310 sec/batch
Epoch 2/10 Iteration: 8500 Avg. Training loss: 4.3636 0.4424 sec/batch
Epoch 2/10 Iteration: 8600 Avg. Training loss: 4.2943 0.4894 sec/batch
Epoch 2/10 Iteration: 8700 Avg. Training loss: 4.2794 0.5119 sec/batch
Epoch 2/10 Iteration: 8800 Avg. Training loss: 4.3286 0.4552 sec/batch
Epoch 2/10 Iteration: 8900 Avg. Training loss: 4.2498 0.4150 sec/batch
Epoch 2/10 Iteration: 9000 Avg. Training loss: 4.3003 0.4156 sec/batch
Nearest to all: prozac, sentenced, canceled, shams, instructors, caribou, roc, summation,
Nearest to six: five, limburg, capita, seven, cimon, three, nine, median,
Nearest to their: either, lethal, handsel, privacy, maghrib, bounty, predominate, compounded,
Nearest to also: are, encompasses, weddings, transports, some, cousteau, arrhythmias, frequently,
Nearest to into: accordingly, repairs, batavia, airstream, hewn, bore, faxing, codebase,
Nearest to some: etc, pty, ip, joys, chanting, sip, have, strictest,
Nearest to called: emporis, interdependence, known, gta, hangul, oshawa, ker, cloisters,
Nearest to in: nanobots, opold, participating, originate, chrominance, tok, sirhan, ranks,
Nearest to consists: censorial, grt, dravidians, expended, shops, hake, stat, manna,
Nearest to numerous: haggard, coelacanth, flick, pathetic, bun, arrangements, injects, fantastical,
Nearest to know: breslau, word, tlatelolco, make, grandees, demeaning, we, fal,
Nearest to versions: disturbed, leprosy, paley, hoe, luria, dvb, audit, bus,
Nearest to shows: hapgood, gingerich, rigid, instantaneously, integrative, bermudez, armalite, protagonists,
Nearest to orthodox: apostolic, transept, byzantium, eucharist, codepoint, lutheran, anointed, regress,
Nearest to scale: casing, forges, rawalpindi, azar, constans, cooling, mainframes, velocity,
Nearest to joseph: bangor, psalter, moyen, naturalist, assisi, subversive, winnings, sabbath,
Epoch 2/10 Iteration: 9100 Avg. Training loss: 4.2975 0.4214 sec/batch
Epoch 2/10 Iteration: 9200 Avg. Training loss: 4.2964 0.4123 sec/batch
Epoch 3/10 Iteration: 9300 Avg. Training loss: 4.3255 0.2061 sec/batch
Epoch 3/10 Iteration: 9400 Avg. Training loss: 4.2500 0.4166 sec/batch
Epoch 3/10 Iteration: 9500 Avg. Training loss: 4.2167 0.4111 sec/batch
Epoch 3/10 Iteration: 9600 Avg. Training loss: 4.2273 0.4117 sec/batch
Epoch 3/10 Iteration: 9700 Avg. Training loss: 4.2243 0.4128 sec/batch
Epoch 3/10 Iteration: 9800 Avg. Training loss: 4.2293 0.4131 sec/batch
Epoch 3/10 Iteration: 9900 Avg. Training loss: 4.2097 0.4124 sec/batch
Epoch 3/10 Iteration: 10000 Avg. Training loss: 4.1819 0.4123 sec/batch
Nearest to all: prozac, canceled, summation, shams, instructors, samadhi, coexisted, sentenced,
Nearest to six: five, limburg, capita, three, median, seven, cimon, dod,
Nearest to their: either, lethal, maghrib, handsel, predominate, bounty, compounded, privacy,
Nearest to also: encompasses, are, weddings, some, transports, arrhythmias, format, cousteau,
Nearest to into: accordingly, repairs, batavia, codebase, hewn, airstream, refilled, faxing,
Nearest to some: etc, pty, joys, sip, ip, chanting, also, audible,
Nearest to called: emporis, known, interdependence, hubs, cloisters, hangul, oshawa, gta,
Nearest to in: nanobots, chrominance, originate, the, participating, tok, border, opold,
Nearest to consists: censorial, grt, dravidians, hake, expended, stat, rho, shops,
Nearest to numerous: coelacanth, haggard, flick, pathetic, arrangements, bun, injects, fantastical,
Nearest to know: breslau, we, word, demeaning, grandees, tlatelolco, make, manner,
Nearest to versions: architectures, disturbed, operating, dvb, bus, mapped, quadra, hoe,
Nearest to shows: hapgood, armalite, integrative, instantaneously, gingerich, rigid, encompasses, bermudez,
Nearest to orthodox: apostolic, byzantium, transept, eucharist, lutheran, codepoint, copleston, christians,
Nearest to scale: casing, forges, azar, rawalpindi, velocity, cooling, declination, constans,
Nearest to joseph: bangor, psalter, moyen, octavian, assisi, naturalist, diversions, affirmed,
Epoch 3/10 Iteration: 10100 Avg. Training loss: 4.2484 0.4152 sec/batch
Epoch 3/10 Iteration: 10200 Avg. Training loss: 4.2153 0.4095 sec/batch
Epoch 3/10 Iteration: 10300 Avg. Training loss: 4.2103 0.4103 sec/batch
Epoch 3/10 Iteration: 10400 Avg. Training loss: 4.1590 0.4115 sec/batch
Epoch 3/10 Iteration: 10500 Avg. Training loss: 4.1929 0.4121 sec/batch
Epoch 3/10 Iteration: 10600 Avg. Training loss: 4.1532 0.4122 sec/batch
Epoch 3/10 Iteration: 10700 Avg. Training loss: 4.1524 0.4091 sec/batch
Epoch 3/10 Iteration: 10800 Avg. Training loss: 4.1819 0.4095 sec/batch
Epoch 3/10 Iteration: 10900 Avg. Training loss: 4.1877 0.4119 sec/batch
Epoch 3/10 Iteration: 11000 Avg. Training loss: 4.1757 0.4117 sec/batch
Nearest to all: prozac, canceled, summation, shams, eitc, coexisted, sentenced, roc,
Nearest to six: five, capita, three, limburg, median, seven, cimon, nine,
Nearest to their: either, however, lethal, predominate, maghrib, privacy, feign, compounded,
Nearest to also: encompasses, are, transports, some, forwarding, abh, arrhythmias, frequently,
Nearest to into: accordingly, codebase, hewn, repairs, batavia, faxing, refilled, airstream,
Nearest to some: etc, joys, pty, sip, chanting, also, have, ip,
Nearest to called: emporis, known, hangul, interdependence, cloisters, gta, or, oshawa,
Nearest to in: nanobots, originate, chrominance, participating, opold, sirhan, the, discussions,
Nearest to consists: censorial, rho, grt, dravidians, expended, shops, guildhall, stat,
Nearest to numerous: coelacanth, arrangements, haggard, flick, pathetic, bun, branding, mcdermott,
Nearest to know: breslau, we, make, word, demeaning, tlatelolco, fal, ionization,
Nearest to versions: disturbed, mapped, pc, architectures, dvb, multiverses, roms, hoe,
Nearest to shows: hapgood, encompasses, integrative, instantaneously, armalite, diomedes, rigid, bermudez,
Nearest to orthodox: apostolic, transept, lutheran, eucharist, codepoint, catholic, byzantium, christians,
Nearest to scale: casing, forges, azar, rawalpindi, declination, constans, velocity, cooling,
Nearest to joseph: bangor, psalter, moyen, octavian, assisi, naturalist, affirmed, diversions,
Epoch 3/10 Iteration: 11100 Avg. Training loss: 4.1775 0.4145 sec/batch
Epoch 3/10 Iteration: 11200 Avg. Training loss: 4.2459 0.4115 sec/batch
Epoch 3/10 Iteration: 11300 Avg. Training loss: 4.1849 0.4129 sec/batch
Epoch 3/10 Iteration: 11400 Avg. Training loss: 4.1217 0.4096 sec/batch
Epoch 3/10 Iteration: 11500 Avg. Training loss: 4.1682 0.4108 sec/batch
Epoch 3/10 Iteration: 11600 Avg. Training loss: 4.1743 0.4121 sec/batch
Epoch 3/10 Iteration: 11700 Avg. Training loss: 4.2096 0.4118 sec/batch
Epoch 3/10 Iteration: 11800 Avg. Training loss: 4.1838 0.4107 sec/batch
Epoch 3/10 Iteration: 11900 Avg. Training loss: 4.1393 0.4166 sec/batch
Epoch 3/10 Iteration: 12000 Avg. Training loss: 4.1925 0.4175 sec/batch
Nearest to all: prozac, summation, canceled, roc, shams, sentenced, eitc, coexisted,
Nearest to six: five, capita, limburg, three, median, nine, seven, cimon,
Nearest to their: either, lethal, maghrib, however, tasman, predominate, compounded, bounty,
Nearest to also: are, encompasses, some, transports, commonly, abh, arrhythmias, frequently,
Nearest to into: accordingly, codebase, hewn, repairs, batavia, airstream, faxing, refilled,
Nearest to some: pty, also, etc, sip, joys, ip, chanting, have,
Nearest to called: emporis, known, hangul, or, gta, cloisters, interdependence, as,
Nearest to in: the, nanobots, originate, tok, sirhan, opold, participating, discussions,
Nearest to consists: censorial, rho, grt, guildhall, schists, stat, hake, unleavened,
Nearest to numerous: coelacanth, arrangements, haggard, flick, bun, branding, mcdermott, slacks,
Nearest to know: breslau, we, make, ionization, word, fal, demeaning, tlatelolco,
Nearest to versions: disturbed, camcorder, multiverses, hoe, dvb, quadra, mapped, paley,
Nearest to shows: hapgood, encompasses, integrative, instantaneously, armalite, diomedes, rigid, lodging,
Nearest to orthodox: apostolic, lutheran, byzantium, transept, eucharist, codepoint, extremity, anointed,
Nearest to scale: casing, forges, azar, declination, rawalpindi, cooling, nb, velocity,
Nearest to joseph: bangor, psalter, moyen, octavian, naturalist, assisi, rilke, schell,
Epoch 3/10 Iteration: 12100 Avg. Training loss: 4.2047 0.4167 sec/batch
Epoch 3/10 Iteration: 12200 Avg. Training loss: 4.1683 0.4123 sec/batch
Epoch 3/10 Iteration: 12300 Avg. Training loss: 4.1801 0.4090 sec/batch
Epoch 3/10 Iteration: 12400 Avg. Training loss: 4.1671 0.4109 sec/batch
Epoch 3/10 Iteration: 12500 Avg. Training loss: 4.1492 0.4146 sec/batch
Epoch 3/10 Iteration: 12600 Avg. Training loss: 4.1256 0.4148 sec/batch
Epoch 3/10 Iteration: 12700 Avg. Training loss: 4.1855 0.4127 sec/batch
Epoch 3/10 Iteration: 12800 Avg. Training loss: 4.1024 0.4166 sec/batch
Epoch 3/10 Iteration: 12900 Avg. Training loss: 4.2091 0.4129 sec/batch
Epoch 3/10 Iteration: 13000 Avg. Training loss: 4.2128 0.4128 sec/batch
Nearest to all: prozac, canceled, sentenced, shams, summation, coexisted, eitc, roc,
Nearest to six: five, three, seven, nine, limburg, one, capita, two,
Nearest to their: either, maghrib, lethal, bounty, reassert, tasman, decided, predominate,
Nearest to also: some, encompasses, are, restructure, cousteau, the, abh, weddings,
Nearest to into: accordingly, batavia, hewn, codebase, repairs, refilled, nitz, airstream,
Nearest to some: also, sip, pty, etc, ip, joys, have, chanting,
Nearest to called: emporis, known, hangul, cloisters, gta, poet, mercenary, calvinistic,
Nearest to in: opold, the, participating, petrograd, newfound, nanobots, tok, avignon,
Nearest to consists: censorial, rho, schists, grt, guildhall, unleavened, hake, shops,
Nearest to numerous: coelacanth, arrangements, haggard, pathetic, mcdermott, bun, prairie, comparative,
Nearest to know: breslau, we, word, make, fal, demeaning, tlatelolco, markets,
Nearest to versions: disturbed, camcorder, hoe, multiverses, obscene, mapped, dvb, quadra,
Nearest to shows: hapgood, encompasses, lodging, integrative, armalite, vai, instantaneously, bermudez,
Nearest to orthodox: apostolic, transept, byzantium, thema, lutheran, codepoint, extremity, eucharist,
Nearest to scale: casing, forges, azar, rawalpindi, nb, declination, guth, mainframes,
Nearest to joseph: bangor, psalter, moyen, naturalist, leases, octavian, diversions, affirmed,
Epoch 3/10 Iteration: 13100 Avg. Training loss: 4.2240 0.4155 sec/batch
Epoch 3/10 Iteration: 13200 Avg. Training loss: 4.1286 0.4123 sec/batch
Epoch 3/10 Iteration: 13300 Avg. Training loss: 4.1575 0.4089 sec/batch
Epoch 3/10 Iteration: 13400 Avg. Training loss: 4.1509 0.4088 sec/batch
Epoch 3/10 Iteration: 13500 Avg. Training loss: 4.0205 0.4269 sec/batch
Epoch 3/10 Iteration: 13600 Avg. Training loss: 4.1452 0.4650 sec/batch
Epoch 3/10 Iteration: 13700 Avg. Training loss: 4.1751 0.4555 sec/batch
Epoch 3/10 Iteration: 13800 Avg. Training loss: 4.1540 0.4573 sec/batch
Epoch 4/10 Iteration: 13900 Avg. Training loss: 4.1662 0.1158 sec/batch
Epoch 4/10 Iteration: 14000 Avg. Training loss: 4.1232 0.4567 sec/batch
Nearest to all: prozac, canceled, summation, fixative, requirement, eitc, instructors, sentenced,
Nearest to six: five, three, seven, capita, nine, median, two, zero,
Nearest to their: either, maghrib, lethal, bounty, tasman, however, compounded, handsel,
Nearest to also: are, encompasses, some, the, cousteau, transports, faceted, restructure,
Nearest to into: batavia, accordingly, codebase, hewn, upward, repairs, airstream, refilled,
Nearest to some: also, sip, pty, etc, ip, joys, have, audible,
Nearest to called: emporis, known, or, hangul, as, cloisters, gta, strictly,
Nearest to in: the, nanobots, participating, originate, petrograd, opold, tok, principally,
Nearest to consists: censorial, rho, grt, parliaments, shops, schists, unleavened, guildhall,
Nearest to numerous: coelacanth, arrangements, dwellers, slacks, bun, haggard, mcdermott, comparative,
Nearest to know: we, breslau, make, markets, demeaning, fal, diagnoses, ask,
Nearest to versions: camcorder, mapped, disturbed, hoe, dvb, operating, multiverses, unlocks,
Nearest to shows: hapgood, armalite, lodging, encompasses, integrative, instantaneously, diomedes, vai,
Nearest to orthodox: apostolic, transept, thema, lutheran, codepoint, eucharist, extremity, moravians,
Nearest to scale: casing, forges, azar, nb, declination, rawalpindi, mainframes, cooling,
Nearest to joseph: psalter, bangor, moyen, diversions, leases, naturalist, rilke, schell,
Epoch 4/10 Iteration: 14100 Avg. Training loss: 4.0575 0.4627 sec/batch
Epoch 4/10 Iteration: 14200 Avg. Training loss: 4.0863 0.4549 sec/batch
Epoch 4/10 Iteration: 14300 Avg. Training loss: 4.1094 0.4551 sec/batch
Epoch 4/10 Iteration: 14400 Avg. Training loss: 4.0598 0.4589 sec/batch
Epoch 4/10 Iteration: 14500 Avg. Training loss: 4.0856 0.4574 sec/batch
Epoch 4/10 Iteration: 14600 Avg. Training loss: 4.0379 0.4569 sec/batch
Epoch 4/10 Iteration: 14700 Avg. Training loss: 4.0873 0.4506 sec/batch
Epoch 4/10 Iteration: 14800 Avg. Training loss: 4.0640 0.4462 sec/batch
Epoch 4/10 Iteration: 14900 Avg. Training loss: 4.1192 0.4379 sec/batch
Epoch 4/10 Iteration: 15000 Avg. Training loss: 4.0288 0.4370 sec/batch
Nearest to all: canceled, prozac, summation, requirement, coexisted, eitc, shams, roc,
Nearest to six: five, three, seven, median, nine, capita, zero, one,
Nearest to their: either, however, maghrib, compounded, lethal, reassert, tasman, bounty,
Nearest to also: are, some, encompasses, the, cousteau, restructure, transports, faceted,
Nearest to into: batavia, accordingly, codebase, hewn, repairs, refilled, airstream, quick,
Nearest to some: also, sip, etc, have, ip, professed, pty, be,
Nearest to called: emporis, known, as, or, cloisters, hangul, collectively, prevenient,
Nearest to in: the, nanobots, petrograd, participating, th, principally, opold, breached,
Nearest to consists: rho, censorial, unleavened, grt, schists, parliaments, shops, guildhall,
Nearest to numerous: arrangements, coelacanth, slacks, dwellers, bun, mcdermott, featuring, responsiveness,
Nearest to know: we, make, fal, ask, breslau, demeaning, markets, diagnoses,
Nearest to versions: camcorder, mapped, disturbed, multiverses, operating, roms, dvb, hoe,
Nearest to shows: hapgood, encompasses, armalite, lodging, integrative, instantaneously, akm, diomedes,
Nearest to orthodox: apostolic, transept, codepoint, lutheran, thema, moravians, extremity, eucharist,
Nearest to scale: casing, forges, azar, nb, declination, rawalpindi, guth, cooling,
Nearest to joseph: bangor, psalter, moyen, schell, diversions, leases, octavian, naturalist,
Epoch 4/10 Iteration: 15100 Avg. Training loss: 4.0299 0.4442 sec/batch
Epoch 4/10 Iteration: 15200 Avg. Training loss: 4.0388 0.4388 sec/batch
Epoch 4/10 Iteration: 15300 Avg. Training loss: 4.0304 0.4356 sec/batch
Epoch 4/10 Iteration: 15400 Avg. Training loss: 4.0384 0.4372 sec/batch
Epoch 4/10 Iteration: 15500 Avg. Training loss: 4.0717 0.4378 sec/batch
Epoch 4/10 Iteration: 15600 Avg. Training loss: 4.0318 0.4378 sec/batch
Epoch 4/10 Iteration: 15700 Avg. Training loss: 4.0665 0.4363 sec/batch
Epoch 4/10 Iteration: 15800 Avg. Training loss: 4.1176 0.4382 sec/batch
Epoch 4/10 Iteration: 15900 Avg. Training loss: 4.0223 0.4395 sec/batch
Epoch 4/10 Iteration: 16000 Avg. Training loss: 4.0486 0.4380 sec/batch
Nearest to all: canceled, summation, prozac, requirement, eitc, coexisted, shams, unfavorable,
Nearest to six: five, three, nine, seven, one, median, zero, two,
Nearest to their: either, however, compounded, lethal, crystallographic, tasman, predominate, contravention,
Nearest to also: are, the, encompasses, some, faceted, superstring, cousteau, commonly,
Nearest to into: accordingly, batavia, repairs, codebase, faxing, refilled, hewn, airstream,
Nearest to some: also, etc, sip, be, succesor, pty, have, ip,
Nearest to called: emporis, as, known, or, hangul, collectively, strictly, cloisters,
Nearest to in: the, electronegative, originate, tok, by, opold, nanobots, of,
Nearest to consists: rho, schists, or, parliaments, grt, censorial, unleavened, shops,
Nearest to numerous: coelacanth, arrangements, slacks, bun, mcdermott, dwellers, branding, featuring,
Nearest to know: we, make, fal, ask, breslau, ionization, demeaning, markets,
Nearest to versions: camcorder, disturbed, multiverses, cdpd, mapped, wise, quadra, roms,
Nearest to shows: hapgood, encompasses, armalite, lodging, instantaneously, integrative, akm, vai,
Nearest to orthodox: apostolic, lutheran, transept, matsui, oriental, moravians, extremity, catholic,
Nearest to scale: forges, nb, casing, azar, rawalpindi, declination, guth, lemnian,
Nearest to joseph: psalter, bangor, moyen, schell, leases, diversions, octavian, naturalist,
Epoch 4/10 Iteration: 16100 Avg. Training loss: 4.0767 0.4433 sec/batch
Epoch 4/10 Iteration: 16200 Avg. Training loss: 4.0862 0.4337 sec/batch
Epoch 4/10 Iteration: 16300 Avg. Training loss: 4.0941 0.4262 sec/batch
Epoch 4/10 Iteration: 16400 Avg. Training loss: 4.0558 0.4112 sec/batch
Epoch 4/10 Iteration: 16500 Avg. Training loss: 4.0671 0.4137 sec/batch
Epoch 4/10 Iteration: 16600 Avg. Training loss: 4.0617 0.4140 sec/batch
Epoch 4/10 Iteration: 16700 Avg. Training loss: 4.0554 0.4127 sec/batch
Epoch 4/10 Iteration: 16800 Avg. Training loss: 4.0659 0.4153 sec/batch
Epoch 4/10 Iteration: 16900 Avg. Training loss: 4.0881 0.4159 sec/batch
Epoch 4/10 Iteration: 17000 Avg. Training loss: 4.0577 0.4139 sec/batch
Nearest to all: summation, canceled, prozac, coexisted, requirement, eitc, unfavorable, groups,
Nearest to six: five, three, nine, seven, one, zero, median, capita,
Nearest to their: either, however, compounded, reassert, contravention, maghrib, decided, predominate,
Nearest to also: some, are, encompasses, the, restructure, superstring, faceted, cousteau,
Nearest to into: batavia, accordingly, repairs, hewn, codebase, refilled, upward, airstream,
Nearest to some: also, etc, succesor, sip, have, chanting, be, professed,
Nearest to called: emporis, as, known, prevenient, messiah, hangul, strictly, or,
Nearest to in: the, opold, of, petrograd, kemal, discussions, th, by,
Nearest to consists: rho, schists, parliaments, grt, guildhall, shops, unleavened, censorial,
Nearest to numerous: coelacanth, arrangements, slacks, bun, mcdermott, dwellers, haggard, prairie,
Nearest to know: we, fal, ask, make, word, breslau, diagnoses, demeaning,
Nearest to versions: camcorder, disturbed, multiverses, cdpd, wise, mapped, hoe, unlocks,
Nearest to shows: hapgood, encompasses, armalite, lodging, instantaneously, integrative, transferable, akm,
Nearest to orthodox: apostolic, matsui, extremity, transept, codepoint, moravians, thema, oriental,
Nearest to scale: forges, nb, azar, casing, rawalpindi, declination, cooling, guth,
Nearest to joseph: bangor, psalter, moyen, schell, diversions, leases, octavian, naturalist,
Epoch 4/10 Iteration: 17100 Avg. Training loss: 4.0509 0.4266 sec/batch
Epoch 4/10 Iteration: 17200 Avg. Training loss: 4.0063 0.4166 sec/batch
Epoch 4/10 Iteration: 17300 Avg. Training loss: 4.0595 0.4158 sec/batch
Epoch 4/10 Iteration: 17400 Avg. Training loss: 4.0761 0.4181 sec/batch
Epoch 4/10 Iteration: 17500 Avg. Training loss: 4.0636 0.4519 sec/batch
Epoch 4/10 Iteration: 17600 Avg. Training loss: 4.1394 0.4351 sec/batch
Epoch 4/10 Iteration: 17700 Avg. Training loss: 4.1271 0.4105 sec/batch
Epoch 4/10 Iteration: 17800 Avg. Training loss: 4.0553 0.4135 sec/batch
Epoch 4/10 Iteration: 17900 Avg. Training loss: 4.0203 0.4129 sec/batch
Epoch 4/10 Iteration: 18000 Avg. Training loss: 4.0471 0.4116 sec/batch
Nearest to all: canceled, requirement, summation, prozac, coexisted, eitc, baasha, shams,
Nearest to six: five, three, seven, zero, nine, one, two, four,
Nearest to their: either, however, contravention, maghrib, reassert, decided, bounty, compounded,
Nearest to also: some, are, the, encompasses, faceted, restructure, superstring, cousteau,
Nearest to into: batavia, accordingly, codebase, repairs, hewn, upward, quick, stp,
Nearest to some: also, etc, succesor, sip, be, have, ip, as,
Nearest to called: emporis, known, as, hangul, cloisters, strictly, messiah, prevenient,
Nearest to in: the, petrograd, originate, discussions, compactly, electronegative, of, opold,
Nearest to consists: rho, schists, parliaments, grt, shops, unleavened, guildhall, or,
Nearest to numerous: coelacanth, arrangements, slacks, haggard, bun, prairie, pathetic, mcdermott,
Nearest to know: we, fal, make, ask, ionization, breslau, word, you,
Nearest to versions: camcorder, disturbed, cdpd, multiverses, wise, unlocks, quadra, roms,
Nearest to shows: hapgood, lodging, encompasses, armalite, transferable, instantaneously, akm, vai,
Nearest to orthodox: apostolic, matsui, thema, moravians, extremity, transept, codepoint, numbering,
Nearest to scale: forges, nb, azar, casing, rawalpindi, constans, guth, bravery,
Nearest to joseph: psalter, bangor, moyen, schell, leases, diversions, nikita, rilke,
Epoch 4/10 Iteration: 18100 Avg. Training loss: 3.9846 0.4199 sec/batch
Epoch 4/10 Iteration: 18200 Avg. Training loss: 4.0377 0.4154 sec/batch
Epoch 4/10 Iteration: 18300 Avg. Training loss: 4.0243 0.4188 sec/batch
Epoch 4/10 Iteration: 18400 Avg. Training loss: 4.0334 0.4183 sec/batch
Epoch 4/10 Iteration: 18500 Avg. Training loss: 4.0687 0.4167 sec/batch
Epoch 5/10 Iteration: 18600 Avg. Training loss: 3.9931 0.4170 sec/batch
Epoch 5/10 Iteration: 18700 Avg. Training loss: 3.9909 0.4175 sec/batch
Epoch 5/10 Iteration: 18800 Avg. Training loss: 3.9813 0.4155 sec/batch
Epoch 5/10 Iteration: 18900 Avg. Training loss: 4.0213 0.4151 sec/batch
Epoch 5/10 Iteration: 19000 Avg. Training loss: 3.9762 0.4165 sec/batch
Nearest to all: canceled, requirement, summation, eitc, prozac, baasha, coexisted, lennon,
Nearest to six: five, three, seven, one, zero, nine, four, eight,
Nearest to their: maghrib, either, however, reassert, muhi, decided, tasman, transsexuals,
Nearest to also: some, encompasses, are, the, restructure, superstring, faceted, cousteau,
Nearest to into: batavia, codebase, accordingly, upward, airstream, quick, repairs, refilled,
Nearest to some: also, be, etc, ip, have, sip, professed, succesor,
Nearest to called: as, emporis, known, prevenient, cloisters, or, messiah, hangul,
Nearest to in: the, petrograd, of, winchester, by, opold, supremacy, principally,
Nearest to consists: rho, schists, parliaments, unleavened, shops, grt, guildhall, stat,
Nearest to numerous: coelacanth, slacks, dwellers, arrangements, prairie, bun, mcdermott, featuring,
Nearest to know: we, ask, make, fal, breslau, ionization, demeaning, you,
Nearest to versions: camcorder, disturbed, cdpd, mapped, wise, operating, quadra, multiverses,
Nearest to shows: hapgood, encompasses, lodging, transferable, instantaneously, armalite, integrative, akm,
Nearest to orthodox: apostolic, moravians, matsui, extremity, thema, transept, codepoint, apostolicae,
Nearest to scale: forges, nb, azar, casing, lemnian, constans, apparent, declination,
Nearest to joseph: bangor, psalter, moyen, schell, diversions, leases, nikita, rilke,
Epoch 5/10 Iteration: 19100 Avg. Training loss: 3.9786 0.4238 sec/batch
Epoch 5/10 Iteration: 19200 Avg. Training loss: 3.9821 0.4130 sec/batch
Epoch 5/10 Iteration: 19300 Avg. Training loss: 4.0143 0.4114 sec/batch
Epoch 5/10 Iteration: 19400 Avg. Training loss: 4.0263 0.4172 sec/batch
Epoch 5/10 Iteration: 19500 Avg. Training loss: 4.0487 0.4180 sec/batch
Epoch 5/10 Iteration: 19600 Avg. Training loss: 3.9965 0.4154 sec/batch
Epoch 5/10 Iteration: 19700 Avg. Training loss: 3.9233 0.4159 sec/batch
Epoch 5/10 Iteration: 19800 Avg. Training loss: 3.9728 0.4164 sec/batch
Epoch 5/10 Iteration: 19900 Avg. Training loss: 3.9342 0.4149 sec/batch
Epoch 5/10 Iteration: 20000 Avg. Training loss: 3.9793 0.4155 sec/batch
Nearest to all: requirement, summation, canceled, eitc, coexisted, prozac, baasha, exactly,
Nearest to six: five, three, zero, seven, nine, one, four, two,
Nearest to their: either, however, reassert, contravention, decided, maghrib, muhi, bounty,
Nearest to also: some, are, the, encompasses, of, restructure, frequently, superstring,
Nearest to into: batavia, accordingly, codebase, shared, repairs, hewn, upward, quickly,
Nearest to some: also, be, have, etc, succesor, professed, as, sip,
Nearest to called: as, emporis, known, or, hangul, messiah, cloisters, collectively,
Nearest to in: the, of, by, petrograd, principally, electronegative, stampede, as,
Nearest to consists: rho, schists, or, parliaments, grt, unleavened, shops, guildhall,
Nearest to numerous: coelacanth, slacks, prairie, arrangements, dwellers, bun, featuring, branding,
Nearest to know: we, make, ask, ionization, fal, breslau, word, tlatelolco,
Nearest to versions: camcorder, disturbed, wise, cdpd, multiverses, operating, unlocks, mapped,
Nearest to shows: hapgood, encompasses, instantaneously, lodging, transferable, armalite, integrative, vai,
Nearest to orthodox: matsui, apostolic, moravians, transept, extremity, thema, apostolicae, catholic,
Nearest to scale: forges, nb, azar, constans, rawalpindi, declination, casing, lemnian,
Nearest to joseph: psalter, bangor, moyen, diversions, leases, schell, nikita, octavian,
Epoch 5/10 Iteration: 20100 Avg. Training loss: 4.0169 0.4215 sec/batch
Epoch 5/10 Iteration: 20200 Avg. Training loss: 3.9768 0.4200 sec/batch
Epoch 5/10 Iteration: 20300 Avg. Training loss: 3.9896 0.4184 sec/batch
Epoch 5/10 Iteration: 20400 Avg. Training loss: 4.0251 0.4189 sec/batch
Epoch 5/10 Iteration: 20500 Avg. Training loss: 4.0398 0.4196 sec/batch
Epoch 5/10 Iteration: 20600 Avg. Training loss: 3.9432 0.4197 sec/batch
Epoch 5/10 Iteration: 20700 Avg. Training loss: 3.9999 0.4185 sec/batch
Epoch 5/10 Iteration: 20800 Avg. Training loss: 4.0175 0.4189 sec/batch
Epoch 5/10 Iteration: 20900 Avg. Training loss: 3.9763 0.4239 sec/batch
Epoch 5/10 Iteration: 21000 Avg. Training loss: 4.0030 0.4184 sec/batch
Nearest to all: summation, coexisted, requirement, eitc, prozac, canceled, of, baasha,
Nearest to six: five, three, one, seven, zero, nine, four, eight,
Nearest to their: either, however, reassert, decided, contravention, crystallographic, transsexuals, counterpart,
Nearest to also: some, the, are, encompasses, of, restructure, frequently, commonly,
Nearest to into: batavia, accordingly, codebase, shared, hewn, detonators, repairs, upward,
Nearest to some: also, be, have, as, succesor, etc, professed, ip,
Nearest to called: as, emporis, known, or, messiah, hangul, strictly, cloisters,
Nearest to in: the, of, petrograd, discussions, by, as, classified, cup,
Nearest to consists: rho, schists, parliaments, unleavened, or, grt, shops, thirds,
Nearest to numerous: coelacanth, slacks, prairie, featuring, arrangements, dwellers, responsiveness, branding,
Nearest to know: we, make, ask, ionization, fal, you, breslau, markets,
Nearest to versions: camcorder, wise, disturbed, rulebooks, cdpd, multiverses, supers, mapped,
Nearest to shows: hapgood, encompasses, lodging, instantaneously, armalite, transferable, ever, vai,
Nearest to orthodox: matsui, oriental, extremity, apostolic, moravians, transept, codepoint, numbering,
Nearest to scale: nb, forges, azar, rawalpindi, casing, constans, declination, lemnian,
Nearest to joseph: psalter, moyen, bangor, schell, diversions, leases, guericke, nikita,
Epoch 5/10 Iteration: 21100 Avg. Training loss: 3.9910 0.4281 sec/batch
Epoch 5/10 Iteration: 21200 Avg. Training loss: 3.9819 0.4214 sec/batch
Epoch 5/10 Iteration: 21300 Avg. Training loss: 3.9745 0.4197 sec/batch
Epoch 5/10 Iteration: 21400 Avg. Training loss: 4.0082 0.4216 sec/batch
Epoch 5/10 Iteration: 21500 Avg. Training loss: 4.0075 0.4215 sec/batch
Epoch 5/10 Iteration: 21600 Avg. Training loss: 4.0193 0.4206 sec/batch
Epoch 5/10 Iteration: 21700 Avg. Training loss: 4.0041 0.4462 sec/batch
Epoch 5/10 Iteration: 21800 Avg. Training loss: 3.9593 0.4197 sec/batch
Epoch 5/10 Iteration: 21900 Avg. Training loss: 3.9595 0.4247 sec/batch
Epoch 5/10 Iteration: 22000 Avg. Training loss: 4.0007 0.4445 sec/batch
Nearest to all: requirement, coexisted, eitc, summation, canceled, groups, prozac, group,
Nearest to six: five, one, three, zero, seven, nine, four, two,
Nearest to their: however, either, decided, crystallographic, reassert, contravention, populate, transsexuals,
Nearest to also: some, the, are, of, encompasses, restructure, other, commonly,
Nearest to into: batavia, codebase, accordingly, shared, hewn, the, visualized, quickly,
Nearest to some: also, have, be, as, the, sip, ip, succesor,
Nearest to called: as, known, emporis, messiah, or, prevenient, strictly, cloisters,
Nearest to in: the, of, petrograd, by, has, as, discussions, to,
Nearest to consists: rho, parliaments, schists, unleavened, or, thirds, shops, grt,
Nearest to numerous: coelacanth, slacks, prairie, arrangements, fantastical, dwellers, mcdermott, bun,
Nearest to know: we, ask, make, you, fal, breslau, word, ionization,
Nearest to versions: camcorder, wise, rulebooks, disturbed, cdpd, multiverses, operating, supers,
Nearest to shows: hapgood, lodging, transferable, encompasses, instantaneously, armalite, ever, vai,
Nearest to orthodox: matsui, moravians, oriental, apostolic, transept, extremity, apostolicae, consubstantiation,
Nearest to scale: forges, nb, azar, rawalpindi, bravery, declination, casing, apparent,
Nearest to joseph: psalter, bangor, moyen, schell, diversions, leases, nikita, guericke,
Epoch 5/10 Iteration: 22100 Avg. Training loss: 3.9614 0.4413 sec/batch
Epoch 5/10 Iteration: 22200 Avg. Training loss: 4.0669 0.4453 sec/batch
Epoch 5/10 Iteration: 22300 Avg. Training loss: 4.0785 0.4442 sec/batch
Epoch 5/10 Iteration: 22400 Avg. Training loss: 4.0217 0.4282 sec/batch
Epoch 5/10 Iteration: 22500 Avg. Training loss: 3.9426 0.4435 sec/batch
Epoch 5/10 Iteration: 22600 Avg. Training loss: 3.9746 0.4332 sec/batch
Epoch 5/10 Iteration: 22700 Avg. Training loss: 3.9705 0.4176 sec/batch
Epoch 5/10 Iteration: 22800 Avg. Training loss: 3.9352 0.4191 sec/batch
Epoch 5/10 Iteration: 22900 Avg. Training loss: 4.0117 0.4389 sec/batch
Epoch 5/10 Iteration: 23000 Avg. Training loss: 3.9596 0.4330 sec/batch
Nearest to all: canceled, dubbed, eitc, coexisted, requirement, summation, prozac, instructors,
Nearest to six: five, one, seven, zero, four, three, nine, two,
Nearest to their: decided, maghrib, crystallographic, transsexuals, reassert, however, bounty, contravention,
Nearest to also: some, the, are, of, encompasses, restructure, superstring, frequently,
Nearest to into: batavia, codebase, hewn, accordingly, quickly, upward, the, shared,
Nearest to some: also, have, be, as, sip, ip, the, etc,
Nearest to called: as, known, emporis, or, messiah, strictly, now, cloisters,
Nearest to in: the, of, as, petrograd, by, to, stampede, after,
Nearest to consists: parliaments, schists, rho, or, unleavened, thirds, shops, grt,
Nearest to numerous: slacks, dwellers, coelacanth, prairie, arrangements, featuring, mcdermott, responsiveness,
Nearest to know: we, ask, you, make, fal, breslau, word, impersonated,
Nearest to versions: camcorder, wise, operating, cdpd, disturbed, rulebooks, multiverses, supers,
Nearest to shows: hapgood, lodging, transferable, encompasses, armalite, instantaneously, ever, indri,
Nearest to orthodox: matsui, thema, moravians, numbering, extremity, apostolic, oriental, transept,
Nearest to scale: forges, nb, azar, rawalpindi, bravery, nims, lemnian, declination,
Nearest to joseph: psalter, moyen, bangor, schell, nikita, diversions, leases, guericke,
Epoch 5/10 Iteration: 23100 Avg. Training loss: 4.0041 0.4184 sec/batch
Epoch 6/10 Iteration: 23200 Avg. Training loss: 3.9771 0.3099 sec/batch
Epoch 6/10 Iteration: 23300 Avg. Training loss: 3.9702 0.4135 sec/batch
Epoch 6/10 Iteration: 23400 Avg. Training loss: 3.9015 0.4128 sec/batch
Epoch 6/10 Iteration: 23500 Avg. Training loss: 3.9675 0.4105 sec/batch
Epoch 6/10 Iteration: 23600 Avg. Training loss: 3.9305 0.4121 sec/batch
Epoch 6/10 Iteration: 23700 Avg. Training loss: 3.9568 0.4128 sec/batch
Epoch 6/10 Iteration: 23800 Avg. Training loss: 3.8636 0.4131 sec/batch
Epoch 6/10 Iteration: 23900 Avg. Training loss: 4.0091 0.4114 sec/batch
Epoch 6/10 Iteration: 24000 Avg. Training loss: 3.9520 0.4123 sec/batch
Nearest to all: coexisted, eitc, dubbed, requirement, summation, baasha, lugh, canceled,
Nearest to six: five, one, three, seven, zero, four, nine, two,
Nearest to their: decided, reassert, either, crystallographic, transsexuals, muhi, maghrib, however,
Nearest to also: some, the, are, of, encompasses, commonly, superstring, often,
Nearest to into: batavia, codebase, shared, visualized, quickly, upward, the, accordingly,
Nearest to some: also, have, be, as, professed, chanting, counsel, others,
Nearest to called: as, known, emporis, messiah, or, now, strictly, prevenient,
Nearest to in: the, of, as, after, to, petrograd, with, stampede,
Nearest to consists: unleavened, parliaments, rho, thirds, schists, or, shops, grt,
Nearest to numerous: dwellers, slacks, coelacanth, prairie, featuring, arrangements, mcdermott, fantastical,
Nearest to know: we, ask, you, make, fal, breslau, word, impersonated,
Nearest to versions: camcorder, wise, disturbed, multiverses, cdpd, rulebooks, operating, supers,
Nearest to shows: hapgood, encompasses, transferable, armalite, lodging, indri, audiences, instantaneously,
Nearest to orthodox: matsui, apostolic, oriental, moravians, thema, apostolicae, extremity, numbering,
Nearest to scale: nb, forges, azar, apparent, bravery, declination, lemnian, definable,
Nearest to joseph: psalter, moyen, bangor, nikita, leases, schell, diversions, guericke,
Epoch 6/10 Iteration: 24100 Avg. Training loss: 3.9465 0.4192 sec/batch
Epoch 6/10 Iteration: 24200 Avg. Training loss: 3.9653 0.4143 sec/batch
Epoch 6/10 Iteration: 24300 Avg. Training loss: 3.8422 0.4919 sec/batch
Epoch 6/10 Iteration: 24400 Avg. Training loss: 3.9066 0.5100 sec/batch
Epoch 6/10 Iteration: 24500 Avg. Training loss: 3.8979 0.4510 sec/batch
Epoch 6/10 Iteration: 24600 Avg. Training loss: 3.9237 0.4257 sec/batch
Epoch 6/10 Iteration: 24700 Avg. Training loss: 3.9366 0.4387 sec/batch
Epoch 6/10 Iteration: 24800 Avg. Training loss: 3.9533 0.4353 sec/batch
Epoch 6/10 Iteration: 24900 Avg. Training loss: 3.9218 0.4369 sec/batch
Epoch 6/10 Iteration: 25000 Avg. Training loss: 3.9440 0.4489 sec/batch
Nearest to all: coexisted, of, eitc, dubbed, requirement, palm, impeached, baasha,
Nearest to six: five, one, zero, four, three, seven, nine, eight,
Nearest to their: either, however, crystallographic, reassert, contravention, transsexuals, decided, to,
Nearest to also: some, the, are, of, encompasses, other, commonly, often,
Nearest to into: batavia, upward, shared, the, visualized, codebase, ganglion, prometheus,
Nearest to some: also, be, have, as, professed, chanting, for, succesor,
Nearest to called: as, known, emporis, or, messiah, now, strictly, collectively,
Nearest to in: the, of, as, stampede, petrograd, electronegative, after, by,
Nearest to consists: or, schists, rho, thirds, unleavened, parliaments, shops, grt,
Nearest to numerous: coelacanth, prairie, slacks, dwellers, visconti, mcdermott, arrangements, fantastical,
Nearest to know: we, you, ask, make, breslau, fal, ionization, word,
Nearest to versions: wise, camcorder, disturbed, multiverses, rulebooks, cdpd, hammurabi, operating,
Nearest to shows: hapgood, transferable, encompasses, armalite, lodging, instantaneously, television, show,
Nearest to orthodox: matsui, apostolic, oriental, moravians, thema, apostolicae, catholic, extremity,
Nearest to scale: nb, forges, azar, declination, bravery, apparent, definable, constans,
Nearest to joseph: psalter, moyen, bangor, nikita, schell, leases, diversions, operated,
Epoch 6/10 Iteration: 25100 Avg. Training loss: 3.9901 0.4468 sec/batch
Epoch 6/10 Iteration: 25200 Avg. Training loss: 3.9441 0.4219 sec/batch
Epoch 6/10 Iteration: 25300 Avg. Training loss: 3.9230 0.4209 sec/batch
Epoch 6/10 Iteration: 25400 Avg. Training loss: 3.9565 0.4344 sec/batch
Epoch 6/10 Iteration: 25500 Avg. Training loss: 3.9355 0.4237 sec/batch
Epoch 6/10 Iteration: 25600 Avg. Training loss: 3.9389 0.4695 sec/batch
Epoch 6/10 Iteration: 25700 Avg. Training loss: 3.9500 0.5055 sec/batch
Epoch 6/10 Iteration: 25800 Avg. Training loss: 3.9478 0.5012 sec/batch
Epoch 6/10 Iteration: 25900 Avg. Training loss: 3.9345 0.4907 sec/batch
Epoch 6/10 Iteration: 26000 Avg. Training loss: 3.9778 0.5126 sec/batch
Nearest to all: eitc, dubbed, coexisted, requirement, summation, impeached, baasha, groups,
Nearest to six: five, one, zero, four, three, seven, nine, two,
Nearest to their: crystallographic, however, either, decided, reassert, contravention, tasman, breaking,
Nearest to also: some, the, of, are, encompasses, in, superstring, frequently,
Nearest to into: batavia, shared, the, codebase, prometheus, visualized, upward, accordingly,
Nearest to some: also, have, be, as, professed, succesor, for, pty,
Nearest to called: as, messiah, emporis, known, or, strictly, now, name,
Nearest to in: the, of, as, by, to, petrograd, after, stampede,
Nearest to consists: schists, rho, thirds, or, grt, unicameral, parliaments, unleavened,
Nearest to numerous: coelacanth, prairie, dwellers, slacks, fantastical, mcdermott, study, common,
Nearest to know: we, you, ask, make, fal, breslau, word, behaviorist,
Nearest to versions: wise, camcorder, disturbed, multiverses, rulebooks, cdpd, hammurabi, supers,
Nearest to shows: hapgood, transferable, encompasses, show, armalite, television, lodging, nou,
Nearest to orthodox: matsui, oriental, apostolic, moravians, extremity, codepoint, thema, consubstantiation,
Nearest to scale: forges, nb, declination, azar, apparent, bravery, definable, transposed,
Nearest to joseph: psalter, nikita, bangor, schell, moyen, leases, diversions, operated,
Epoch 6/10 Iteration: 26100 Avg. Training loss: 3.9490 0.5058 sec/batch
Epoch 6/10 Iteration: 26200 Avg. Training loss: 3.9484 0.4950 sec/batch
Epoch 6/10 Iteration: 26300 Avg. Training loss: 3.9389 0.4948 sec/batch
Epoch 6/10 Iteration: 26400 Avg. Training loss: 3.9140 0.5051 sec/batch
Epoch 6/10 Iteration: 26500 Avg. Training loss: 3.9229 0.4778 sec/batch
Epoch 6/10 Iteration: 26600 Avg. Training loss: 3.9516 0.4165 sec/batch
Epoch 6/10 Iteration: 26700 Avg. Training loss: 3.9181 0.4201 sec/batch
Epoch 6/10 Iteration: 26800 Avg. Training loss: 4.0316 0.4123 sec/batch
Epoch 6/10 Iteration: 26900 Avg. Training loss: 3.9878 0.4142 sec/batch
Epoch 6/10 Iteration: 27000 Avg. Training loss: 4.0146 0.4081 sec/batch
Nearest to all: dubbed, eitc, baasha, lugh, coexisted, palm, of, requirement,
Nearest to six: five, one, four, zero, seven, three, nine, two,
Nearest to their: reassert, contravention, decided, crystallographic, however, muhi, either, transsexuals,
Nearest to also: some, the, of, are, other, encompasses, grapheme, to,
Nearest to into: batavia, the, codebase, upward, prometheus, visualized, quickly, shared,
Nearest to some: also, have, be, as, professed, succesor, the, are,
Nearest to called: as, messiah, emporis, known, or, now, strictly, name,
Nearest to in: the, of, as, after, by, to, petrograd, stampede,
Nearest to consists: schists, or, rho, thirds, unleavened, unicameral, manna, grt,
Nearest to numerous: prairie, coelacanth, dwellers, visconti, slacks, fantastical, featuring, mcdermott,
Nearest to know: we, you, ask, make, fal, breslau, impersonated, behaviorist,
Nearest to versions: camcorder, wise, disturbed, rulebooks, multiverses, cdpd, hammurabi, supers,
Nearest to shows: hapgood, transferable, lodging, encompasses, ever, television, armalite, nou,
Nearest to orthodox: matsui, moravians, thema, oriental, apostolicae, apostolic, extremity, numbering,
Nearest to scale: forges, nb, bravery, azar, apparent, rawalpindi, declination, definable,
Nearest to joseph: nikita, bangor, moyen, psalter, schell, leases, operated, diversions,
Epoch 6/10 Iteration: 27100 Avg. Training loss: 3.9169 0.4203 sec/batch
Epoch 6/10 Iteration: 27200 Avg. Training loss: 3.8786 0.4600 sec/batch
Epoch 6/10 Iteration: 27300 Avg. Training loss: 4.0042 0.5217 sec/batch
Epoch 6/10 Iteration: 27400 Avg. Training loss: 3.8696 0.5055 sec/batch
Epoch 6/10 Iteration: 27500 Avg. Training loss: 3.9696 0.5113 sec/batch
Epoch 6/10 Iteration: 27600 Avg. Training loss: 3.9396 0.5040 sec/batch
Epoch 6/10 Iteration: 27700 Avg. Training loss: 3.9439 0.5008 sec/batch
Epoch 7/10 Iteration: 27800 Avg. Training loss: 3.9846 0.2601 sec/batch
Epoch 7/10 Iteration: 27900 Avg. Training loss: 3.8871 0.5112 sec/batch
Epoch 7/10 Iteration: 28000 Avg. Training loss: 3.9039 0.5022 sec/batch
Nearest to all: dubbed, baasha, eitc, coexisted, lugh, lennon, palm, instructors,
Nearest to six: five, one, four, zero, seven, three, eight, two,
Nearest to their: maghrib, crystallographic, decided, however, these, reassert, muhi, transsexuals,
Nearest to also: some, the, of, are, encompasses, in, other, superstring,
Nearest to into: batavia, upward, codebase, the, from, prometheus, visualized, quickly,
Nearest to some: also, have, be, as, professed, others, the, to,
Nearest to called: as, messiah, or, known, emporis, strictly, now, name,
Nearest to in: the, of, as, to, and, by, petrograd, after,
Nearest to consists: schists, unicameral, thirds, rho, parliaments, unleavened, grt, or,
Nearest to numerous: prairie, dwellers, coelacanth, slacks, fantastical, arrangements, study, visconti,
Nearest to know: we, you, ask, make, breslau, fal, posthuman, diagnoses,
Nearest to versions: camcorder, wise, rulebooks, operating, disturbed, cdpd, multiverses, documented,
Nearest to shows: hapgood, transferable, lodging, encompasses, show, television, vlaanderen, integrating,
Nearest to orthodox: matsui, moravians, apostolicae, thema, oriental, apostolic, extremity, numbering,
Nearest to scale: nb, forges, bravery, transposed, azar, definable, declination, lemnian,
Nearest to joseph: nikita, bangor, moyen, psalter, schell, leases, diversions, operated,
Epoch 7/10 Iteration: 28100 Avg. Training loss: 3.9202 0.5145 sec/batch
Epoch 7/10 Iteration: 28200 Avg. Training loss: 3.9537 0.5060 sec/batch
Epoch 7/10 Iteration: 28300 Avg. Training loss: 3.9032 0.5153 sec/batch
Epoch 7/10 Iteration: 28400 Avg. Training loss: 3.8928 0.5064 sec/batch
Epoch 7/10 Iteration: 28500 Avg. Training loss: 3.8668 0.4653 sec/batch
Epoch 7/10 Iteration: 28600 Avg. Training loss: 3.9137 0.4500 sec/batch
Epoch 7/10 Iteration: 28700 Avg. Training loss: 3.9130 0.4551 sec/batch
Epoch 7/10 Iteration: 28800 Avg. Training loss: 3.9174 0.4512 sec/batch
Epoch 7/10 Iteration: 28900 Avg. Training loss: 3.8474 0.4348 sec/batch
Epoch 7/10 Iteration: 29000 Avg. Training loss: 3.8814 0.4733 sec/batch
Nearest to all: of, eitc, coexisted, dubbed, baasha, in, requirement, a,
Nearest to six: five, one, zero, four, seven, three, two, eight,
Nearest to their: these, however, either, they, crystallographic, reassert, maghrib, contravention,
Nearest to also: the, some, of, in, are, other, encompasses, to,
Nearest to into: batavia, the, from, upward, which, shared, continents, visualized,
Nearest to some: also, have, be, as, professed, to, the, are,
Nearest to called: as, messiah, known, now, or, emporis, name, the,
Nearest to in: the, of, to, and, as, after, also, by,
Nearest to consists: schists, thirds, unicameral, or, unleavened, manna, shops, rho,
Nearest to numerous: prairie, coelacanth, dwellers, arrangements, visconti, slacks, branding, study,
Nearest to know: we, you, ask, make, breslau, fal, thy, posthuman,
Nearest to versions: camcorder, wise, multiverses, rulebooks, disturbed, operating, cdpd, hammurabi,
Nearest to shows: hapgood, transferable, encompasses, television, lodging, episode, show, vlaanderen,
Nearest to orthodox: matsui, moravians, oriental, thema, apostolicae, apostolic, extremity, catholic,
Nearest to scale: forges, nb, bravery, azar, declination, lemnian, definable, apparent,
Nearest to joseph: nikita, leases, bangor, psalter, schell, moyen, diversions, muni,
Epoch 7/10 Iteration: 29100 Avg. Training loss: 3.8697 0.4623 sec/batch
Epoch 7/10 Iteration: 29200 Avg. Training loss: 3.8423 0.4858 sec/batch
Epoch 7/10 Iteration: 29300 Avg. Training loss: 3.8702 0.4715 sec/batch
Epoch 7/10 Iteration: 29400 Avg. Training loss: 3.9326 0.4696 sec/batch
Epoch 7/10 Iteration: 29500 Avg. Training loss: 3.9079 0.4471 sec/batch
Epoch 7/10 Iteration: 29600 Avg. Training loss: 3.8734 0.4250 sec/batch
Epoch 7/10 Iteration: 29700 Avg. Training loss: 3.9551 0.4315 sec/batch
Epoch 7/10 Iteration: 29800 Avg. Training loss: 3.8991 0.4279 sec/batch
Epoch 7/10 Iteration: 29900 Avg. Training loss: 3.8881 0.4241 sec/batch
Epoch 7/10 Iteration: 30000 Avg. Training loss: 3.9131 0.4293 sec/batch
Nearest to all: of, coexisted, dubbed, baasha, eitc, in, a, palm,
Nearest to six: five, one, zero, four, seven, three, two, nine,
Nearest to their: these, however, they, contravention, either, crystallographic, breaking, decided,
Nearest to also: the, some, of, in, are, other, to, commonly,
Nearest to into: batavia, the, from, visualized, shared, upward, codebase, mesogens,
Nearest to some: also, have, be, as, professed, are, the, for,
Nearest to called: as, messiah, or, the, now, emporis, known, name,
Nearest to in: the, of, as, to, by, after, and, also,
Nearest to consists: or, schists, thirds, unicameral, unleavened, rho, shops, manna,
Nearest to numerous: prairie, coelacanth, dwellers, slacks, fantastical, common, branding, visconti,
Nearest to know: we, you, ask, make, breslau, fal, thy, posthuman,
Nearest to versions: wise, rulebooks, camcorder, disturbed, multiverses, cdpd, documented, hammurabi,
Nearest to shows: hapgood, transferable, episode, show, television, encompasses, from, lodging,
Nearest to orthodox: matsui, oriental, thema, moravians, apostolicae, apostolic, extremity, catholic,
Nearest to scale: nb, forges, azar, bravery, lemnian, transposed, declination, apparent,
Nearest to joseph: nikita, leases, psalter, schell, bangor, moyen, diversions, operated,
Epoch 7/10 Iteration: 30100 Avg. Training loss: 3.9143 0.4333 sec/batch
Epoch 7/10 Iteration: 30200 Avg. Training loss: 3.9183 0.4208 sec/batch
Epoch 7/10 Iteration: 30300 Avg. Training loss: 3.8817 0.4801 sec/batch
Epoch 7/10 Iteration: 30400 Avg. Training loss: 3.8725 0.5241 sec/batch
Epoch 7/10 Iteration: 30500 Avg. Training loss: 3.9331 0.4616 sec/batch
Epoch 7/10 Iteration: 30600 Avg. Training loss: 3.9327 0.4165 sec/batch
Epoch 7/10 Iteration: 30700 Avg. Training loss: 3.8935 0.4222 sec/batch
Epoch 7/10 Iteration: 30800 Avg. Training loss: 3.8931 0.4146 sec/batch
Epoch 7/10 Iteration: 30900 Avg. Training loss: 3.9170 0.4139 sec/batch
Epoch 7/10 Iteration: 31000 Avg. Training loss: 3.8551 0.4178 sec/batch
Nearest to all: of, dubbed, a, in, coexisted, baasha, eitc, but,
Nearest to six: five, one, four, zero, seven, three, two, nine,
Nearest to their: these, however, they, to, contravention, either, were, reassert,
Nearest to also: some, the, of, in, to, are, other, encompasses,
Nearest to into: batavia, the, from, visualized, upward, which, prometheus, and,
Nearest to some: also, have, as, be, are, the, to, for,
Nearest to called: as, messiah, or, now, the, known, name, emporis,
Nearest to in: the, of, to, and, as, by, after, which,
Nearest to consists: schists, unicameral, thirds, unleavened, manna, or, shops, rho,
Nearest to numerous: prairie, coelacanth, dwellers, common, fantastical, arrangements, slacks, jezreel,
Nearest to know: we, you, ask, make, breslau, fal, thy, posthuman,
Nearest to versions: wise, rulebooks, camcorder, disturbed, multiverses, cdpd, documented, jaya,
Nearest to shows: hapgood, transferable, television, encompasses, episode, show, nou, lodging,
Nearest to orthodox: matsui, oriental, thema, apostolicae, apostolic, moravians, catholic, consubstantiation,
Nearest to scale: nb, forges, bravery, azar, lemnian, apparent, declination, transposed,
Nearest to joseph: leases, nikita, schell, bangor, psalter, moyen, diversions, remaking,
Epoch 7/10 Iteration: 31100 Avg. Training loss: 3.8893 0.4248 sec/batch
Epoch 7/10 Iteration: 31200 Avg. Training loss: 3.9482 0.4179 sec/batch
Epoch 7/10 Iteration: 31300 Avg. Training loss: 3.8646 0.4209 sec/batch
Epoch 7/10 Iteration: 31400 Avg. Training loss: 3.9961 0.4171 sec/batch
Epoch 7/10 Iteration: 31500 Avg. Training loss: 3.9738 0.4386 sec/batch
Epoch 7/10 Iteration: 31600 Avg. Training loss: 3.9560 0.4301 sec/batch
Epoch 7/10 Iteration: 31700 Avg. Training loss: 3.8957 0.4328 sec/batch
Epoch 7/10 Iteration: 31800 Avg. Training loss: 3.8799 0.4207 sec/batch
Epoch 7/10 Iteration: 31900 Avg. Training loss: 3.8760 0.4125 sec/batch
Epoch 7/10 Iteration: 32000 Avg. Training loss: 3.8007 0.4232 sec/batch
Nearest to all: of, dubbed, in, baasha, a, eitc, be, single,
Nearest to six: five, zero, one, seven, four, three, two, nine,
Nearest to their: these, were, however, contravention, reassert, they, decided, breaking,
Nearest to also: the, some, in, of, to, are, other, on,
Nearest to into: batavia, the, from, which, upward, visualized, and, codebase,
Nearest to some: also, have, be, as, are, the, for, others,
Nearest to called: as, or, now, the, messiah, known, name, emporis,
Nearest to in: the, of, to, as, and, a, after, also,
Nearest to consists: thirds, unicameral, schists, or, branch, manna, unleavened, grt,
Nearest to numerous: prairie, slacks, coelacanth, dwellers, fantastical, visconti, common, responsiveness,
Nearest to know: we, you, ask, breslau, make, fal, thy, how,
Nearest to versions: wise, rulebooks, camcorder, disturbed, multiverses, documented, cdpd, operating,
Nearest to shows: hapgood, transferable, show, television, lodging, encompasses, margrethe, episode,
Nearest to orthodox: matsui, oriental, thema, moravians, apostolicae, apostolic, numbering, extremity,
Nearest to scale: nb, forges, bravery, azar, declination, abusive, apparent, statistically,
Nearest to joseph: nikita, leases, psalter, bangor, schell, moyen, diversions, subsidized,
Epoch 7/10 Iteration: 32100 Avg. Training loss: 3.9340 0.4234 sec/batch
Epoch 7/10 Iteration: 32200 Avg. Training loss: 3.9045 0.4161 sec/batch
Epoch 7/10 Iteration: 32300 Avg. Training loss: 3.9196 0.4134 sec/batch
Epoch 8/10 Iteration: 32400 Avg. Training loss: 3.9712 0.1045 sec/batch
Epoch 8/10 Iteration: 32500 Avg. Training loss: 3.8813 0.4171 sec/batch
Epoch 8/10 Iteration: 32600 Avg. Training loss: 3.8465 0.4154 sec/batch
Epoch 8/10 Iteration: 32700 Avg. Training loss: 3.8825 0.4186 sec/batch
Epoch 8/10 Iteration: 32800 Avg. Training loss: 3.8917 0.4166 sec/batch
Epoch 8/10 Iteration: 32900 Avg. Training loss: 3.8303 0.4249 sec/batch
Epoch 8/10 Iteration: 33000 Avg. Training loss: 3.8716 0.4231 sec/batch
Nearest to all: dubbed, baasha, in, of, coexisted, single, eitc, and,
Nearest to six: five, one, zero, seven, four, three, eight, two,
Nearest to their: these, were, they, however, muhi, reassert, to, maghrib,
Nearest to also: some, the, in, of, to, other, are, kinetics,
Nearest to into: batavia, the, from, which, upward, mesogens, codebase, visualized,
Nearest to some: also, have, be, as, are, to, others, the,
Nearest to called: as, the, now, or, messiah, name, known, strictly,
Nearest to in: the, of, to, as, after, and, a, by,
Nearest to consists: thirds, unicameral, schists, unleavened, manna, shops, branch, lalr,
Nearest to numerous: dwellers, common, prairie, slacks, coelacanth, fantastical, responsiveness, beren,
Nearest to know: we, you, ask, breslau, make, fal, thank, posthuman,
Nearest to versions: wise, camcorder, documented, cdpd, multiverses, rulebooks, disturbed, operating,
Nearest to shows: hapgood, transferable, encompasses, television, show, vlaanderen, margrethe, lodging,
Nearest to orthodox: matsui, oriental, moravians, thema, apostolicae, apostolic, consubstantiation, numbering,
Nearest to scale: nb, forges, azar, bravery, declination, lemnian, transposed, mato,
Nearest to joseph: nikita, schell, leases, diversions, psalter, bangor, moyen, occupied,
Epoch 8/10 Iteration: 33100 Avg. Training loss: 3.8223 0.4279 sec/batch
Epoch 8/10 Iteration: 33200 Avg. Training loss: 3.8668 0.4178 sec/batch
Epoch 8/10 Iteration: 33300 Avg. Training loss: 3.8929 0.4180 sec/batch
Epoch 8/10 Iteration: 33400 Avg. Training loss: 3.9173 0.4200 sec/batch
Epoch 8/10 Iteration: 33500 Avg. Training loss: 3.8252 0.4258 sec/batch
Epoch 8/10 Iteration: 33600 Avg. Training loss: 3.8141 0.4272 sec/batch
Epoch 8/10 Iteration: 33700 Avg. Training loss: 3.9020 0.4214 sec/batch
Epoch 8/10 Iteration: 33800 Avg. Training loss: 3.7789 0.4205 sec/batch
Epoch 8/10 Iteration: 33900 Avg. Training loss: 3.8444 0.4202 sec/batch
Epoch 8/10 Iteration: 34000 Avg. Training loss: 3.8955 0.4205 sec/batch
Nearest to all: in, of, baasha, the, be, are, a, single,
Nearest to six: five, zero, four, one, seven, three, eight, two,
Nearest to their: they, however, to, these, breaking, contravention, were, and,
Nearest to also: some, the, in, of, are, other, to, for,
Nearest to into: the, batavia, from, which, and, visualized, mesogens, upward,
Nearest to some: also, have, be, as, are, the, for, to,
Nearest to called: as, the, or, now, messiah, known, name, strictly,
Nearest to in: the, of, as, to, and, a, also, by,
Nearest to consists: thirds, or, schists, unicameral, unleavened, clavichord, lalr, manna,
Nearest to numerous: common, dwellers, prairie, coelacanth, slacks, responsiveness, panthers, fantastical,
Nearest to know: we, you, ask, make, breslau, fal, how, thy,
Nearest to versions: wise, camcorder, disturbed, multiverses, rulebooks, documented, cdpd, early,
Nearest to shows: hapgood, television, transferable, encompasses, show, vlaanderen, episode, nou,
Nearest to orthodox: matsui, oriental, apostolicae, moravians, apostolic, thema, catholic, eastern,
Nearest to scale: nb, forges, azar, declination, bravery, mainframes, lemnian, definable,
Nearest to joseph: nikita, leases, psalter, schell, bangor, diversions, moyen, muni,
Epoch 8/10 Iteration: 34100 Avg. Training loss: 3.8591 0.4263 sec/batch
Epoch 8/10 Iteration: 34200 Avg. Training loss: 3.8849 0.4211 sec/batch
Epoch 8/10 Iteration: 34300 Avg. Training loss: 3.9739 0.4180 sec/batch
Epoch 8/10 Iteration: 34400 Avg. Training loss: 3.8374 0.4114 sec/batch
Epoch 8/10 Iteration: 34500 Avg. Training loss: 3.8403 0.4159 sec/batch
Epoch 8/10 Iteration: 34600 Avg. Training loss: 3.8828 0.4192 sec/batch
Epoch 8/10 Iteration: 34700 Avg. Training loss: 3.8868 0.4202 sec/batch
Epoch 8/10 Iteration: 34800 Avg. Training loss: 3.9169 0.4204 sec/batch
Epoch 8/10 Iteration: 34900 Avg. Training loss: 3.8924 0.4204 sec/batch
Epoch 8/10 Iteration: 35000 Avg. Training loss: 3.8710 0.4198 sec/batch
Nearest to all: in, of, be, single, but, a, baasha, dubbed,
Nearest to six: five, one, four, zero, three, seven, eight, nine,
Nearest to their: they, these, to, however, breaking, decided, crystallographic, were,
Nearest to also: some, the, in, of, to, other, are, for,
Nearest to into: batavia, the, from, and, codebase, mesogens, which, visualized,
Nearest to some: also, have, be, are, as, for, the, to,
Nearest to called: as, or, the, now, messiah, name, known, of,
Nearest to in: the, of, to, as, and, also, by, a,
Nearest to consists: thirds, unicameral, schists, or, unleavened, manna, branch, clavichord,
Nearest to numerous: common, dwellers, coelacanth, responsiveness, prairie, slacks, panthers, branding,
Nearest to know: we, you, ask, make, breslau, how, fal, thank,
Nearest to versions: wise, camcorder, rulebooks, documented, multiverses, disturbed, early, vga,
Nearest to shows: television, hapgood, episode, transferable, encompasses, from, show, nou,
Nearest to orthodox: matsui, oriental, apostolicae, moravians, thema, apostolic, floundered, catholic,
Nearest to scale: nb, forges, azar, declination, transposed, bravery, mato, apparent,
Nearest to joseph: nikita, schell, leases, bangor, moyen, psalter, diversions, muni,
Epoch 8/10 Iteration: 35100 Avg. Training loss: 3.8978 0.4246 sec/batch
Epoch 8/10 Iteration: 35200 Avg. Training loss: 3.8593 0.4164 sec/batch
Epoch 8/10 Iteration: 35300 Avg. Training loss: 3.8851 0.4192 sec/batch
Epoch 8/10 Iteration: 35400 Avg. Training loss: 3.8940 0.4139 sec/batch
Epoch 8/10 Iteration: 35500 Avg. Training loss: 3.8800 0.4142 sec/batch
Epoch 8/10 Iteration: 35600 Avg. Training loss: 3.8930 0.4172 sec/batch
Epoch 8/10 Iteration: 35700 Avg. Training loss: 3.8161 0.4176 sec/batch
Epoch 8/10 Iteration: 35800 Avg. Training loss: 3.8656 0.4259 sec/batch
Epoch 8/10 Iteration: 35900 Avg. Training loss: 3.8967 0.4375 sec/batch
Epoch 8/10 Iteration: 36000 Avg. Training loss: 3.9600 0.4348 sec/batch
Nearest to all: of, in, baasha, single, be, the, are, dubbed,
Nearest to six: five, four, one, seven, zero, three, two, eight,
Nearest to their: to, these, they, however, were, contravention, decided, allow,
Nearest to also: some, the, of, in, to, other, and, are,
Nearest to into: batavia, the, from, which, and, codebase, continents, mesogens,
Nearest to some: also, have, be, are, as, the, to, for,
Nearest to called: as, now, or, messiah, the, known, name, emporis,
Nearest to in: the, of, to, as, and, a, after, also,
Nearest to consists: thirds, unicameral, schists, or, unleavened, manna, culminates, shops,
Nearest to numerous: common, coelacanth, slacks, fantastical, dwellers, prairie, responsiveness, panthers,
Nearest to know: we, you, ask, breslau, make, how, fal, thank,
Nearest to versions: wise, rulebooks, documented, camcorder, multiverses, disturbed, vga, ms,
Nearest to shows: hapgood, transferable, television, show, nou, episode, margrethe, from,
Nearest to orthodox: matsui, oriental, moravians, apostolicae, thema, apostolic, numbering, floundered,
Nearest to scale: nb, forges, azar, bravery, declination, transposed, abusive, mainframes,
Nearest to joseph: nikita, schell, occupied, moyen, bangor, leases, psalter, diversions,
Epoch 8/10 Iteration: 36100 Avg. Training loss: 3.9779 0.4377 sec/batch
Epoch 8/10 Iteration: 36200 Avg. Training loss: 3.9325 0.4341 sec/batch
Epoch 8/10 Iteration: 36300 Avg. Training loss: 3.8651 0.4344 sec/batch
Epoch 8/10 Iteration: 36400 Avg. Training loss: 3.8233 0.4348 sec/batch
Epoch 8/10 Iteration: 36500 Avg. Training loss: 3.8875 0.4286 sec/batch
Epoch 8/10 Iteration: 36600 Avg. Training loss: 3.8093 0.4177 sec/batch
Epoch 8/10 Iteration: 36700 Avg. Training loss: 3.8672 0.4228 sec/batch
Epoch 8/10 Iteration: 36800 Avg. Training loss: 3.8815 0.4177 sec/batch
Epoch 8/10 Iteration: 36900 Avg. Training loss: 3.8849 0.4169 sec/batch
Epoch 8/10 Iteration: 37000 Avg. Training loss: 3.9129 0.4171 sec/batch
Nearest to all: in, of, single, are, a, and, has, be,
Nearest to six: five, four, one, seven, zero, three, two, eight,
Nearest to their: these, to, they, were, and, breaking, however, allow,
Nearest to also: the, some, in, of, to, other, and, are,
Nearest to into: the, from, batavia, mesogens, which, and, codebase, visualized,
Nearest to some: also, have, be, are, as, the, for, to,
Nearest to called: as, the, or, now, messiah, known, by, name,
Nearest to in: the, of, as, to, and, a, which, also,
Nearest to consists: unicameral, thirds, or, unleavened, schists, shops, branch, grt,
Nearest to numerous: slacks, common, dwellers, responsiveness, coelacanth, prairie, fantastical, panthers,
Nearest to know: we, you, ask, breslau, make, how, thank, fal,
Nearest to versions: wise, camcorder, rulebooks, vga, documented, ms, operating, cdpd,
Nearest to shows: television, hapgood, show, transferable, episode, margrethe, lodging, vlaanderen,
Nearest to orthodox: matsui, oriental, moravians, thema, apostolicae, numbering, floundered, gloucestershire,
Nearest to scale: nb, forges, transposed, declination, mato, azar, greenwich, bravery,
Nearest to joseph: nikita, bangor, moyen, schell, leases, occupied, diversions, psalter,
Epoch 9/10 Iteration: 37100 Avg. Training loss: 3.8819 0.4155 sec/batch
Epoch 9/10 Iteration: 37200 Avg. Training loss: 3.8671 0.4165 sec/batch
Epoch 9/10 Iteration: 37300 Avg. Training loss: 3.8321 0.4231 sec/batch
Epoch 9/10 Iteration: 37400 Avg. Training loss: 3.8800 0.4168 sec/batch
Epoch 9/10 Iteration: 37500 Avg. Training loss: 3.8461 0.4171 sec/batch
Epoch 9/10 Iteration: 37600 Avg. Training loss: 3.8435 0.4198 sec/batch
Epoch 9/10 Iteration: 37700 Avg. Training loss: 3.7929 0.4224 sec/batch
Epoch 9/10 Iteration: 37800 Avg. Training loss: 3.8605 0.4270 sec/batch
Epoch 9/10 Iteration: 37900 Avg. Training loss: 3.8945 0.4344 sec/batch
Epoch 9/10 Iteration: 38000 Avg. Training loss: 3.8773 0.4556 sec/batch
Nearest to all: single, in, of, baasha, a, are, be, and,
Nearest to six: five, four, one, seven, zero, three, two, eight,
Nearest to their: to, they, these, however, breaking, and, allow, were,
Nearest to also: the, some, in, of, other, to, and, by,
Nearest to into: from, the, batavia, which, mesogens, and, continents, visualized,
Nearest to some: also, have, be, are, for, as, to, the,
Nearest to called: as, or, the, now, known, name, of, messiah,
Nearest to in: the, of, to, and, as, which, also, a,
Nearest to consists: schists, thirds, unicameral, unleavened, shops, or, manna, culminates,
Nearest to numerous: slacks, dwellers, common, responsiveness, coelacanth, fantastical, prairie, branding,
Nearest to know: we, you, ask, breslau, make, how, thank, fal,
Nearest to versions: wise, rulebooks, camcorder, documented, vga, cdpd, disturbed, operating,
Nearest to shows: television, hapgood, episode, show, transferable, margrethe, gala, vlaanderen,
Nearest to orthodox: matsui, oriental, moravians, apostolicae, thema, apostolic, numbering, eastern,
Nearest to scale: nb, forges, transposed, declination, azar, mato, bravery, greenwich,
Nearest to joseph: nikita, occupied, bangor, leases, moyen, psalter, diversions, schell,
Epoch 9/10 Iteration: 38100 Avg. Training loss: 3.8689 0.4831 sec/batch
Epoch 9/10 Iteration: 38200 Avg. Training loss: 3.7891 0.4596 sec/batch
Epoch 9/10 Iteration: 38300 Avg. Training loss: 3.8139 0.4639 sec/batch
Epoch 9/10 Iteration: 38400 Avg. Training loss: 3.7645 0.4573 sec/batch
Epoch 9/10 Iteration: 38500 Avg. Training loss: 3.8451 0.4543 sec/batch
Epoch 9/10 Iteration: 38600 Avg. Training loss: 3.8508 0.4539 sec/batch
Epoch 9/10 Iteration: 38700 Avg. Training loss: 3.8453 0.4557 sec/batch
Epoch 9/10 Iteration: 38800 Avg. Training loss: 3.8405 0.4513 sec/batch
Epoch 9/10 Iteration: 38900 Avg. Training loss: 3.8771 0.4522 sec/batch
Epoch 9/10 Iteration: 39000 Avg. Training loss: 3.9052 0.4577 sec/batch
Nearest to all: of, in, are, a, be, and, the, single,
Nearest to six: five, four, one, seven, zero, three, eight, two,
Nearest to their: to, they, these, however, breaking, time, were, allow,
Nearest to also: the, some, of, in, to, other, and, are,
Nearest to into: from, the, batavia, mesogens, and, which, visualized, prometheus,
Nearest to some: also, have, be, are, to, for, as, the,
Nearest to called: as, the, or, now, by, also, of, name,
Nearest to in: the, of, to, also, as, a, which, and,
Nearest to consists: or, schists, thirds, unicameral, unleavened, shops, rho, culminates,
Nearest to numerous: slacks, dwellers, common, coelacanth, responsiveness, panthers, prairie, fantastical,
Nearest to know: we, you, ask, how, make, breslau, thy, posthuman,
Nearest to versions: wise, rulebooks, camcorder, documented, disturbed, cdpd, backlight, vga,
Nearest to shows: television, hapgood, show, episode, transferable, margrethe, vlaanderen, encompasses,
Nearest to orthodox: matsui, oriental, apostolicae, thema, moravians, numbering, apostolic, transept,
Nearest to scale: nb, forges, transposed, azar, declination, mato, mainframes, extortion,
Nearest to joseph: nikita, leases, bangor, schell, diversions, occupied, moyen, psalter,
Epoch 9/10 Iteration: 39100 Avg. Training loss: 3.8343 0.4561 sec/batch
Epoch 9/10 Iteration: 39200 Avg. Training loss: 3.8403 0.4544 sec/batch
Epoch 9/10 Iteration: 39300 Avg. Training loss: 3.8664 0.4574 sec/batch
Epoch 9/10 Iteration: 39400 Avg. Training loss: 3.8541 0.4614 sec/batch
Epoch 9/10 Iteration: 39500 Avg. Training loss: 3.8591 0.4583 sec/batch
Epoch 9/10 Iteration: 39600 Avg. Training loss: 3.8770 0.4607 sec/batch
Epoch 9/10 Iteration: 39700 Avg. Training loss: 3.8536 0.4616 sec/batch
Epoch 9/10 Iteration: 39800 Avg. Training loss: 3.8219 0.4591 sec/batch
Epoch 9/10 Iteration: 39900 Avg. Training loss: 3.8723 0.4603 sec/batch
Epoch 9/10 Iteration: 40000 Avg. Training loss: 3.9082 0.4598 sec/batch
Nearest to all: of, and, are, single, be, in, a, the,
Nearest to six: five, four, one, zero, seven, two, three, eight,
Nearest to their: to, these, they, and, were, however, breaking, allow,
Nearest to also: the, some, in, of, to, and, other, by,
Nearest to into: from, the, batavia, and, mesogens, visualized, which, prometheus,
Nearest to some: also, have, are, be, to, as, the, such,
Nearest to called: as, or, the, of, name, now, known, messiah,
Nearest to in: the, of, to, also, as, and, a, which,
Nearest to consists: thirds, schists, or, shops, manna, unleavened, unicameral, culminates,
Nearest to numerous: common, responsiveness, dwellers, fantastical, slacks, coelacanth, oriya, panthers,
Nearest to know: we, you, ask, how, breslau, make, thy, posthuman,
Nearest to versions: wise, rulebooks, camcorder, documented, cdpd, backlight, vga, operating,
Nearest to shows: television, show, transferable, hapgood, margrethe, episode, nou, vlaanderen,
Nearest to orthodox: matsui, oriental, apostolicae, thema, moravians, numbering, transept, apostolic,
Nearest to scale: nb, forges, transposed, declination, azar, mato, apparent, greenwich,
Nearest to joseph: nikita, leases, bangor, schell, occupied, diversions, psalter, operated,
Epoch 9/10 Iteration: 40100 Avg. Training loss: 3.8535 0.4639 sec/batch
Epoch 9/10 Iteration: 40200 Avg. Training loss: 3.8428 0.4570 sec/batch
Epoch 9/10 Iteration: 40300 Avg. Training loss: 3.8467 0.4566 sec/batch
Epoch 9/10 Iteration: 40400 Avg. Training loss: 3.8503 0.4520 sec/batch
Epoch 9/10 Iteration: 40500 Avg. Training loss: 3.9185 0.4583 sec/batch
Epoch 9/10 Iteration: 40600 Avg. Training loss: 3.8142 0.4628 sec/batch
Epoch 9/10 Iteration: 40700 Avg. Training loss: 3.9331 0.4620 sec/batch
Epoch 9/10 Iteration: 40800 Avg. Training loss: 3.9236 0.4547 sec/batch
Epoch 9/10 Iteration: 40900 Avg. Training loss: 3.8890 0.4544 sec/batch
Epoch 9/10 Iteration: 41000 Avg. Training loss: 3.8130 0.4550 sec/batch
Nearest to all: of, single, and, a, are, be, has, the,
Nearest to six: five, four, seven, one, zero, two, three, eight,
Nearest to their: these, they, to, and, were, allow, jamal, however,
Nearest to also: the, some, of, in, to, and, by, other,
Nearest to into: from, the, batavia, and, visualized, which, mesogens, quickly,
Nearest to some: also, have, are, be, as, the, such, to,
Nearest to called: as, or, the, of, by, also, name, now,
Nearest to in: the, of, as, to, a, also, and, which,
Nearest to consists: or, thirds, unicameral, schists, manna, unleavened, culminates, clavichord,
Nearest to numerous: dwellers, common, slacks, responsiveness, fantastical, beren, panthers, coelacanth,
Nearest to know: we, you, ask, how, make, breslau, thy, i,
Nearest to versions: wise, rulebooks, camcorder, jaya, documented, disturbed, cdpd, vga,
Nearest to shows: television, margrethe, transferable, show, hapgood, episode, nou, arrival,
Nearest to orthodox: matsui, apostolicae, thema, oriental, moravians, numbering, transept, tractates,
Nearest to scale: nb, forges, transposed, declination, azar, mainframes, bravery, extortion,
Nearest to joseph: nikita, occupied, leases, bangor, schell, psalter, moyen, operated,
Epoch 9/10 Iteration: 41100 Avg. Training loss: 3.8360 0.4605 sec/batch
Epoch 9/10 Iteration: 41200 Avg. Training loss: 3.8426 0.4531 sec/batch
Epoch 9/10 Iteration: 41300 Avg. Training loss: 3.8197 0.4569 sec/batch
Epoch 9/10 Iteration: 41400 Avg. Training loss: 3.8747 0.4532 sec/batch
Epoch 9/10 Iteration: 41500 Avg. Training loss: 3.8428 0.4568 sec/batch
Epoch 9/10 Iteration: 41600 Avg. Training loss: 3.8824 0.4558 sec/batch
Epoch 10/10 Iteration: 41700 Avg. Training loss: 3.8591 0.3539 sec/batch
Epoch 10/10 Iteration: 41800 Avg. Training loss: 3.8547 0.4569 sec/batch
Epoch 10/10 Iteration: 41900 Avg. Training loss: 3.8232 0.4416 sec/batch
Epoch 10/10 Iteration: 42000 Avg. Training loss: 3.8735 0.4311 sec/batch
Nearest to all: and, single, of, are, in, the, a, has,
Nearest to six: five, one, four, seven, zero, eight, three, two,
Nearest to their: these, to, they, and, were, allow, jamal, however,
Nearest to also: the, some, in, of, to, and, other, as,
Nearest to into: from, the, batavia, and, which, mesogens, continents, quickly,
Nearest to some: also, have, be, are, as, for, to, the,
Nearest to called: as, or, the, name, of, by, also, known,
Nearest to in: the, of, to, and, as, a, is, also,
Nearest to consists: unicameral, or, culminates, unleavened, thirds, schists, clavichord, manna,
Nearest to numerous: dwellers, common, slacks, coelacanth, prairie, panthers, beren, responsiveness,
Nearest to know: you, we, ask, how, breslau, make, thy, thank,
Nearest to versions: wise, rulebooks, camcorder, documented, jaya, monosodium, cdpd, disturbed,
Nearest to shows: television, transferable, hapgood, show, margrethe, episode, gala, arrival,
Nearest to orthodox: matsui, oriental, apostolicae, moravians, thema, numbering, tractates, mcpherson,
Nearest to scale: nb, transposed, forges, declination, azar, mato, innumerable, forearm,
Nearest to joseph: nikita, occupied, bangor, leases, schell, diversions, operated, moyen,
Epoch 10/10 Iteration: 42100 Avg. Training loss: 3.7930 0.4357 sec/batch
Epoch 10/10 Iteration: 42200 Avg. Training loss: 3.8296 0.4282 sec/batch
Epoch 10/10 Iteration: 42300 Avg. Training loss: 3.7825 0.4287 sec/batch
Epoch 10/10 Iteration: 42400 Avg. Training loss: 3.8581 0.4283 sec/batch
Epoch 10/10 Iteration: 42500 Avg. Training loss: 3.8317 0.4295 sec/batch
Epoch 10/10 Iteration: 42600 Avg. Training loss: 3.8562 0.4308 sec/batch
Epoch 10/10 Iteration: 42700 Avg. Training loss: 3.8754 0.4315 sec/batch
Epoch 10/10 Iteration: 42800 Avg. Training loss: 3.7444 0.4302 sec/batch
Epoch 10/10 Iteration: 42900 Avg. Training loss: 3.8213 0.4333 sec/batch
Epoch 10/10 Iteration: 43000 Avg. Training loss: 3.7880 0.4310 sec/batch
Nearest to all: of, are, a, the, and, be, has, in,
Nearest to six: five, one, four, seven, zero, eight, two, three,
Nearest to their: these, they, however, allow, and, to, differentia, breaking,
Nearest to also: the, some, of, in, to, and, other, are,
Nearest to into: from, the, and, batavia, with, which, mesogens, quickly,
Nearest to some: also, have, be, are, as, to, the, such,
Nearest to called: as, or, the, of, also, by, known, now,
Nearest to in: the, of, a, to, and, as, is, also,
Nearest to consists: or, schists, manna, unleavened, unicameral, and, thirds, clavichord,
Nearest to numerous: dwellers, common, slacks, coelacanth, oriya, laws, panthers, study,
Nearest to know: we, you, ask, how, make, breslau, thy, fal,
Nearest to versions: wise, rulebooks, camcorder, cdpd, documented, jaya, disturbed, monosodium,
Nearest to shows: television, transferable, episode, margrethe, hapgood, show, arrival, gala,
Nearest to orthodox: matsui, oriental, apostolicae, thema, moravians, numbering, tags, zariski,
Nearest to scale: nb, forges, declination, transposed, innumerable, azar, proterozoic, withstand,
Nearest to joseph: nikita, bangor, leases, occupied, operated, diversions, latter, schell,
Epoch 10/10 Iteration: 43100 Avg. Training loss: 3.7551 0.4381 sec/batch
Epoch 10/10 Iteration: 43200 Avg. Training loss: 3.8094 0.4304 sec/batch
Epoch 10/10 Iteration: 43300 Avg. Training loss: 3.8502 0.4324 sec/batch
Epoch 10/10 Iteration: 43400 Avg. Training loss: 3.8330 0.4305 sec/batch
Epoch 10/10 Iteration: 43500 Avg. Training loss: 3.8256 0.4294 sec/batch
Epoch 10/10 Iteration: 43600 Avg. Training loss: 3.9160 0.4300 sec/batch
Epoch 10/10 Iteration: 43700 Avg. Training loss: 3.8046 0.4312 sec/batch
Epoch 10/10 Iteration: 43800 Avg. Training loss: 3.8368 0.4296 sec/batch
Epoch 10/10 Iteration: 43900 Avg. Training loss: 3.8532 0.4296 sec/batch
Epoch 10/10 Iteration: 44000 Avg. Training loss: 3.8340 0.4311 sec/batch
Nearest to all: of, and, are, a, be, the, single, in,
Nearest to six: five, one, four, seven, zero, eight, two, three,
Nearest to their: these, they, to, and, however, allow, breaking, were,
Nearest to also: the, some, of, in, to, and, other, as,
Nearest to into: from, the, mesogens, and, batavia, which, with, visualized,
Nearest to some: also, have, be, as, are, to, the, for,
Nearest to called: as, or, the, of, by, known, name, also,
Nearest to in: the, of, to, a, and, as, also, which,
Nearest to consists: or, unleavened, unicameral, manna, and, schists, thirds, clavichord,
Nearest to numerous: common, dwellers, slacks, fantastical, laws, responsiveness, panthers, oriya,
Nearest to know: you, we, ask, how, breslau, make, thy, posthuman,
Nearest to versions: wise, rulebooks, camcorder, monosodium, documented, jaya, disturbed, backlight,
Nearest to shows: television, episode, show, hapgood, transferable, lcds, margrethe, gala,
Nearest to orthodox: matsui, oriental, apostolicae, thema, moravians, tags, transept, numbering,
Nearest to scale: nb, transposed, forges, azar, declination, proterozoic, extortion, innumerable,
Nearest to joseph: nikita, bangor, occupied, diversions, leases, operated, schell, psalter,
Epoch 10/10 Iteration: 44100 Avg. Training loss: 3.8603 0.4342 sec/batch
Epoch 10/10 Iteration: 44200 Avg. Training loss: 3.8521 0.4283 sec/batch
Epoch 10/10 Iteration: 44300 Avg. Training loss: 3.8386 0.4290 sec/batch
Epoch 10/10 Iteration: 44400 Avg. Training loss: 3.8214 0.4266 sec/batch
Epoch 10/10 Iteration: 44500 Avg. Training loss: 3.8795 0.4275 sec/batch
Epoch 10/10 Iteration: 44600 Avg. Training loss: 3.8319 0.4282 sec/batch
Epoch 10/10 Iteration: 44700 Avg. Training loss: 3.8542 0.4363 sec/batch
Epoch 10/10 Iteration: 44800 Avg. Training loss: 3.8354 0.4393 sec/batch
Epoch 10/10 Iteration: 44900 Avg. Training loss: 3.8167 0.4394 sec/batch
Epoch 10/10 Iteration: 45000 Avg. Training loss: 3.8325 0.4409 sec/batch
Nearest to all: of, are, the, a, and, single, be, in,
Nearest to six: four, five, one, seven, zero, three, two, eight,
Nearest to their: these, they, and, to, allow, however, the, many,
Nearest to also: the, in, of, some, to, and, other, by,
Nearest to into: from, the, mesogens, and, batavia, with, which, continents,
Nearest to some: also, have, as, be, are, to, the, this,
Nearest to called: as, or, the, by, of, known, name, also,
Nearest to in: the, of, to, as, and, a, also, which,
Nearest to consists: or, schists, unicameral, manna, unleavened, thirds, by, clavichord,
Nearest to numerous: common, laws, slacks, oriya, dwellers, responsiveness, usually, fantastical,
Nearest to know: you, we, ask, how, make, breslau, thy, i,
Nearest to versions: wise, rulebooks, documented, camcorder, jaya, backlight, patanjali, operating,
Nearest to shows: television, show, margrethe, episode, transferable, arrival, nou, gala,
Nearest to orthodox: matsui, oriental, apostolicae, thema, moravians, transept, muscles, numbering,
Nearest to scale: nb, transposed, forges, azar, declination, extortion, withstand, innumerable,
Nearest to joseph: nikita, occupied, bangor, diversions, psalter, leases, operated, schell,
Epoch 10/10 Iteration: 45100 Avg. Training loss: 3.8715 0.4485 sec/batch
Epoch 10/10 Iteration: 45200 Avg. Training loss: 3.8006 0.4411 sec/batch
Epoch 10/10 Iteration: 45300 Avg. Training loss: 3.9403 0.4379 sec/batch
Epoch 10/10 Iteration: 45400 Avg. Training loss: 3.8559 0.4396 sec/batch
Epoch 10/10 Iteration: 45500 Avg. Training loss: 3.8953 0.4411 sec/batch
Epoch 10/10 Iteration: 45600 Avg. Training loss: 3.8248 0.4431 sec/batch
Epoch 10/10 Iteration: 45700 Avg. Training loss: 3.7687 0.4434 sec/batch
Epoch 10/10 Iteration: 45800 Avg. Training loss: 3.8744 0.4378 sec/batch
Epoch 10/10 Iteration: 45900 Avg. Training loss: 3.7739 0.4422 sec/batch
Epoch 10/10 Iteration: 46000 Avg. Training loss: 3.8561 0.4502 sec/batch
Nearest to all: a, of, and, the, are, single, be, in,
Nearest to six: four, one, seven, five, zero, eight, three, two,
Nearest to their: these, they, and, allow, to, were, breaking, however,
Nearest to also: the, in, some, of, and, to, other, by,
Nearest to into: from, the, and, which, mesogens, batavia, continents, with,
Nearest to some: also, have, as, are, be, such, this, to,
Nearest to called: as, or, the, by, of, also, name, known,
Nearest to in: the, of, as, to, and, also, a, which,
Nearest to consists: unicameral, or, thirds, manna, unleavened, schists, rajasthan, by,
Nearest to numerous: slacks, panthers, common, dwellers, oriya, responsiveness, fantastical, baritone,
Nearest to know: you, we, ask, how, breslau, make, i, thy,
Nearest to versions: wise, rulebooks, jaya, documented, camcorder, vga, greenlandic, patanjali,
Nearest to shows: television, show, margrethe, transferable, episode, gala, hapgood, arrival,
Nearest to orthodox: matsui, thema, oriental, apostolicae, moravians, numbering, ignores, muscles,
Nearest to scale: nb, transposed, forges, azar, withstand, extortion, innumerable, proterozoic,
Nearest to joseph: nikita, bangor, occupied, operated, leases, winnings, diversions, latter,
Epoch 10/10 Iteration: 46100 Avg. Training loss: 3.8352 0.4480 sec/batch
Epoch 10/10 Iteration: 46200 Avg. Training loss: 3.8313 0.4366 sec/batch

Restore the trained network if you need to:



In [ ]:

    
with train_graph.as_default():
    saver = tf.train.Saver()

with tf.Session(graph=train_graph) as sess:
    saver.restore(sess, tf.train.latest_checkpoint('checkpoints'))
    embed_mat = sess.run(embedding)

Visualizing the word vectors

Below we'll use T-SNE to visualize how our high-dimensional word vectors cluster together. T-SNE is used to project these vectors into two dimensions while preserving local stucture. Check out this post from Christopher Olah to learn more about T-SNE and other ways to visualize high-dimensional data.



In [17]:

    
%matplotlib inline
%config InlineBackend.figure_format = 'retina'

import matplotlib.pyplot as plt
from sklearn.manifold import TSNE









    



/home/fernanda/.miniconda/envs/tensorflow/lib/python3.5/site-packages/matplotlib/font_manager.py:280: UserWarning: Matplotlib is building the font cache using fc-list. This may take a moment.
  'Matplotlib is building the font cache using fc-list. '



In [18]:

    
viz_words = 500
tsne = TSNE()
embed_tsne = tsne.fit_transform(embed_mat[:viz_words, :])



In [19]:

    
fig, ax = plt.subplots(figsize=(14, 14))
for idx in range(viz_words):
    plt.scatter(*embed_tsne[idx, :], color='steelblue')
    plt.annotate(int_to_vocab[idx], (embed_tsne[idx, 0], embed_tsne[idx, 1]), alpha=0.7)



In [ ]: