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 translations.

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 language and words, 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 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()

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 [5]:

    
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 [6]:

    
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 [7]:

    
def get_target(words, idx, window_size=5):
    ''' Get a list of words in a window around an index. '''
    
    c = random.randint(0, window_size)
    start = max(0, idx - c)
    end = min(len(words), idx + c + 1)
    return words[start:end]

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 [8]:

    
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 one-hot encoded vectors. 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. This weight matrix is usually called the embedding matrix or embedding look-up table. 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 [9]:

    
train_graph = tf.Graph()
with train_graph.as_default():
    inputs = tf.placeholder(tf.int32, shape=[None], name = 'inputs')
    labels = tf.placeholder(tf.float32, shape=[None, None], name = 'labels')

Embedding

The embedding matrix has a size of the number of words by the number of neurons 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 one-hot encoded vectors for our inputs. When you do the matrix multiplication of the one-hot vector with the embedding matrix, you end up selecting only one row out of the entire matrix:

You don't actually need to do the matrix multiplication, you just need to select the row in the embedding matrix that corresponds to the input word. Then, the embedding matrix becomes a lookup table, you're looking up a vector the size of the hidden layer that represents the input word.

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. This TensorFlow tutorial will help if you get stuck.



In [10]:

    
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) # use tf.nn.embedding_lookup to get the hidden layer output

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 [11]:

    
# 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 [12]:

    
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 [13]:

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









    



mkdir: checkpoints: File exists

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 [ ]:

    
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: 5.5774 0.5127 sec/batch
Epoch 1/10 Iteration: 200 Avg. Training loss: 5.4496 0.4183 sec/batch
Epoch 1/10 Iteration: 300 Avg. Training loss: 5.2990 0.4119 sec/batch
Epoch 1/10 Iteration: 400 Avg. Training loss: 5.3338 0.4571 sec/batch
Epoch 1/10 Iteration: 500 Avg. Training loss: 5.1963 0.5887 sec/batch
Epoch 1/10 Iteration: 600 Avg. Training loss: 5.1839 0.6820 sec/batch
Epoch 1/10 Iteration: 700 Avg. Training loss: 5.2045 0.6096 sec/batch
Epoch 1/10 Iteration: 800 Avg. Training loss: 5.1018 0.5181 sec/batch
Epoch 1/10 Iteration: 900 Avg. Training loss: 5.0765 0.5203 sec/batch
Epoch 1/10 Iteration: 1000 Avg. Training loss: 5.0332 0.5306 sec/batch
Nearest to four: indies, provisional, fes, valentino, bachman, fermented, ins, sung,
Nearest to so: gela, complained, dusseldorf, airframe, manifests, hitomaro, islamic, sake,
Nearest to d: eisenach, clocks, smashing, reptiles, relies, rem, whipping, strain,
Nearest to they: christian, seasonally, variability, alongside, dedicating, synth, tomino, inflicted,
Nearest to been: pencils, peltier, kircher, declension, squaw, charisma, mater, oviedo,
Nearest to with: diable, consequence, rica, quintessentially, locating, coburg, likeable, syndicates,
Nearest to i: denies, sharada, migration, rhymes, awakened, gypsys, mannerism, rescind,
Nearest to have: typhoons, bridgman, yarrow, rossby, humane, opinions, alpes, shortness,
Nearest to bible: ve, pencils, barbarous, fgth, maskable, personnel, homogenization, drip,
Nearest to troops: udf, celtic, ebcdic, ethidium, jamal, parry, acceptable, samoan,
Nearest to articles: disconnected, intricacies, suicidal, recordable, disrupted, hauled, coppersmith, erzherzog,
Nearest to additional: sites, raisin, sole, blackbody, kfreebsd, geniuses, caseless, amnesiac,
Nearest to shown: territories, poulenc, nicholson, machir, odr, disclosure, armando, eulogies,
Nearest to heavy: vangelis, eckernf, wiesel, pipelining, amplify, bce, assessment, verg,
Nearest to professional: marcia, grimm, artefacts, caster, superconducting, torpedoes, ambrose, rivi,
Nearest to recorded: gorky, balliol, scoffed, danorum, intermodal, chaplin, privates, hairy,
Epoch 1/10 Iteration: 1100 Avg. Training loss: 5.0614 0.5066 sec/batch
Epoch 1/10 Iteration: 1200 Avg. Training loss: 4.9964 0.5090 sec/batch
Epoch 1/10 Iteration: 1300 Avg. Training loss: 4.9882 0.5364 sec/batch
Epoch 1/10 Iteration: 1400 Avg. Training loss: 4.8669 0.5034 sec/batch
Epoch 1/10 Iteration: 1500 Avg. Training loss: 4.8459 0.5193 sec/batch
Epoch 1/10 Iteration: 1600 Avg. Training loss: 4.8342 0.5102 sec/batch
Epoch 1/10 Iteration: 1700 Avg. Training loss: 4.7752 0.5056 sec/batch
Epoch 1/10 Iteration: 1800 Avg. Training loss: 4.7436 0.4992 sec/batch
Epoch 1/10 Iteration: 1900 Avg. Training loss: 4.6906 0.6007 sec/batch
Epoch 1/10 Iteration: 2000 Avg. Training loss: 4.6838 0.4390 sec/batch
Nearest to four: indies, provisional, one, fermented, valentino, fes, sung, play,
Nearest to so: complained, gela, islamic, dusseldorf, sake, hitomaro, manifests, airframe,
Nearest to d: eisenach, clocks, whipping, loser, organising, reptiles, rca, enthusiasts,
Nearest to they: christian, alongside, involved, variability, dedicating, seasonally, forgery, ideology,
Nearest to been: pencils, kircher, squaw, peltier, declension, guilty, mater, inevitable,
Nearest to with: consequence, ever, diable, prescription, rica, indeed, exploration, quintessentially,
Nearest to i: migration, this, denies, moisture, gypsys, if, gluon, rescind,
Nearest to have: typhoons, opinions, eliminating, yarrow, noticed, averages, existence, industry,
Nearest to bible: pencils, ve, names, them, maskable, homogenization, barbarous, fgth,
Nearest to troops: udf, celtic, ebcdic, acceptable, jamal, making, accepting, ethidium,
Nearest to articles: suicidal, recordable, disconnected, disrupted, intricacies, hauled, blackfoot, winning,
Nearest to additional: sole, sites, field, raisin, amnesiac, botanist, noin, auto,
Nearest to shown: territories, disclosure, abstract, poulenc, nicholson, spectrum, flogging, eulogies,
Nearest to heavy: vangelis, pipelining, besides, assessment, eckernf, bce, minimum, schists,
Nearest to professional: marcia, appian, artefacts, grimm, torpedoes, caster, superconducting, ambrose,
Nearest to recorded: balliol, gorky, for, particularly, harsh, chaplin, hairy, scoffed,
Epoch 1/10 Iteration: 2100 Avg. Training loss: 4.6245 0.4443 sec/batch
Epoch 1/10 Iteration: 2200 Avg. Training loss: 4.6082 0.4387 sec/batch
Epoch 1/10 Iteration: 2300 Avg. Training loss: 4.5602 0.4370 sec/batch
Epoch 1/10 Iteration: 2400 Avg. Training loss: 4.5503 0.4398 sec/batch
Epoch 1/10 Iteration: 2500 Avg. Training loss: 4.4649 0.4429 sec/batch
Epoch 1/10 Iteration: 2600 Avg. Training loss: 4.5141 0.4492 sec/batch
Epoch 1/10 Iteration: 2700 Avg. Training loss: 4.4810 0.4525 sec/batch
Epoch 1/10 Iteration: 2800 Avg. Training loss: 4.4484 0.5195 sec/batch
Epoch 1/10 Iteration: 2900 Avg. Training loss: 4.4443 0.4389 sec/batch
Epoch 1/10 Iteration: 3000 Avg. Training loss: 4.4499 0.4393 sec/batch
Nearest to four: indies, sung, one, provisional, valentino, surveying, median, fes,
Nearest to so: complained, gela, manifests, airframe, dusseldorf, islamic, hitomaro, sake,
Nearest to d: eisenach, rca, clocks, relies, loser, whipping, opposing, enthusiasts,
Nearest to they: christian, alongside, plausible, ideology, seasonally, dedicating, variability, conscious,
Nearest to been: pencils, peltier, inevitable, guilty, declension, kircher, squaw, charisma,
Nearest to with: consequence, diable, exploration, preventive, executive, capabilities, screen, rica,
Nearest to i: denies, migration, if, gypsys, this, moisture, evaluation, rhymes,
Nearest to have: typhoons, opinions, eliminating, averages, humane, affiliated, bridgman, shortness,
Nearest to bible: ve, names, pencils, them, personnel, translations, homogenization, lot,
Nearest to troops: udf, celtic, acceptable, accepting, ebcdic, jamal, making, parry,
Nearest to articles: suicidal, disrupted, recordable, hauled, winning, disconnected, intricacies, primal,
Nearest to additional: sole, sites, field, raisin, auto, botanist, blackbody, geniuses,
Nearest to shown: territories, disclosure, abstract, righteousness, spectrum, subpages, eulogies, flogging,
Nearest to heavy: vangelis, assessment, bce, pipelining, eckernf, minimum, besides, wiesel,
Nearest to professional: marcia, grimm, artefacts, appian, rivi, torpedoes, schilling, caster,
Nearest to recorded: balliol, gorky, particularly, chaplin, secretly, for, harmonious, hairy,
Epoch 1/10 Iteration: 3100 Avg. Training loss: 4.4548 0.5121 sec/batch
Epoch 1/10 Iteration: 3200 Avg. Training loss: 4.4218 0.6654 sec/batch
Epoch 1/10 Iteration: 3300 Avg. Training loss: 4.3872 0.5657 sec/batch
Epoch 1/10 Iteration: 3400 Avg. Training loss: 4.3815 0.5423 sec/batch
Epoch 1/10 Iteration: 3500 Avg. Training loss: 4.4455 0.4927 sec/batch
Epoch 1/10 Iteration: 3600 Avg. Training loss: 4.3890 0.6747 sec/batch
Epoch 1/10 Iteration: 3700 Avg. Training loss: 4.3862 0.5494 sec/batch
Epoch 1/10 Iteration: 3800 Avg. Training loss: 4.3832 0.4876 sec/batch
Epoch 1/10 Iteration: 3900 Avg. Training loss: 4.3446 0.5642 sec/batch
Epoch 1/10 Iteration: 4000 Avg. Training loss: 4.3206 0.7429 sec/batch
Nearest to four: indies, one, valentino, sung, provisional, viscount, median, surveying,
Nearest to so: gela, manifests, airframe, complained, dusseldorf, sake, cultic, hitomaro,
Nearest to d: eisenach, inventor, clocks, rca, luis, reptiles, relies, censor,
Nearest to they: christian, seasonally, alongside, plausible, ideology, variability, veritable, gates,
Nearest to been: pencils, peltier, inevitable, guilty, declension, kircher, charisma, squaw,
Nearest to with: consequence, exploration, preventive, diable, rica, accustomed, syndicates, screen,
Nearest to i: denies, if, rhymes, gypsys, gyula, jay, schism, gluon,
Nearest to have: opinions, humane, typhoons, eliminating, averages, shortness, yarrow, bridgman,
Nearest to bible: ve, pencils, translations, names, them, homogenization, personnel, familial,
Nearest to troops: udf, celtic, samoan, acceptable, vesper, ebcdic, accepting, parry,
Nearest to articles: recordable, disrupted, suicidal, disconnected, winning, hauled, intricacies, tley,
Nearest to additional: sites, sole, blackbody, field, scrambled, chastity, geniuses, raisin,
Nearest to shown: territories, disclosure, righteousness, abstract, subpages, sabotage, kropotkin, nicholson,
Nearest to heavy: vangelis, assessment, pipelining, bce, wiesel, eckernf, minimum, amplify,
Nearest to professional: marcia, artefacts, grimm, rivi, appian, schilling, ambrose, torpedoes,
Nearest to recorded: balliol, secretly, chaplin, gorky, particularly, harmonious, affiliate, branded,
Epoch 1/10 Iteration: 4100 Avg. Training loss: 4.3205 0.4995 sec/batch
Epoch 1/10 Iteration: 4200 Avg. Training loss: 4.3297 0.6686 sec/batch
Epoch 1/10 Iteration: 4300 Avg. Training loss: 4.2864 0.5902 sec/batch
Epoch 1/10 Iteration: 4400 Avg. Training loss: 4.3188 0.4948 sec/batch
Epoch 1/10 Iteration: 4500 Avg. Training loss: 4.3314 0.4944 sec/batch
Epoch 1/10 Iteration: 4600 Avg. Training loss: 4.3037 0.5539 sec/batch
Epoch 2/10 Iteration: 4700 Avg. Training loss: 4.2713 0.4516 sec/batch
Epoch 2/10 Iteration: 4800 Avg. Training loss: 4.2400 0.6103 sec/batch
Epoch 2/10 Iteration: 4900 Avg. Training loss: 4.2003 0.5331 sec/batch
Epoch 2/10 Iteration: 5000 Avg. Training loss: 4.1990 0.4932 sec/batch
Nearest to four: one, valentino, physicist, two, indies, viscount, amelia, surveying,
Nearest to so: manifests, airframe, gela, complained, dusseldorf, wellcome, clarity, sake,
Nearest to d: inventor, eisenach, musician, nobel, eight, luis, prize, jaime,
Nearest to they: christian, seasonally, plausible, alongside, veritable, utterly, brilliantly, ideology,
Nearest to been: peltier, pencils, inevitable, guilty, kircher, declension, saigon, charisma,
Nearest to with: exploration, consequence, preventive, wray, accustomed, syndicates, diable, bungee,
Nearest to i: denies, rhymes, gypsys, me, if, awakened, gyula, ludicrous,
Nearest to have: humane, opinions, typhoons, shortness, averages, eliminating, manichean, mooring,
Nearest to bible: translations, pencils, ve, names, homogenization, them, tiberian, barbarous,
Nearest to troops: udf, samoan, vesper, celtic, acceptable, accepting, lords, miraculously,
Nearest to articles: suicidal, recordable, disconnected, disrupted, tley, intricacies, coppersmith, atari,
Nearest to additional: sites, scrambled, chastity, field, caseless, raisin, sole, blackbody,
Nearest to shown: territories, disclosure, righteousness, abstract, kropotkin, subpages, spectrum, hakim,
Nearest to heavy: vangelis, assessment, pipelining, bce, wiesel, amplify, organise, minimum,
Nearest to professional: marcia, artefacts, grimm, rivi, schilling, reinhardt, torpedoes, appian,
Nearest to recorded: balliol, chaplin, secretly, gorky, affiliate, hairy, taciturn, lanchester,
Epoch 2/10 Iteration: 5100 Avg. Training loss: 4.1708 0.5046 sec/batch
Epoch 2/10 Iteration: 5200 Avg. Training loss: 4.1671 0.5380 sec/batch
Epoch 2/10 Iteration: 5300 Avg. Training loss: 4.1396 0.4830 sec/batch
Epoch 2/10 Iteration: 5400 Avg. Training loss: 4.1965 0.5038 sec/batch
Epoch 2/10 Iteration: 5500 Avg. Training loss: 4.1714 0.4793 sec/batch
Epoch 2/10 Iteration: 5600 Avg. Training loss: 4.1471 0.4788 sec/batch
Epoch 2/10 Iteration: 5700 Avg. Training loss: 4.1279 0.4922 sec/batch
Epoch 2/10 Iteration: 5800 Avg. Training loss: 4.0823 0.5527 sec/batch
Epoch 2/10 Iteration: 5900 Avg. Training loss: 4.1141 0.5005 sec/batch
Epoch 2/10 Iteration: 6000 Avg. Training loss: 4.0742 0.5038 sec/batch
Nearest to four: two, one, median, valentino, indies, amelia, physicist, daglow,
Nearest to so: gela, airframe, complained, manifests, sake, dusseldorf, wellcome, clarity,
Nearest to d: inventor, eisenach, jaime, nobel, rca, luis, musician, mit,
Nearest to they: christian, seasonally, alongside, plausible, utterly, brilliantly, veritable, ideology,
Nearest to been: peltier, inevitable, pencils, guilty, kircher, declension, saigon, enter,
Nearest to with: wray, exploration, preventive, botulism, amplifying, capabilities, alleviated, consequence,
Nearest to i: if, denies, rhymes, awakened, me, sharada, gypsys, ludicrous,
Nearest to have: humane, opinions, shortness, averages, eliminating, typhoons, manichean, mooring,
Nearest to bible: translations, pencils, names, ve, tiberian, homogenization, torah, theology,
Nearest to troops: udf, samoan, vesper, army, celtic, acceptable, accepting, lords,
Nearest to articles: suicidal, disconnected, recordable, disrupted, tley, primal, intricacies, coppersmith,
Nearest to additional: sites, scrambled, caseless, blackbody, amour, raisin, field, domain,
Nearest to shown: disclosure, righteousness, territories, abstract, subpages, kropotkin, hakim, alleviation,
Nearest to heavy: vangelis, pipelining, assessment, organise, amplify, wiesel, bagger, minimum,
Nearest to professional: marcia, schilling, grimm, artefacts, rivi, marl, petersen, reinhardt,
Nearest to recorded: balliol, chaplin, affiliate, gorky, secretly, hairy, stanhope, collingridge,
Epoch 2/10 Iteration: 6100 Avg. Training loss: 4.0998 0.5227 sec/batch
Epoch 2/10 Iteration: 6200 Avg. Training loss: 4.0678 0.5535 sec/batch
Epoch 2/10 Iteration: 6300 Avg. Training loss: 4.0867 0.4929 sec/batch
Epoch 2/10 Iteration: 6400 Avg. Training loss: 4.0567 0.4881 sec/batch
Epoch 2/10 Iteration: 6500 Avg. Training loss: 4.0732 0.4947 sec/batch
Epoch 2/10 Iteration: 6600 Avg. Training loss: 4.1007 0.5269 sec/batch
Epoch 2/10 Iteration: 6700 Avg. Training loss: 4.0522 0.4878 sec/batch
Epoch 2/10 Iteration: 6800 Avg. Training loss: 4.0554 0.4906 sec/batch
Epoch 2/10 Iteration: 6900 Avg. Training loss: 4.0712 0.6142 sec/batch
Epoch 2/10 Iteration: 7000 Avg. Training loss: 4.0371 0.4919 sec/batch
Nearest to four: two, one, median, valentino, five, runways, six, amelia,
Nearest to so: airframe, gela, complained, manifests, wellcome, clarity, string, aramean,
Nearest to d: inventor, eisenach, rca, nobel, jaime, mit, rem, biochemist,
Nearest to they: christian, plausible, alongside, seasonally, utterly, etre, brilliantly, yup,
Nearest to been: peltier, inevitable, pencils, infiltrators, declension, barometer, highlighted, saigon,
Nearest to with: wray, exploration, accustomed, amplifying, alleviated, likeable, screen, botulism,
Nearest to i: if, me, denies, sharada, rhymes, awakened, gypsys, ludicrous,
Nearest to have: humane, opinions, shortness, eliminating, manichean, longhair, averages, mooring,
Nearest to bible: translations, names, tiberian, torah, theology, pencils, passages, jesus,
Nearest to troops: udf, samoan, army, vesper, acceptable, celtic, miraculously, ashburton,
Nearest to articles: suicidal, recordable, disconnected, primal, coppersmith, disrupted, tley, specification,
Nearest to additional: sites, scrambled, blackbody, cryptographically, caseless, field, kfreebsd, raisin,
Nearest to shown: disclosure, abstract, righteousness, spectrum, subpages, hakim, alleviation, colours,
Nearest to heavy: vangelis, pipelining, organise, amplify, minimum, assessment, rightmost, bagger,
Nearest to professional: marcia, schilling, grimm, rivi, artefacts, marl, petersen, debilitating,
Nearest to recorded: balliol, chaplin, hairy, stanhope, taciturn, collingridge, affiliate, secretly,
Epoch 2/10 Iteration: 7100 Avg. Training loss: 4.0270 0.5460 sec/batch
Epoch 2/10 Iteration: 7200 Avg. Training loss: 4.0727 0.4653 sec/batch
Epoch 2/10 Iteration: 7300 Avg. Training loss: 4.0333 0.4060 sec/batch
Epoch 2/10 Iteration: 7400 Avg. Training loss: 4.0201 0.4080 sec/batch
Epoch 2/10 Iteration: 7500 Avg. Training loss: 4.0683 0.4138 sec/batch
Epoch 2/10 Iteration: 7600 Avg. Training loss: 4.0390 0.4770 sec/batch
Epoch 2/10 Iteration: 7700 Avg. Training loss: 4.0428 0.4356 sec/batch
Epoch 2/10 Iteration: 7800 Avg. Training loss: 4.0435 0.4331 sec/batch
Epoch 2/10 Iteration: 7900 Avg. Training loss: 4.0042 0.4787 sec/batch
Epoch 2/10 Iteration: 8000 Avg. Training loss: 3.9987 0.7035 sec/batch
Nearest to four: median, two, one, six, five, runways, zero, keres,
Nearest to so: manifests, complained, airframe, gela, cultic, sake, normals, wellcome,
Nearest to d: inventor, eisenach, rca, biochemist, mit, nobel, rem, jaime,
Nearest to they: plausible, alongside, christian, utterly, seasonally, etre, conscious, ideology,
Nearest to been: peltier, inevitable, pencils, barometer, highlighted, brenda, declension, infiltrators,
Nearest to with: wray, exploration, accustomed, amplifying, botulism, semiautomatic, alleviated, likeable,
Nearest to i: if, me, sharada, awakened, denies, gypsys, rhymes, you,
Nearest to have: humane, manichean, pollinated, shortness, opinions, eliminating, averages, longhair,
Nearest to bible: translations, tiberian, torah, passages, jesus, spirit, pencils, ve,
Nearest to troops: army, samoan, udf, vesper, acceptable, ashburton, war, celtic,
Nearest to articles: recordable, suicidal, disconnected, disrupted, specification, primal, tley, coppersmith,
Nearest to additional: scrambled, caseless, sites, cryptographically, blackbody, kfreebsd, amour, domain,
Nearest to shown: righteousness, disclosure, subpages, abstract, hakim, alleviation, colours, spectrum,
Nearest to heavy: vangelis, pipelining, organise, amplify, minimum, assessment, rightmost, bagger,
Nearest to professional: marcia, schilling, grimm, rivi, artefacts, marl, petersen, amateur,
Nearest to recorded: chaplin, balliol, affiliate, collingridge, stanhope, taciturn, proselytes, hairy,
Epoch 2/10 Iteration: 8100 Avg. Training loss: 3.9854 0.6601 sec/batch
Epoch 2/10 Iteration: 8200 Avg. Training loss: 3.9900 0.6409 sec/batch
Epoch 2/10 Iteration: 8300 Avg. Training loss: 4.0479 0.6793 sec/batch
Epoch 2/10 Iteration: 8400 Avg. Training loss: 4.0427 0.6346 sec/batch
Epoch 2/10 Iteration: 8500 Avg. Training loss: 4.0400 0.6650 sec/batch
Epoch 2/10 Iteration: 8600 Avg. Training loss: 3.9646 0.6639 sec/batch
Epoch 2/10 Iteration: 8700 Avg. Training loss: 3.9495 0.6437 sec/batch
Epoch 2/10 Iteration: 8800 Avg. Training loss: 4.0140 0.6596 sec/batch
Epoch 2/10 Iteration: 8900 Avg. Training loss: 3.9180 0.4605 sec/batch
Epoch 2/10 Iteration: 9000 Avg. Training loss: 3.9681 0.4301 sec/batch
Nearest to four: one, two, runways, keres, six, five, zero, airports,
Nearest to so: manifests, airframe, gela, cultic, sake, complained, decayed, lances,
Nearest to d: inventor, biochemist, nobel, eisenach, prize, astronomer, jaime, physician,
Nearest to they: plausible, alongside, seasonally, christian, utterly, inflicted, variability, exceedingly,
Nearest to been: peltier, inevitable, pencils, highlighted, barometer, declension, infiltrators, driest,
Nearest to with: wray, amplifying, botulism, accustomed, exploration, semiautomatic, kebir, gran,
Nearest to i: if, me, awakened, rhymes, sharada, you, denies, myself,
Nearest to have: humane, manichean, pollinated, eliminating, mooring, averages, shortness, opinions,
Nearest to bible: translations, tiberian, torah, passages, spirit, theology, names, jesus,
Nearest to troops: army, samoan, udf, acceptable, vesper, ashburton, occupation, miraculously,
Nearest to articles: suicidal, recordable, disconnected, faq, midyear, coppersmith, tley, specification,
Nearest to additional: caseless, scrambled, blackbody, cryptographically, sites, superiority, kfreebsd, skipper,
Nearest to shown: righteousness, disclosure, subpages, colours, abstract, alleviation, spectrum, hakim,
Nearest to heavy: vangelis, organise, pipelining, firing, amplify, bagger, sunflower, metres,
Nearest to professional: marcia, schilling, rivi, grimm, artefacts, petersen, marl, amateur,
Nearest to recorded: collingridge, stanhope, chaplin, affiliate, balliol, hairy, proselytes, taciturn,
Epoch 2/10 Iteration: 9100 Avg. Training loss: 3.9781 0.4252 sec/batch
Epoch 2/10 Iteration: 9200 Avg. Training loss: 3.9512 0.4157 sec/batch
Epoch 3/10 Iteration: 9300 Avg. Training loss: 4.0112 0.1927 sec/batch
Epoch 3/10 Iteration: 9400 Avg. Training loss: 3.9330 0.4200 sec/batch
Epoch 3/10 Iteration: 9500 Avg. Training loss: 3.9058 0.4191 sec/batch
Epoch 3/10 Iteration: 9600 Avg. Training loss: 3.9126 0.5105 sec/batch
Epoch 3/10 Iteration: 9700 Avg. Training loss: 3.8855 0.4400 sec/batch
Epoch 3/10 Iteration: 9800 Avg. Training loss: 3.8666 0.4438 sec/batch
Epoch 3/10 Iteration: 9900 Avg. Training loss: 3.9107 0.4259 sec/batch
Epoch 3/10 Iteration: 10000 Avg. Training loss: 3.8332 0.4295 sec/batch
Nearest to four: keres, two, runways, one, median, five, six, pix,
Nearest to so: gela, manifests, complained, airframe, sake, pronominal, cultic, lances,
Nearest to d: inventor, biochemist, nobel, eisenach, jaime, johann, durkheim, engineer,
Nearest to they: alongside, christian, plausible, seasonally, inflicted, utterly, variability, vertebrate,
Nearest to been: peltier, inevitable, pencils, highlighted, brenda, barometer, informed, alleged,
Nearest to with: wray, botulism, accustomed, remedied, semiautomatic, amplifying, delhomme, exploration,
Nearest to i: if, me, awakened, rhymes, sharada, myself, crowned, ludicrous,
Nearest to have: manichean, humane, pollinated, opinions, mooring, averages, longhair, shortness,
Nearest to bible: translations, tiberian, torah, passages, spirit, tobit, protestantism, names,
Nearest to troops: army, samoan, udf, ashburton, acceptable, vesper, luftwaffe, miraculously,
Nearest to articles: suicidal, recordable, disconnected, midyear, faq, coppersmith, primal, disrupted,
Nearest to additional: caseless, scrambled, cryptographically, blackbody, sites, superiority, kfreebsd, clause,
Nearest to shown: righteousness, subpages, disclosure, colours, alleviation, hakim, wherein, odr,
Nearest to heavy: vangelis, organise, pipelining, firing, amplify, bagger, rightmost, wiesel,
Nearest to professional: marcia, schilling, amateur, rivi, marl, grimm, petersen, debilitating,
Nearest to recorded: collingridge, stanhope, chaplin, balliol, affiliate, hairy, proselytes, taciturn,
Epoch 3/10 Iteration: 10100 Avg. Training loss: 3.9248 0.4304 sec/batch
Epoch 3/10 Iteration: 10200 Avg. Training loss: 3.8680 0.4271 sec/batch
Epoch 3/10 Iteration: 10300 Avg. Training loss: 3.9026 0.5496 sec/batch
Epoch 3/10 Iteration: 10400 Avg. Training loss: 3.8291 0.6362 sec/batch
Epoch 3/10 Iteration: 10500 Avg. Training loss: 3.8580 0.6673 sec/batch
Epoch 3/10 Iteration: 10600 Avg. Training loss: 3.8262 0.6872 sec/batch
Epoch 3/10 Iteration: 10700 Avg. Training loss: 3.8132 0.6671 sec/batch
Epoch 3/10 Iteration: 10800 Avg. Training loss: 3.8250 0.6484 sec/batch
Epoch 3/10 Iteration: 10900 Avg. Training loss: 3.8635 0.6424 sec/batch
Epoch 3/10 Iteration: 11000 Avg. Training loss: 3.8393 0.6646 sec/batch
Nearest to four: keres, median, two, runways, five, pix, six, zero,
Nearest to so: airframe, gela, manifests, sake, lances, complained, pronominal, cultic,
Nearest to d: inventor, eisenach, durkheim, biochemist, rca, nobel, ure, debussy,
Nearest to they: plausible, nutrient, inflicted, variability, harvesting, alongside, vertebrate, seasonally,
Nearest to been: peltier, inevitable, brenda, barometer, highlighted, pencils, protozoa, informed,
Nearest to with: wray, botulism, accustomed, remedied, exploration, semiautomatic, amplifying, mems,
Nearest to i: if, me, awakened, rhymes, sharada, myself, ludicrous, denies,
Nearest to have: manichean, humane, mooring, shortness, pollinated, longhair, eliminating, opinions,
Nearest to bible: translations, tiberian, torah, passages, spirit, tobit, protestantism, names,
Nearest to troops: army, samoan, ashburton, udf, acceptable, luftwaffe, miraculously, intercept,
Nearest to articles: suicidal, disconnected, recordable, midyear, faq, primal, blog, coppersmith,
Nearest to additional: scrambled, caseless, cryptographically, sites, blackbody, kfreebsd, powermac, superiority,
Nearest to shown: righteousness, subpages, disclosure, colours, abstract, blurs, isomers, wherein,
Nearest to heavy: vangelis, organise, bagger, firing, wiped, pipelining, sunflower, amplify,
Nearest to professional: schilling, amateur, marcia, marl, debilitating, rivi, petersen, artefacts,
Nearest to recorded: collingridge, hairy, chaplin, stanhope, guided, balliol, proselytes, affiliate,
Epoch 3/10 Iteration: 11100 Avg. Training loss: 3.8345 0.6848 sec/batch
Epoch 3/10 Iteration: 11200 Avg. Training loss: 3.8744 0.5967 sec/batch
Epoch 3/10 Iteration: 11300 Avg. Training loss: 3.8391 0.4522 sec/batch
Epoch 3/10 Iteration: 11400 Avg. Training loss: 3.8124 0.5076 sec/batch
Epoch 3/10 Iteration: 11500 Avg. Training loss: 3.8474 0.4855 sec/batch
Epoch 3/10 Iteration: 11600 Avg. Training loss: 3.8458 0.4861 sec/batch
Epoch 3/10 Iteration: 11700 Avg. Training loss: 3.8672 0.4733 sec/batch
Epoch 3/10 Iteration: 11800 Avg. Training loss: 3.8128 0.4299 sec/batch
Epoch 3/10 Iteration: 11900 Avg. Training loss: 3.7828 0.4101 sec/batch
Epoch 3/10 Iteration: 12000 Avg. Training loss: 3.8409 0.4184 sec/batch
Nearest to four: keres, two, median, runways, five, one, six, zero,
Nearest to so: gela, airframe, manifests, sake, lances, cultic, complained, decayed,
Nearest to d: inventor, biochemist, eisenach, durkheim, nobel, ure, rca, jaime,
Nearest to they: plausible, alongside, inflicted, christian, harvesting, variability, etre, seasonally,
Nearest to been: peltier, highlighted, brenda, inevitable, barometer, oviedo, protozoa, driest,
Nearest to with: wray, botulism, exploration, accustomed, semiautomatic, remedied, alleviated, amplifying,
Nearest to i: me, if, awakened, rhymes, myself, sharada, crowned, you,
Nearest to have: manichean, humane, longhair, shortness, pollinated, mooring, eliminating, opinions,
Nearest to bible: translations, tiberian, torah, passages, tobit, spirit, hebrew, names,
Nearest to troops: army, samoan, ashburton, luftwaffe, acceptable, udf, intercept, miraculously,
Nearest to articles: suicidal, faq, blog, recordable, disconnected, midyear, specification, coppersmith,
Nearest to additional: scrambled, caseless, cryptographically, sites, straddle, skipper, attract, blackbody,
Nearest to shown: righteousness, subpages, disclosure, alleviation, tetrahedral, colours, blurs, abstract,
Nearest to heavy: vangelis, organise, bagger, wiped, pipelining, propel, firing, sunflower,
Nearest to professional: amateur, schilling, marcia, marl, debilitating, rivi, artefacts, grimm,
Nearest to recorded: collingridge, stanhope, chaplin, hairy, band, balliol, proselytes, lanchester,
Epoch 3/10 Iteration: 12100 Avg. Training loss: 3.8688 0.4151 sec/batch
Epoch 3/10 Iteration: 12200 Avg. Training loss: 3.8316 0.4157 sec/batch
Epoch 3/10 Iteration: 12300 Avg. Training loss: 3.8341 0.4182 sec/batch
Epoch 3/10 Iteration: 12400 Avg. Training loss: 3.8551 0.4105 sec/batch
Epoch 3/10 Iteration: 12500 Avg. Training loss: 3.8146 0.4229 sec/batch
Epoch 3/10 Iteration: 12600 Avg. Training loss: 3.7714 0.4246 sec/batch
Epoch 3/10 Iteration: 12700 Avg. Training loss: 3.8350 0.4400 sec/batch
Epoch 3/10 Iteration: 12800 Avg. Training loss: 3.7819 0.4347 sec/batch
Epoch 3/10 Iteration: 12900 Avg. Training loss: 3.8593 0.4450 sec/batch
Epoch 3/10 Iteration: 13000 Avg. Training loss: 3.8696 0.4308 sec/batch
Nearest to four: keres, two, five, one, six, runways, zero, median,
Nearest to so: cultic, gela, manifests, airframe, sake, lances, complained, pronominal,
Nearest to d: inventor, biochemist, eisenach, nobel, jaime, durkheim, alexandre, ure,
Nearest to they: alongside, inflicted, christian, plausible, harvesting, variability, etre, foe,
Nearest to been: peltier, brenda, highlighted, oviedo, inevitable, informed, barometer, alleged,
Nearest to with: wray, accustomed, semiautomatic, remedied, botulism, exploration, curtail, guillotined,
Nearest to i: me, if, myself, awakened, rhymes, crowned, sharada, you,
Nearest to have: manichean, humane, longhair, supernaturally, eliminating, gelderland, pollinated, shortness,
Nearest to bible: translations, tiberian, tobit, torah, passages, names, ve, spirit,
Nearest to troops: army, samoan, ashburton, luftwaffe, udf, acceptable, vandals, intercept,
Nearest to articles: suicidal, blog, faq, disconnected, recordable, midyear, newswire, coppersmith,
Nearest to additional: caseless, straddle, scrambled, cryptographically, sites, attract, skipper, superiority,
Nearest to shown: subpages, righteousness, disclosure, colours, alleviation, telford, tetrahedral, plums,
Nearest to heavy: vangelis, organise, bagger, wiped, propel, diskettes, cortisol, firing,
Nearest to professional: amateur, schilling, marcia, marl, graduation, negro, petersen, debilitating,
Nearest to recorded: collingridge, stanhope, proselytes, chaplin, band, song, guided, taciturn,
Epoch 3/10 Iteration: 13100 Avg. Training loss: 3.8951 0.4253 sec/batch
Epoch 3/10 Iteration: 13200 Avg. Training loss: 3.8158 0.4105 sec/batch
Epoch 3/10 Iteration: 13300 Avg. Training loss: 3.7934 0.4056 sec/batch
Epoch 3/10 Iteration: 13400 Avg. Training loss: 3.7902 0.4094 sec/batch
Epoch 3/10 Iteration: 13500 Avg. Training loss: 3.7018 0.4108 sec/batch
Epoch 3/10 Iteration: 13600 Avg. Training loss: 3.8113 0.4116 sec/batch
Epoch 3/10 Iteration: 13700 Avg. Training loss: 3.7906 1.9707 sec/batch
Epoch 3/10 Iteration: 13800 Avg. Training loss: 3.7724 0.6068 sec/batch
Epoch 4/10 Iteration: 13900 Avg. Training loss: 3.8121 0.0910 sec/batch
Epoch 4/10 Iteration: 14000 Avg. Training loss: 3.7799 0.4722 sec/batch
Nearest to four: keres, two, five, zero, one, six, runways, median,
Nearest to so: airframe, manifests, gela, cultic, sake, pronominal, lances, decayed,
Nearest to d: inventor, biochemist, eisenach, durkheim, jaime, nobel, isambard, debussy,
Nearest to they: alongside, plausible, harvesting, variability, inflicted, nutrient, seasonally, vertebrate,
Nearest to been: peltier, highlighted, brenda, protozoa, inevitable, barometer, oviedo, informed,
Nearest to with: wray, accustomed, semiautomatic, botulism, remedied, vitamin, curtail, amplifying,
Nearest to i: me, if, myself, awakened, rhymes, crowned, you, sharada,
Nearest to have: manichean, humane, longhair, supernaturally, inspires, pollinated, shortness, vlsi,
Nearest to bible: translations, tiberian, tobit, torah, psalms, names, passages, hebrew,
Nearest to troops: army, samoan, ashburton, luftwaffe, udf, acceptable, vandals, intercept,
Nearest to articles: suicidal, faq, disconnected, blog, recordable, midyear, federations, newswire,
Nearest to additional: caseless, cryptographically, straddle, scrambled, sites, skipper, attract, superiority,
Nearest to shown: subpages, disclosure, righteousness, tetrahedral, colours, wherein, alleviation, hakim,
Nearest to heavy: vangelis, organise, bagger, propel, wiped, cortisol, runeword, amplify,
Nearest to professional: amateur, schilling, marcia, marl, negro, conference, graduation, debilitating,
Nearest to recorded: collingridge, guided, band, stanhope, hairy, chaplin, lanchester, proselytes,
Epoch 4/10 Iteration: 14100 Avg. Training loss: 3.7317 0.4579 sec/batch
Epoch 4/10 Iteration: 14200 Avg. Training loss: 3.7481 0.4401 sec/batch
Epoch 4/10 Iteration: 14300 Avg. Training loss: 3.7546 0.4374 sec/batch
Epoch 4/10 Iteration: 14400 Avg. Training loss: 3.7233 0.5701 sec/batch
Epoch 4/10 Iteration: 14500 Avg. Training loss: 3.7225 0.6379 sec/batch
Epoch 4/10 Iteration: 14600 Avg. Training loss: 3.7139 0.6251 sec/batch
Epoch 4/10 Iteration: 14700 Avg. Training loss: 3.7518 0.6305 sec/batch
Epoch 4/10 Iteration: 14800 Avg. Training loss: 3.7775 0.6407 sec/batch
Epoch 4/10 Iteration: 14900 Avg. Training loss: 3.7584 0.6283 sec/batch
Epoch 4/10 Iteration: 15000 Avg. Training loss: 3.7027 0.6585 sec/batch
Nearest to four: keres, two, five, median, zero, six, household, one,
Nearest to so: airframe, lances, gela, manifests, cultic, pronominal, complained, decayed,
Nearest to d: inventor, durkheim, eisenach, biochemist, jaime, isambard, debussy, nobel,
Nearest to they: alongside, nutrient, harvesting, variability, christian, plausible, seasonally, cans,
Nearest to been: peltier, brenda, highlighted, protozoa, barometer, superiority, inevitable, informed,
Nearest to with: wray, accustomed, semiautomatic, botulism, remedied, curtail, amplifying, exploration,
Nearest to i: me, if, myself, awakened, rhymes, you, crowned, sharada,
Nearest to have: manichean, longhair, pollinated, mooring, humane, gelderland, shawnee, supernaturally,
Nearest to bible: translations, tiberian, torah, tobit, hebrew, psalms, rashi, passages,
Nearest to troops: army, ashburton, samoan, luftwaffe, udf, acceptable, intercept, unable,
Nearest to articles: suicidal, recordable, disconnected, blog, faq, midyear, newswire, federations,
Nearest to additional: cryptographically, straddle, caseless, sites, clause, superiority, kfreebsd, scrambled,
Nearest to shown: righteousness, disclosure, subpages, wherein, alleviation, tetrahedral, colours, telford,
Nearest to heavy: vangelis, organise, bagger, propel, cavalry, pinning, wiped, cortisol,
Nearest to professional: amateur, schilling, marl, negro, debilitating, marcia, conference, graduation,
Nearest to recorded: collingridge, guided, band, song, proselytes, hairy, stanhope, chaplin,
Epoch 4/10 Iteration: 15100 Avg. Training loss: 3.6883 0.6408 sec/batch
Epoch 4/10 Iteration: 15200 Avg. Training loss: 3.7049 0.6331 sec/batch
Epoch 4/10 Iteration: 15300 Avg. Training loss: 3.6560 0.6426 sec/batch
Epoch 4/10 Iteration: 15400 Avg. Training loss: 3.7308 36.4869 sec/batch
Epoch 4/10 Iteration: 15500 Avg. Training loss: 3.6982 0.6139 sec/batch
Epoch 4/10 Iteration: 15600 Avg. Training loss: 3.7216 0.6234 sec/batch
Epoch 4/10 Iteration: 15700 Avg. Training loss: 3.7170 0.6358 sec/batch
Epoch 4/10 Iteration: 15800 Avg. Training loss: 3.7565 0.6284 sec/batch
Epoch 4/10 Iteration: 15900 Avg. Training loss: 3.7011 0.6125 sec/batch
Epoch 4/10 Iteration: 16000 Avg. Training loss: 3.6932 0.6443 sec/batch
Nearest to four: keres, five, two, median, six, household, pix, zero,
Nearest to so: airframe, gela, cultic, vec, manifests, decayed, sake, lances,
Nearest to d: biochemist, inventor, durkheim, eisenach, isambard, jaime, ure, debussy,
Nearest to they: alongside, harvesting, christian, variability, nutrient, plausible, inflicted, etre,
Nearest to been: peltier, highlighted, brenda, protozoa, barometer, informed, sapiens, inevitable,
Nearest to with: wray, accustomed, botulism, semiautomatic, remedied, curtail, transitioning, vitamin,
Nearest to i: me, if, myself, you, awakened, rhymes, ve, sharada,
Nearest to have: manichean, longhair, mooring, gelderland, inspires, pollinated, supernaturally, humane,
Nearest to bible: translations, tiberian, tobit, torah, passages, hebrew, rashi, psalms,
Nearest to troops: army, ashburton, samoan, udf, luftwaffe, acceptable, soldiers, vandals,
Nearest to articles: suicidal, faq, recordable, disconnected, blog, midyear, newswire, federations,
Nearest to additional: cryptographically, straddle, caseless, sites, clause, attract, kfreebsd, superiority,
Nearest to shown: disclosure, righteousness, subpages, tetrahedral, wherein, alleviation, colours, carnac,
Nearest to heavy: vangelis, organise, bagger, propel, cavalry, amplify, wiped, cortisol,
Nearest to professional: amateur, schilling, marl, graduation, marcia, conference, negro, debilitating,
Nearest to recorded: collingridge, band, song, guided, hairy, proselytes, secretly, recordings,
Epoch 4/10 Iteration: 16100 Avg. Training loss: 3.7113 0.6328 sec/batch
Epoch 4/10 Iteration: 16200 Avg. Training loss: 3.7510 0.5898 sec/batch
Epoch 4/10 Iteration: 16300 Avg. Training loss: 3.7371 0.5844 sec/batch
Epoch 4/10 Iteration: 16400 Avg. Training loss: 3.6926 0.5688 sec/batch
Epoch 4/10 Iteration: 16500 Avg. Training loss: 3.7190 0.5652 sec/batch
Epoch 4/10 Iteration: 16600 Avg. Training loss: 3.7203 1.1250 sec/batch
Epoch 4/10 Iteration: 16700 Avg. Training loss: 3.7393 1.1033 sec/batch
Epoch 4/10 Iteration: 16800 Avg. Training loss: 3.7076 0.9977 sec/batch
Epoch 4/10 Iteration: 16900 Avg. Training loss: 3.7339 1.0242 sec/batch
Epoch 4/10 Iteration: 17000 Avg. Training loss: 3.7375 1.0773 sec/batch
Nearest to four: keres, five, two, median, six, household, pix, ckel,
Nearest to so: manifests, airframe, cultic, gela, sake, complained, lances, involuntarily,
Nearest to d: biochemist, inventor, durkheim, eisenach, jaime, isambard, debussy, ure,
Nearest to they: harvesting, alongside, christian, begging, etre, inflicted, veritable, plausible,
Nearest to been: peltier, brenda, highlighted, barometer, protozoa, informed, oviedo, kannapolis,
Nearest to with: wray, accustomed, semiautomatic, botulism, curtail, asme, remedied, kebir,
Nearest to i: me, myself, if, ve, awakened, you, crowned, rhymes,
Nearest to have: manichean, longhair, supernaturally, gelderland, pollinated, mooring, inspires, shawnee,
Nearest to bible: translations, tiberian, torah, tobit, rabbis, passages, rashi, hebrew,
Nearest to troops: army, ashburton, samoan, luftwaffe, udf, stalingrad, peacekeeping, attacked,
Nearest to articles: recordable, newswire, blog, suicidal, faq, disconnected, pages, membership,
Nearest to additional: straddle, cryptographically, clause, caseless, attract, skipper, superiority, sites,
Nearest to shown: righteousness, subpages, disclosure, tetrahedral, alleviation, telford, numerous, carnac,
Nearest to heavy: vangelis, organise, propel, cortisol, bagger, amplify, runeword, cavalry,
Nearest to professional: amateur, schilling, marl, marcia, negro, graduation, association, conference,
Nearest to recorded: collingridge, band, song, guided, flavian, hairy, lanchester, recordings,
Epoch 4/10 Iteration: 17100 Avg. Training loss: 3.6999 0.9918 sec/batch
Epoch 4/10 Iteration: 17200 Avg. Training loss: 3.6461 1.1992 sec/batch
Epoch 4/10 Iteration: 17300 Avg. Training loss: 3.6883 1.0472 sec/batch
Epoch 4/10 Iteration: 17400 Avg. Training loss: 3.7413 1.3572 sec/batch
Epoch 4/10 Iteration: 17500 Avg. Training loss: 3.7145 1.3890 sec/batch
Epoch 4/10 Iteration: 17600 Avg. Training loss: 3.7860 1.3863 sec/batch
Epoch 4/10 Iteration: 17700 Avg. Training loss: 3.8051 1.3267 sec/batch
Epoch 4/10 Iteration: 17800 Avg. Training loss: 3.6633 1.2529 sec/batch
Epoch 4/10 Iteration: 17900 Avg. Training loss: 3.6809 1.1236 sec/batch
Epoch 4/10 Iteration: 18000 Avg. Training loss: 3.6919 1.2579 sec/batch
Nearest to four: keres, five, median, two, six, rate, household, zero,
Nearest to so: manifests, cultic, airframe, gela, vec, sake, complained, lances,
Nearest to d: inventor, biochemist, durkheim, jaime, eisenach, nobel, isambard, debussy,
Nearest to they: harvesting, alongside, christian, nutrient, begging, veritable, workload, riaa,
Nearest to been: peltier, brenda, highlighted, informed, barometer, protozoa, oviedo, kannapolis,
Nearest to with: wray, accustomed, semiautomatic, botulism, asme, vitamin, remedied, curtail,
Nearest to i: me, myself, if, ve, my, you, rhymes, awakened,
Nearest to have: manichean, longhair, supernaturally, pollinated, inspires, vlsi, mooring, humane,
Nearest to bible: translations, tiberian, psalms, rashi, tobit, passages, homogenization, names,
Nearest to troops: army, ashburton, samoan, acceptable, udf, vandals, celtic, attacked,
Nearest to articles: newswire, recordable, suicidal, blog, membership, federations, disconnected, faq,
Nearest to additional: straddle, clause, cryptographically, caseless, sites, skipper, superiority, attract,
Nearest to shown: subpages, disclosure, righteousness, tetrahedral, alleviation, telford, wherein, carnac,
Nearest to heavy: vangelis, organise, propel, cortisol, bagger, runeword, amplify, schists,
Nearest to professional: amateur, marcia, graduation, negro, schilling, association, graduates, marl,
Nearest to recorded: collingridge, song, band, guided, recordings, flavian, pianist, secretly,
Epoch 4/10 Iteration: 18100 Avg. Training loss: 3.6222 1.1411 sec/batch
Epoch 4/10 Iteration: 18200 Avg. Training loss: 3.6915 1.1536 sec/batch
Epoch 4/10 Iteration: 18300 Avg. Training loss: 3.7066 1.1090 sec/batch
Epoch 4/10 Iteration: 18400 Avg. Training loss: 3.6594 1.1061 sec/batch
Epoch 4/10 Iteration: 18500 Avg. Training loss: 3.7239 1.1048 sec/batch
Epoch 5/10 Iteration: 18600 Avg. Training loss: 3.7030 1.1279 sec/batch
Epoch 5/10 Iteration: 18700 Avg. Training loss: 3.6545 1.1315 sec/batch
Epoch 5/10 Iteration: 18800 Avg. Training loss: 3.6503 1.2219 sec/batch
Epoch 5/10 Iteration: 18900 Avg. Training loss: 3.6499 1.1265 sec/batch
Epoch 5/10 Iteration: 19000 Avg. Training loss: 3.6312 1.1055 sec/batch
Nearest to four: keres, five, six, two, zero, rate, ckel, one,
Nearest to so: manifests, cultic, airframe, vec, sake, gela, complained, lances,
Nearest to d: inventor, biochemist, durkheim, jaime, eisenach, isambard, debussy, nobel,
Nearest to they: harvesting, alongside, christian, nutrient, begging, veritable, fakes, southerners,
Nearest to been: peltier, brenda, highlighted, protozoa, informed, sapiens, superiority, kannapolis,
Nearest to with: wray, accustomed, asme, semiautomatic, sherpa, remedied, vitamin, curtail,
Nearest to i: me, myself, if, ve, my, rhymes, you, crowned,
Nearest to have: manichean, supernaturally, longhair, inspires, gelderland, shawnee, hesitates, mooring,
Nearest to bible: translations, tiberian, tobit, psalms, rashi, hebrew, passages, officiate,
Nearest to troops: army, ashburton, samoan, luftwaffe, stalingrad, acceptable, celtic, armed,
Nearest to articles: membership, newswire, federations, suicidal, disconnected, covenants, recordable, stela,
Nearest to additional: straddle, caseless, clause, cryptographically, sites, superiority, skipper, pinnacles,
Nearest to shown: subpages, righteousness, tetrahedral, disclosure, wherein, alleviation, telford, carnac,
Nearest to heavy: vangelis, organise, bagger, cavalry, propel, amplify, runeword, cortisol,
Nearest to professional: amateur, marcia, marl, educational, graduation, schilling, graduates, association,
Nearest to recorded: collingridge, guided, song, band, recordings, flavian, secretly, lanchester,
Epoch 5/10 Iteration: 19100 Avg. Training loss: 3.6326 1.1241 sec/batch
Epoch 5/10 Iteration: 19200 Avg. Training loss: 3.6136 1.0964 sec/batch
Epoch 5/10 Iteration: 19300 Avg. Training loss: 3.6593 1.0966 sec/batch
Epoch 5/10 Iteration: 19400 Avg. Training loss: 3.6854 1.0885 sec/batch
Epoch 5/10 Iteration: 19500 Avg. Training loss: 3.6460 1.1700 sec/batch
Epoch 5/10 Iteration: 19600 Avg. Training loss: 3.6429 1.1094 sec/batch
Epoch 5/10 Iteration: 19700 Avg. Training loss: 3.5796 1.0558 sec/batch
Epoch 5/10 Iteration: 19800 Avg. Training loss: 3.6076 1.0517 sec/batch
Epoch 5/10 Iteration: 19900 Avg. Training loss: 3.5692 1.1093 sec/batch
Epoch 5/10 Iteration: 20000 Avg. Training loss: 3.6345 1.0636 sec/batch
Nearest to four: five, median, keres, two, six, household, rate, zero,
Nearest to so: airframe, cultic, gela, manifests, lances, sake, vec, complained,
Nearest to d: durkheim, inventor, biochemist, eisenach, jaime, isambard, debussy, ure,
Nearest to they: nutrient, harvesting, alongside, begging, workload, fakes, toned, introns,
Nearest to been: peltier, brenda, highlighted, protozoa, coalescence, barometer, informed, engenders,
Nearest to with: wray, accustomed, asme, mems, botulism, semiautomatic, sherpa, remedied,
Nearest to i: me, myself, ve, if, my, rhymes, you, awakened,
Nearest to have: manichean, inspires, longhair, supernaturally, mooring, hesitates, gelderland, rossby,
Nearest to bible: translations, tiberian, tobit, passages, homogenization, psalms, officiate, rashi,
Nearest to troops: army, ashburton, samoan, acceptable, luftwaffe, celtic, udf, stalingrad,
Nearest to articles: newswire, membership, covenants, federations, disconnected, pages, recordable, belgaum,
Nearest to additional: straddle, cryptographically, caseless, sites, superiority, skipper, clause, attract,
Nearest to shown: subpages, wherein, righteousness, alleviation, disclosure, tetrahedral, telford, dizionario,
Nearest to heavy: vangelis, organise, propel, cavalry, bagger, amplify, runeword, garage,
Nearest to professional: amateur, association, graduation, marl, graduates, educational, schilling, negro,
Nearest to recorded: collingridge, song, band, recordings, guided, secretly, piled, lanchester,
Epoch 5/10 Iteration: 20100 Avg. Training loss: 3.6395 1.0859 sec/batch
Epoch 5/10 Iteration: 20200 Avg. Training loss: 3.6531 1.0804 sec/batch
Epoch 5/10 Iteration: 20300 Avg. Training loss: 3.5996 1.0937 sec/batch
Epoch 5/10 Iteration: 20400 Avg. Training loss: 3.6556 1.1536 sec/batch
Epoch 5/10 Iteration: 20500 Avg. Training loss: 3.7056 1.2504 sec/batch
Epoch 5/10 Iteration: 20600 Avg. Training loss: 3.5924 1.3822 sec/batch
Epoch 5/10 Iteration: 20700 Avg. Training loss: 3.6457 1.3852 sec/batch
Epoch 5/10 Iteration: 20800 Avg. Training loss: 3.6343 1.2736 sec/batch
Epoch 5/10 Iteration: 20900 Avg. Training loss: 3.6443 1.2989 sec/batch
Epoch 5/10 Iteration: 21000 Avg. Training loss: 3.6287 1.1257 sec/batch
Nearest to four: five, keres, two, rate, household, six, median, nine,
Nearest to so: manifests, airframe, cultic, gela, vec, sake, lances, complained,
Nearest to d: durkheim, inventor, biochemist, eisenach, jaime, isambard, ure, debussy,
Nearest to they: alongside, begging, harvesting, workload, nutrient, christian, fakes, veritable,
Nearest to been: peltier, highlighted, brenda, protozoa, barometer, informed, coalescence, sapiens,
Nearest to with: wray, accustomed, sherpa, asme, remedied, guillotined, curtail, botulism,
Nearest to i: me, myself, ve, my, if, you, rhymes, awakened,
Nearest to have: manichean, inspires, supernaturally, longhair, hesitates, dwells, gelderland, mooring,
Nearest to bible: translations, tiberian, tobit, passages, psalms, homogenization, officiate, swann,
Nearest to troops: army, ashburton, samoan, acceptable, peacekeeping, celtic, stalingrad, udf,
Nearest to articles: newswire, membership, federations, pages, covenants, recordable, wikipedia, suicidal,
Nearest to additional: straddle, caseless, sites, cryptographically, skipper, clause, trumbull, superiority,
Nearest to shown: subpages, righteousness, asymmetric, wherein, disclosure, alleviation, tetrahedral, downplay,
Nearest to heavy: vangelis, organise, propel, cavalry, bagger, runeword, amplify, schists,
Nearest to professional: amateur, graduation, association, marl, marcia, baseball, graduates, educational,
Nearest to recorded: collingridge, song, recordings, band, secretly, guided, lanchester, cobain,
Epoch 5/10 Iteration: 21100 Avg. Training loss: 3.6543 1.3037 sec/batch
Epoch 5/10 Iteration: 21200 Avg. Training loss: 3.6347 1.1088 sec/batch
Epoch 5/10 Iteration: 21300 Avg. Training loss: 3.6259 1.0823 sec/batch
Epoch 5/10 Iteration: 21400 Avg. Training loss: 3.6589 1.1135 sec/batch
Epoch 5/10 Iteration: 21500 Avg. Training loss: 3.6868 1.0803 sec/batch
Epoch 5/10 Iteration: 21600 Avg. Training loss: 3.6334 1.0778 sec/batch
Epoch 5/10 Iteration: 21700 Avg. Training loss: 3.6172 1.0787 sec/batch
Epoch 5/10 Iteration: 21800 Avg. Training loss: 3.5709 1.0148 sec/batch
Epoch 5/10 Iteration: 21900 Avg. Training loss: 3.6171 1.0639 sec/batch
Epoch 5/10 Iteration: 22000 Avg. Training loss: 3.6829 1.0511 sec/batch
Nearest to four: five, keres, six, rate, two, median, household, runways,
Nearest to so: airframe, cultic, manifests, gela, lances, vec, complained, sake,
Nearest to d: inventor, durkheim, biochemist, eisenach, jaime, isambard, ure, debussy,
Nearest to they: alongside, nutrient, harvesting, workload, begging, riaa, veritable, christian,
Nearest to been: peltier, brenda, highlighted, protozoa, coalescence, barometer, engenders, oviedo,
Nearest to with: wray, accustomed, curtail, asme, sherpa, kebir, remedied, botulism,
Nearest to i: me, myself, ve, my, if, you, rhymes, awakened,
Nearest to have: manichean, inspires, supernaturally, hesitates, longhair, gelderland, dwells, vlsi,
Nearest to bible: translations, tiberian, tobit, prefaces, passages, swann, officiate, homogenization,
Nearest to troops: army, ashburton, samoan, acceptable, peacekeeping, celtic, luftwaffe, soldiers,
Nearest to articles: newswire, federations, membership, pages, recordable, covenants, wikipedia, midyear,
Nearest to additional: straddle, sites, caseless, clause, skipper, adsorption, superiority, cryptographically,
Nearest to shown: subpages, alleviation, wherein, righteousness, asymmetric, tetrahedral, disclosure, projective,
Nearest to heavy: vangelis, organise, propel, cavalry, runeword, bagger, amplify, schists,
Nearest to professional: amateur, graduation, association, marl, baseball, marcia, graduates, educational,
Nearest to recorded: collingridge, song, guided, recordings, band, flavian, secretly, lanchester,
Epoch 5/10 Iteration: 22100 Avg. Training loss: 3.6428 1.0150 sec/batch
Epoch 5/10 Iteration: 22200 Avg. Training loss: 3.7544 1.0316 sec/batch
Epoch 5/10 Iteration: 22300 Avg. Training loss: 3.6905 1.0401 sec/batch
Epoch 5/10 Iteration: 22400 Avg. Training loss: 3.6739 1.0005 sec/batch
Epoch 5/10 Iteration: 22500 Avg. Training loss: 3.6164 1.0286 sec/batch
Epoch 5/10 Iteration: 22600 Avg. Training loss: 3.5803 0.9881 sec/batch
Epoch 5/10 Iteration: 22700 Avg. Training loss: 3.6237 1.0060 sec/batch
Epoch 5/10 Iteration: 22800 Avg. Training loss: 3.5604 1.0077 sec/batch
Epoch 5/10 Iteration: 22900 Avg. Training loss: 3.6370 1.0715 sec/batch
Epoch 5/10 Iteration: 23000 Avg. Training loss: 3.6154 0.9892 sec/batch
Nearest to four: keres, five, rate, two, six, nine, runways, zero,
Nearest to so: airframe, cultic, manifests, vec, gela, complained, sake, lances,
Nearest to d: jaime, inventor, durkheim, biochemist, eisenach, isambard, ure, debussy,
Nearest to they: alongside, nutrient, workload, riaa, harvesting, begging, christian, veritable,
Nearest to been: peltier, brenda, highlighted, protozoa, coalescence, barometer, engenders, korchnoi,
Nearest to with: wray, accustomed, sherpa, asme, curtail, remedied, unleashes, orthogonal,
Nearest to i: me, myself, ve, my, if, you, rhymes, want,
Nearest to have: manichean, inspires, supernaturally, hesitates, dwells, longhair, vlsi, pollinated,
Nearest to bible: translations, tiberian, tobit, swann, passages, officiate, names, psalms,
Nearest to troops: army, ashburton, samoan, acceptable, celtic, luftwaffe, peacekeeping, udf,
Nearest to articles: newswire, federations, pages, membership, recordable, wikipedia, mifflin, covenants,
Nearest to additional: straddle, caseless, clause, trumbull, skipper, sites, kosinski, cryptographically,
Nearest to shown: subpages, tetrahedral, righteousness, disclosure, alleviation, asymmetric, dizionario, wherein,
Nearest to heavy: vangelis, organise, propel, runeword, bagger, garage, amplify, monopolist,
Nearest to professional: amateur, marcia, graduation, pianist, association, graduates, baseball, marl,
Nearest to recorded: collingridge, song, recordings, band, guided, secretly, cobain, piled,
Epoch 5/10 Iteration: 23100 Avg. Training loss: 3.6405 1.0470 sec/batch
Epoch 6/10 Iteration: 23200 Avg. Training loss: 3.6385 0.6503 sec/batch
Epoch 6/10 Iteration: 23300 Avg. Training loss: 3.6100 1.0224 sec/batch
Epoch 6/10 Iteration: 23400 Avg. Training loss: 3.5847 1.0198 sec/batch
Epoch 6/10 Iteration: 23500 Avg. Training loss: 3.6169 0.9868 sec/batch
Epoch 6/10 Iteration: 23600 Avg. Training loss: 3.5749 1.0346 sec/batch
Epoch 6/10 Iteration: 23700 Avg. Training loss: 3.6024 1.0496 sec/batch
Epoch 6/10 Iteration: 23800 Avg. Training loss: 3.5581 1.1119 sec/batch
Epoch 6/10 Iteration: 23900 Avg. Training loss: 3.5851 1.3456 sec/batch
Epoch 6/10 Iteration: 24000 Avg. Training loss: 3.5950 1.2524 sec/batch
Nearest to four: keres, five, rate, six, two, runways, zero, household,
Nearest to so: airframe, cultic, complained, manifests, gela, sake, lances, sittings,
Nearest to d: inventor, durkheim, jaime, biochemist, eisenach, isambard, debussy, ure,
Nearest to they: nutrient, alongside, harvesting, workload, begging, christian, introns, riaa,
Nearest to been: peltier, brenda, highlighted, protozoa, coalescence, engenders, korchnoi, informed,
Nearest to with: wray, accustomed, sherpa, remedied, asme, curtail, botulism, plenty,
Nearest to i: me, myself, ve, my, you, rhymes, if, want,
Nearest to have: manichean, inspires, supernaturally, hesitates, gelderland, longhair, dwells, shawnee,
Nearest to bible: translations, tiberian, tobit, passages, psalms, hebrew, swann, names,
Nearest to troops: army, ashburton, celtic, samoan, acceptable, attacked, luftwaffe, peacekeeping,
Nearest to articles: newswire, membership, federations, pages, covenants, wikipedia, mifflin, recordable,
Nearest to additional: straddle, trumbull, clause, kosinski, skipper, caseless, cryptographically, sites,
Nearest to shown: subpages, alleviation, ostensibly, disclosure, wherein, carnac, tetrahedral, righteousness,
Nearest to heavy: vangelis, organise, propel, amplify, bagger, cavalry, runeword, pinning,
Nearest to professional: amateur, graduation, association, graduates, baseball, marcia, marl, schilling,
Nearest to recorded: collingridge, song, recordings, guided, band, secretly, recording, lanchester,
Epoch 6/10 Iteration: 24100 Avg. Training loss: 3.5779 1.3133 sec/batch
Epoch 6/10 Iteration: 24200 Avg. Training loss: 3.6012 1.4288 sec/batch
Epoch 6/10 Iteration: 24300 Avg. Training loss: 3.4971 1.0762 sec/batch
Epoch 6/10 Iteration: 24400 Avg. Training loss: 3.5675 1.0436 sec/batch
Epoch 6/10 Iteration: 24500 Avg. Training loss: 3.5519 0.9833 sec/batch
Epoch 6/10 Iteration: 24600 Avg. Training loss: 3.5123 1.0580 sec/batch
Epoch 6/10 Iteration: 24700 Avg. Training loss: 3.5611 0.9911 sec/batch
Epoch 6/10 Iteration: 24800 Avg. Training loss: 3.5977 0.9594 sec/batch
Epoch 6/10 Iteration: 24900 Avg. Training loss: 3.5536 0.9705 sec/batch
Epoch 6/10 Iteration: 25000 Avg. Training loss: 3.5647 0.9929 sec/batch
Nearest to four: five, keres, rate, six, two, median, zero, household,
Nearest to so: airframe, cultic, gela, manifests, vec, complained, sake, hoo,
Nearest to d: eisenach, jaime, biochemist, durkheim, isambard, inventor, debussy, hydropower,
Nearest to they: nutrient, workload, begging, introns, alongside, harvesting, ince, such,
Nearest to been: peltier, highlighted, protozoa, brenda, coalescence, kannapolis, engenders, has,
Nearest to with: wray, accustomed, remedied, vitamin, curtail, sherpa, asme, plenty,
Nearest to i: me, myself, ve, my, you, if, saying, want,
Nearest to have: manichean, inspires, hesitates, compare, dwells, supernaturally, longhair, gelderland,
Nearest to bible: translations, tiberian, tobit, swann, homogenization, passages, names, psalms,
Nearest to troops: army, ashburton, samoan, luftwaffe, celtic, acceptable, peacekeeping, reichswehr,
Nearest to articles: newswire, federations, pages, covenants, membership, wikipedia, dmoz, websites,
Nearest to additional: straddle, trumbull, clause, cryptographically, caseless, kosinski, skipper, sites,
Nearest to shown: subpages, alleviation, ostensibly, tetrahedral, admitting, wherein, disclosure, asymmetric,
Nearest to heavy: vangelis, organise, cavalry, amplify, propel, bagger, runeword, cortisol,
Nearest to professional: amateur, graduation, graduates, marl, educational, association, schilling, marcia,
Nearest to recorded: collingridge, song, recordings, guided, band, secretly, cobain, recording,
Epoch 6/10 Iteration: 25100 Avg. Training loss: 3.6654 0.9573 sec/batch
Epoch 6/10 Iteration: 25200 Avg. Training loss: 3.5741 0.9620 sec/batch
Epoch 6/10 Iteration: 25300 Avg. Training loss: 3.5492 1.0576 sec/batch
Epoch 6/10 Iteration: 25400 Avg. Training loss: 3.5878 0.9644 sec/batch
Epoch 6/10 Iteration: 25500 Avg. Training loss: 3.5733 0.9393 sec/batch
Epoch 6/10 Iteration: 25600 Avg. Training loss: 3.5683 0.9798 sec/batch
Epoch 6/10 Iteration: 25700 Avg. Training loss: 3.6012 0.9593 sec/batch
Epoch 6/10 Iteration: 25800 Avg. Training loss: 3.5915 0.9719 sec/batch
Epoch 6/10 Iteration: 25900 Avg. Training loss: 3.5555 0.9440 sec/batch
Epoch 6/10 Iteration: 26000 Avg. Training loss: 3.6132 0.9683 sec/batch
Nearest to four: five, keres, rate, two, runways, nine, median, six,
Nearest to so: cultic, airframe, manifests, gela, vec, complained, sake, hoo,
Nearest to d: jaime, biochemist, eisenach, durkheim, inventor, isambard, ure, debussy,
Nearest to they: begging, nutrient, alongside, workload, veritable, harvesting, christian, introns,
Nearest to been: peltier, highlighted, brenda, protozoa, has, coalescence, kannapolis, recent,
Nearest to with: wray, sherpa, accustomed, vitamin, curtail, guillotined, asme, remedied,
Nearest to i: me, myself, ve, my, rhymes, want, saying, you,
Nearest to have: manichean, inspires, hesitates, supernaturally, compare, dwells, gelderland, rakhine,
Nearest to bible: translations, tiberian, swann, homogenization, names, tobit, endnotes, passages,
Nearest to troops: army, ashburton, luftwaffe, celtic, samoan, regrouped, soldiers, attacked,
Nearest to articles: newswire, pages, federations, membership, covenants, wikipedia, mifflin, dmoz,
Nearest to additional: straddle, clause, trumbull, cryptographically, skipper, caseless, adsorption, kosinski,
Nearest to shown: subpages, alleviation, tetrahedral, ostensibly, projective, asymmetric, telford, dizionario,
Nearest to heavy: vangelis, organise, amplify, cavalry, propel, monopolist, cortisol, bagger,
Nearest to professional: amateur, graduation, graduates, marl, baseball, association, marcia, educational,
Nearest to recorded: collingridge, song, recordings, secretly, band, guided, recording, lanchester,
Epoch 6/10 Iteration: 26100 Avg. Training loss: 3.6029 0.9504 sec/batch
Epoch 6/10 Iteration: 26200 Avg. Training loss: 3.5919 0.9599 sec/batch
Epoch 6/10 Iteration: 26300 Avg. Training loss: 3.5581 1.0450 sec/batch
Epoch 6/10 Iteration: 26400 Avg. Training loss: 3.5496 0.9501 sec/batch
Epoch 6/10 Iteration: 26500 Avg. Training loss: 3.5679 0.8724 sec/batch
Epoch 6/10 Iteration: 26600 Avg. Training loss: 3.6081 1.0085 sec/batch
Epoch 6/10 Iteration: 26700 Avg. Training loss: 3.5859 1.0317 sec/batch
Epoch 6/10 Iteration: 26800 Avg. Training loss: 3.6782 1.0190 sec/batch
Epoch 6/10 Iteration: 26900 Avg. Training loss: 3.6699 1.0708 sec/batch
Epoch 6/10 Iteration: 27000 Avg. Training loss: 3.6253 1.0079 sec/batch
Nearest to four: keres, five, six, rate, nine, runways, two, zero,
Nearest to so: cultic, airframe, manifests, gela, lances, vec, complained, hoo,
Nearest to d: jaime, inventor, eisenach, durkheim, isambard, biochemist, narrator, ure,
Nearest to they: nutrient, alongside, workload, begging, riaa, harvesting, introns, veritable,
Nearest to been: peltier, brenda, highlighted, has, protozoa, coalescence, kannapolis, recent,
Nearest to with: wray, sherpa, accustomed, curtail, guillotined, unleashes, plenty, asme,
Nearest to i: me, myself, ve, my, want, saying, rhymes, you,
Nearest to have: manichean, inspires, hesitates, dwells, supernaturally, compare, headless, longhair,
Nearest to bible: swann, tiberian, translations, names, endnotes, homogenization, tobit, clavichord,
Nearest to troops: army, ashburton, samoan, celtic, reichswehr, attacked, luftwaffe, sacu,
Nearest to articles: newswire, membership, federations, pages, mifflin, dmoz, wikipedia, belgaum,
Nearest to additional: straddle, clause, trumbull, kosinski, foretelling, skipper, winner, caseless,
Nearest to shown: subpages, tetrahedral, ostensibly, alleviation, asymmetric, disclosure, wherein, hashem,
Nearest to heavy: vangelis, organise, amplify, propel, cortisol, eretz, monopolist, cavalry,
Nearest to professional: amateur, graduates, graduation, baseball, marcia, wrestling, association, pianist,
Nearest to recorded: collingridge, song, recordings, guided, secretly, band, flavian, compilation,
Epoch 6/10 Iteration: 27100 Avg. Training loss: 3.5579 1.0812 sec/batch
Epoch 6/10 Iteration: 27200 Avg. Training loss: 3.5052 0.9635 sec/batch
Epoch 6/10 Iteration: 27300 Avg. Training loss: 3.5748 1.3424 sec/batch
Epoch 6/10 Iteration: 27400 Avg. Training loss: 3.5114 1.3121 sec/batch
Epoch 6/10 Iteration: 27500 Avg. Training loss: 3.5992 1.0554 sec/batch
Epoch 6/10 Iteration: 27600 Avg. Training loss: 3.5967 1.0559 sec/batch
Epoch 6/10 Iteration: 27700 Avg. Training loss: 3.5630 1.0210 sec/batch
Epoch 7/10 Iteration: 27800 Avg. Training loss: 3.6556 0.4251 sec/batch
Epoch 7/10 Iteration: 27900 Avg. Training loss: 3.5610 1.0990 sec/batch
Epoch 7/10 Iteration: 28000 Avg. Training loss: 3.5525 0.9676 sec/batch
Nearest to four: keres, five, rate, six, two, runways, ckel, zero,
Nearest to so: cultic, airframe, gela, manifests, vec, lances, complained, hoo,
Nearest to d: durkheim, inventor, jaime, eisenach, isambard, biochemist, narrator, ure,
Nearest to they: begging, nutrient, workload, alongside, were, veritable, introns, harvesting,
Nearest to been: peltier, highlighted, brenda, has, protozoa, korchnoi, recent, member,
Nearest to with: wray, sherpa, accustomed, curtail, guillotined, unleashes, asme, plenty,
Nearest to i: myself, me, ve, my, want, rhymes, saying, you,
Nearest to have: inspires, manichean, hesitates, dwells, supernaturally, rakhine, headless, compare,
Nearest to bible: tiberian, swann, translations, tobit, endnotes, loyola, homogenization, names,
Nearest to troops: army, ashburton, celtic, samoan, reichswehr, udf, luftwaffe, sacu,
Nearest to articles: newswire, membership, federations, covenants, pages, wikipedia, mifflin, belgaum,
Nearest to additional: straddle, clause, trumbull, kosinski, foretelling, caseless, skipper, doran,
Nearest to shown: tetrahedral, subpages, asymmetric, ostensibly, projective, alleviation, downplay, dizionario,
Nearest to heavy: vangelis, organise, propel, amplify, cortisol, garage, eretz, bagger,
Nearest to professional: amateur, graduates, wrestling, marcia, graduation, baseball, association, educational,
Nearest to recorded: collingridge, recordings, song, band, guided, secretly, cobain, flavian,
Epoch 7/10 Iteration: 28100 Avg. Training loss: 3.6019 1.0506 sec/batch
Epoch 7/10 Iteration: 28200 Avg. Training loss: 3.5589 1.0196 sec/batch
Epoch 7/10 Iteration: 28300 Avg. Training loss: 3.5409 0.9729 sec/batch
Epoch 7/10 Iteration: 28400 Avg. Training loss: 3.5422 1.0804 sec/batch
Epoch 7/10 Iteration: 28500 Avg. Training loss: 3.5151 0.9413 sec/batch
Epoch 7/10 Iteration: 28600 Avg. Training loss: 3.5517 1.0200 sec/batch
Epoch 7/10 Iteration: 28700 Avg. Training loss: 3.5416 1.0672 sec/batch
Epoch 7/10 Iteration: 28800 Avg. Training loss: 3.5933 0.9147 sec/batch
Epoch 7/10 Iteration: 28900 Avg. Training loss: 3.4669 1.0584 sec/batch
Epoch 7/10 Iteration: 29000 Avg. Training loss: 3.4962 0.9843 sec/batch
Nearest to four: five, keres, rate, six, two, runways, zero, median,
Nearest to so: airframe, cultic, gela, lances, vec, lebesgue, complained, manifests,
Nearest to d: durkheim, eisenach, inventor, isambard, jaime, narrator, biochemist, hydropower,
Nearest to they: nutrient, begging, alongside, workload, were, harvesting, riaa, introns,
Nearest to been: has, brenda, peltier, highlighted, protozoa, recent, coalescence, engenders,
Nearest to with: wray, sherpa, accustomed, asme, curtail, plenty, unleashes, dagesh,
Nearest to i: myself, me, ve, my, want, rhymes, saying, you,
Nearest to have: manichean, inspires, hesitates, rakhine, gelderland, compare, longhair, rossby,
Nearest to bible: translations, tiberian, swann, tobit, hebrew, passages, endnotes, names,
Nearest to troops: army, ashburton, celtic, soldiers, luftwaffe, samoan, sacu, reichswehr,
Nearest to articles: newswire, membership, federations, covenants, pages, mifflin, wikipedia, belgaum,
Nearest to additional: straddle, trumbull, clause, kosinski, adsorption, skipper, foretelling, caseless,
Nearest to shown: tetrahedral, ostensibly, alleviation, subpages, downplay, asymmetric, carnac, wherein,
Nearest to heavy: vangelis, organise, propel, armoured, cortisol, cavalry, garage, monopolist,
Nearest to professional: amateur, graduates, graduation, wrestling, association, baseball, educational, marl,
Nearest to recorded: collingridge, recordings, song, band, secretly, guided, recording, flavian,
Epoch 7/10 Iteration: 29100 Avg. Training loss: 3.5403 0.9519 sec/batch
Epoch 7/10 Iteration: 29200 Avg. Training loss: 3.4548 0.9967 sec/batch
Epoch 7/10 Iteration: 29300 Avg. Training loss: 3.5402 0.9989 sec/batch
Epoch 7/10 Iteration: 29400 Avg. Training loss: 3.5691 0.9390 sec/batch
Epoch 7/10 Iteration: 29500 Avg. Training loss: 3.5175 0.9705 sec/batch
Epoch 7/10 Iteration: 29600 Avg. Training loss: 3.5388 0.9731 sec/batch
Epoch 7/10 Iteration: 29700 Avg. Training loss: 3.5900 0.9791 sec/batch
Epoch 7/10 Iteration: 29800 Avg. Training loss: 3.5327 1.0459 sec/batch
Epoch 7/10 Iteration: 29900 Avg. Training loss: 3.5320 0.9517 sec/batch
Epoch 7/10 Iteration: 30000 Avg. Training loss: 3.5371 0.9993 sec/batch
Nearest to four: keres, five, rate, six, two, median, runways, couples,
Nearest to so: airframe, cultic, gela, vec, manifests, lances, lebesgue, complained,
Nearest to d: durkheim, eisenach, jaime, isambard, narrator, inventor, ure, biochemist,
Nearest to they: begging, nutrient, workload, were, alongside, christian, introns, kosher,
Nearest to been: has, brenda, highlighted, peltier, protozoa, recent, coalescence, korchnoi,
Nearest to with: wray, accustomed, sherpa, curtail, vitamin, asme, guillotined, dagesh,
Nearest to i: myself, me, ve, my, want, saying, rhymes, you,
Nearest to have: manichean, inspires, hesitates, compare, dwells, rakhine, supernaturally, longhair,
Nearest to bible: swann, tiberian, translations, endnotes, homogenization, tobit, hebrew, loading,
Nearest to troops: army, ashburton, celtic, luftwaffe, reichswehr, soldiers, sacu, samoan,
Nearest to articles: newswire, covenants, pages, federations, membership, wikipedia, encyclop, mifflin,
Nearest to additional: straddle, trumbull, clause, kosinski, adsorption, skipper, winner, hardwoods,
Nearest to shown: tetrahedral, asymmetric, block, projective, alleviation, ostensibly, wherein, subpages,
Nearest to heavy: vangelis, organise, propel, monopolist, garage, amplify, bagger, cortisol,
Nearest to professional: amateur, graduates, wrestling, graduation, pianist, baseball, association, marcia,
Nearest to recorded: song, collingridge, recordings, band, guided, secretly, recording, compilation,
Epoch 7/10 Iteration: 30100 Avg. Training loss: 3.5582 0.9966 sec/batch
Epoch 7/10 Iteration: 30200 Avg. Training loss: 3.5917 0.9697 sec/batch
Epoch 7/10 Iteration: 30300 Avg. Training loss: 3.5079 0.9925 sec/batch
Epoch 7/10 Iteration: 30400 Avg. Training loss: 3.5098 0.9980 sec/batch
Epoch 7/10 Iteration: 30500 Avg. Training loss: 3.5518 1.0788 sec/batch
Epoch 7/10 Iteration: 30600 Avg. Training loss: 3.5363 0.9977 sec/batch
Epoch 7/10 Iteration: 30700 Avg. Training loss: 3.5460 1.0018 sec/batch
Epoch 7/10 Iteration: 30800 Avg. Training loss: 3.5352 1.1768 sec/batch
Epoch 7/10 Iteration: 30900 Avg. Training loss: 3.5642 1.3462 sec/batch
Epoch 7/10 Iteration: 31000 Avg. Training loss: 3.5071 1.1141 sec/batch
Nearest to four: five, keres, rate, six, two, median, runways, nine,
Nearest to so: airframe, cultic, gela, lances, manifests, aramean, complained, lebesgue,
Nearest to d: durkheim, eisenach, jaime, isambard, hydropower, narrator, inventor, ure,
Nearest to they: begging, nutrient, were, alongside, introns, harvesting, workload, veritable,
Nearest to been: has, brenda, highlighted, recent, peltier, protozoa, coalescence, member,
Nearest to with: wray, sherpa, accustomed, curtail, vitamin, asme, unleashes, guillotined,
Nearest to i: myself, me, ve, my, want, saying, you, rhymes,
Nearest to have: manichean, inspires, hesitates, rakhine, fora, rica, dwells, longhair,
Nearest to bible: swann, endnotes, tiberian, homogenization, translations, loyola, tobit, passages,
Nearest to troops: army, ashburton, celtic, soldiers, attacked, reichswehr, vandals, regrouped,
Nearest to articles: newswire, wikipedia, pages, covenants, federations, membership, websites, belgaum,
Nearest to additional: straddle, trumbull, clause, adsorption, hardwoods, kosinski, skipper, winner,
Nearest to shown: tetrahedral, asymmetric, ostensibly, block, alleviation, hashem, wherein, subpages,
Nearest to heavy: vangelis, organise, garage, monopolist, orbison, cortisol, amplify, propel,
Nearest to professional: graduates, amateur, wrestling, baseball, association, graduation, pianist, negro,
Nearest to recorded: song, collingridge, recordings, recording, band, flavian, compilation, phenomenological,
Epoch 7/10 Iteration: 31100 Avg. Training loss: 3.5360 1.0573 sec/batch
Epoch 7/10 Iteration: 31200 Avg. Training loss: 3.5607 1.0357 sec/batch
Epoch 7/10 Iteration: 31300 Avg. Training loss: 3.5516 1.1308 sec/batch
Epoch 7/10 Iteration: 31400 Avg. Training loss: 3.6094 1.0174 sec/batch
Epoch 7/10 Iteration: 31500 Avg. Training loss: 3.6256 1.0597 sec/batch
Epoch 7/10 Iteration: 31600 Avg. Training loss: 3.6078 1.0296 sec/batch
Epoch 7/10 Iteration: 31700 Avg. Training loss: 3.5686 1.0257 sec/batch
Epoch 7/10 Iteration: 31800 Avg. Training loss: 3.4849 0.9875 sec/batch
Epoch 7/10 Iteration: 31900 Avg. Training loss: 3.5372 0.9982 sec/batch
Epoch 7/10 Iteration: 32000 Avg. Training loss: 3.4446 0.9938 sec/batch
Nearest to four: keres, five, rate, six, two, median, runways, zero,
Nearest to so: airframe, cultic, lances, gela, manifests, vec, aramean, complained,
Nearest to d: eisenach, jaime, durkheim, narrator, inventor, isambard, ure, hydropower,
Nearest to they: nutrient, were, workload, riaa, veritable, begging, introns, harvesting,
Nearest to been: has, peltier, brenda, highlighted, recent, korchnoi, kannapolis, coalescence,
Nearest to with: wray, sherpa, accustomed, vitamin, guillotined, unleashes, plenty, curtail,
Nearest to i: myself, me, ve, my, want, rhymes, saying, you,
Nearest to have: manichean, inspires, rica, dwells, headless, longhair, hesitates, subunit,
Nearest to bible: tiberian, swann, endnotes, homogenization, ve, loyola, clavichord, loading,
Nearest to troops: army, ashburton, soldiers, celtic, reichswehr, udf, samoan, sacu,
Nearest to articles: newswire, federations, wikipedia, covenants, pages, belgaum, membership, mifflin,
Nearest to additional: straddle, trumbull, kosinski, clause, adsorption, skipper, winner, foretelling,
Nearest to shown: tetrahedral, block, asymmetric, ostensibly, hashem, alleviation, projective, subpages,
Nearest to heavy: vangelis, organise, garage, monopolist, orbison, cortisol, corbijn, propel,
Nearest to professional: graduates, wrestling, baseball, amateur, wrestler, pianist, negro, marcia,
Nearest to recorded: song, collingridge, recordings, recording, band, flavian, compilation, guided,
Epoch 7/10 Iteration: 32100 Avg. Training loss: 3.5639 1.0084 sec/batch
Epoch 7/10 Iteration: 32200 Avg. Training loss: 3.5536 1.0240 sec/batch
Epoch 7/10 Iteration: 32300 Avg. Training loss: 3.5072 0.9662 sec/batch
Epoch 8/10 Iteration: 32400 Avg. Training loss: 3.5751 0.0546 sec/batch
Epoch 8/10 Iteration: 32500 Avg. Training loss: 3.5478 0.4362 sec/batch
Epoch 8/10 Iteration: 32600 Avg. Training loss: 3.5151 0.4288 sec/batch
Epoch 8/10 Iteration: 32700 Avg. Training loss: 3.5196 0.4292 sec/batch
Epoch 8/10 Iteration: 32800 Avg. Training loss: 3.5406 0.4284 sec/batch
Epoch 8/10 Iteration: 32900 Avg. Training loss: 3.4846 0.4272 sec/batch
Epoch 8/10 Iteration: 33000 Avg. Training loss: 3.5141 0.4277 sec/batch
Nearest to four: keres, five, six, rate, median, nine, couples, combi,
Nearest to so: airframe, cultic, lances, gela, manifests, complained, lebesgue, aramean,
Nearest to d: inventor, jaime, eisenach, durkheim, isambard, narrator, debussy, biochemist,
Nearest to they: were, nutrient, begging, christian, harvesting, alongside, workload, veritable,
Nearest to been: has, peltier, brenda, highlighted, recent, protozoa, korchnoi, coalescence,
Nearest to with: wray, sherpa, curtail, accustomed, vitamin, plenty, guillotined, asme,
Nearest to i: myself, me, ve, my, want, rhymes, saying, sure,
Nearest to have: manichean, inspires, dwells, hesitates, longhair, headless, rakhine, subunit,
Nearest to bible: endnotes, swann, tiberian, homogenization, ve, loading, loyola, clavichord,
Nearest to troops: army, ashburton, soldiers, celtic, sacu, brigade, reichswehr, udf,
Nearest to articles: newswire, covenants, federations, wikipedia, membership, belgaum, pages, websites,
Nearest to additional: straddle, kosinski, trumbull, skipper, clause, pickford, license, herlihy,
Nearest to shown: tetrahedral, block, asymmetric, ostensibly, alleviation, downplay, admitting, hashem,
Nearest to heavy: vangelis, organise, garage, monopolist, orbison, bagger, propel, armoured,
Nearest to professional: graduates, amateur, baseball, wrestling, pianist, marcia, association, graduation,
Nearest to recorded: recordings, collingridge, song, recording, band, flavian, compilation, guided,
Epoch 8/10 Iteration: 33100 Avg. Training loss: 3.4633 0.5543 sec/batch
Epoch 8/10 Iteration: 33200 Avg. Training loss: 3.5277 0.6762 sec/batch
Epoch 8/10 Iteration: 33300 Avg. Training loss: 3.5423 0.6264 sec/batch
Epoch 8/10 Iteration: 33400 Avg. Training loss: 3.5416 0.6374 sec/batch
Epoch 8/10 Iteration: 33500 Avg. Training loss: 3.4805 0.6561 sec/batch
Epoch 8/10 Iteration: 33600 Avg. Training loss: 3.4852 0.6325 sec/batch
Epoch 8/10 Iteration: 33700 Avg. Training loss: 3.4820 0.6525 sec/batch
Epoch 8/10 Iteration: 33800 Avg. Training loss: 3.4219 0.5781 sec/batch
Epoch 8/10 Iteration: 33900 Avg. Training loss: 3.4949 0.5026 sec/batch
Epoch 8/10 Iteration: 34000 Avg. Training loss: 3.5208 0.4829 sec/batch
Nearest to four: five, keres, six, rate, median, couples, per, two,
Nearest to so: airframe, cultic, lances, gela, lebesgue, sake, medes, complained,
Nearest to d: narrator, hydropower, inventor, jaime, durkheim, eisenach, isambard, frisbee,
Nearest to they: were, begging, nutrient, workload, alongside, introns, veritable, such,
Nearest to been: has, brenda, highlighted, recent, peltier, protozoa, coalescence, kannapolis,
Nearest to with: wray, curtail, sherpa, accustomed, vitamin, asme, plenty, remedied,
Nearest to i: myself, me, ve, my, want, saying, rhymes, you,
Nearest to have: manichean, inspires, subunit, hesitates, longhair, dwells, headless, been,
Nearest to bible: swann, homogenization, tiberian, endnotes, officiate, yehoshua, tobit, ve,
Nearest to troops: army, celtic, ashburton, sacu, paleography, soldiers, vandals, udf,
Nearest to articles: newswire, covenants, federations, pages, belgaum, websites, membership, wikipedia,
Nearest to additional: straddle, trumbull, kosinski, clause, skipper, winner, hardwoods, license,
Nearest to shown: block, tetrahedral, asymmetric, alleviation, admitting, ostensibly, transmembrane, wherein,
Nearest to heavy: vangelis, organise, bagger, armoured, orbison, garage, cavalry, propel,
Nearest to professional: graduates, wrestling, amateur, baseball, association, graduation, pianist, marl,
Nearest to recorded: recordings, collingridge, song, recording, band, compilation, flavian, guided,
Epoch 8/10 Iteration: 34100 Avg. Training loss: 3.5231 0.4839 sec/batch
Epoch 8/10 Iteration: 34200 Avg. Training loss: 3.5095 0.4770 sec/batch
Epoch 8/10 Iteration: 34300 Avg. Training loss: 3.5544 0.4663 sec/batch
Epoch 8/10 Iteration: 34400 Avg. Training loss: 3.5052 0.4619 sec/batch
Epoch 8/10 Iteration: 34500 Avg. Training loss: 3.4937 0.4649 sec/batch
Epoch 8/10 Iteration: 34600 Avg. Training loss: 3.4917 0.4681 sec/batch
Epoch 8/10 Iteration: 34700 Avg. Training loss: 3.5087 0.4593 sec/batch
Epoch 8/10 Iteration: 34800 Avg. Training loss: 3.5290 0.4630 sec/batch
Epoch 8/10 Iteration: 34900 Avg. Training loss: 3.4885 0.4611 sec/batch
Epoch 8/10 Iteration: 35000 Avg. Training loss: 3.5449 0.4589 sec/batch
Nearest to four: five, keres, six, rate, median, two, births, nine,
Nearest to so: airframe, cultic, gela, lebesgue, lances, manifests, ttel, sake,
Nearest to d: jaime, narrator, inventor, isambard, durkheim, eisenach, hydropower, ure,
Nearest to they: were, begging, workload, veritable, alongside, introns, nutrient, kon,
Nearest to been: has, highlighted, recent, brenda, peltier, protozoa, coalescence, korchnoi,
Nearest to with: wray, sherpa, accustomed, vitamin, curtail, asme, guillotined, plenty,
Nearest to i: myself, ve, me, want, my, saying, rhymes, windy,
Nearest to have: manichean, inspires, hesitates, been, subunit, merisi, rica, longhair,
Nearest to bible: swann, homogenization, tiberian, endnotes, ve, officiate, yehoshua, tobit,
Nearest to troops: army, celtic, sacu, ashburton, soldiers, attacked, paleography, reichswehr,
Nearest to articles: newswire, pages, federations, wikipedia, covenants, belgaum, encyclop, membership,
Nearest to additional: straddle, trumbull, kosinski, clause, winner, adsorption, skipper, doran,
Nearest to shown: block, asymmetric, tetrahedral, alleviation, ostensibly, hashem, downplay, below,
Nearest to heavy: vangelis, organise, monopolist, garage, orbison, bagger, cavalry, armoured,
Nearest to professional: wrestling, graduates, amateur, baseball, association, pianist, marl, wrestler,
Nearest to recorded: collingridge, song, recordings, recording, band, coltrane, phenomenological, flavian,
Epoch 8/10 Iteration: 35100 Avg. Training loss: 3.5017 0.4654 sec/batch
Epoch 8/10 Iteration: 35200 Avg. Training loss: 3.5075 0.4606 sec/batch
Epoch 8/10 Iteration: 35300 Avg. Training loss: 3.5690 0.4674 sec/batch
Epoch 8/10 Iteration: 35400 Avg. Training loss: 3.5357 0.4791 sec/batch
Epoch 8/10 Iteration: 35500 Avg. Training loss: 3.5266 0.5032 sec/batch
Epoch 8/10 Iteration: 35600 Avg. Training loss: 3.4993 0.7116 sec/batch
Epoch 8/10 Iteration: 35700 Avg. Training loss: 3.4983 0.4763 sec/batch
Epoch 8/10 Iteration: 35800 Avg. Training loss: 3.5017 0.4767 sec/batch
Epoch 8/10 Iteration: 35900 Avg. Training loss: 3.5811 0.4760 sec/batch
Epoch 8/10 Iteration: 36000 Avg. Training loss: 3.5041 0.5474 sec/batch
Nearest to four: five, keres, six, nine, births, rate, two, combi,
Nearest to so: lances, airframe, cultic, gela, lebesgue, ttel, comneni, medes,
Nearest to d: inventor, jaime, narrator, isambard, eisenach, marat, poet, durkheim,
Nearest to they: were, workload, begging, alongside, veritable, introns, kon, disgraced,
Nearest to been: has, brenda, recent, highlighted, peltier, coalescence, korchnoi, announced,
Nearest to with: wray, sherpa, accustomed, vitamin, kebir, curtail, plenty, guise,
Nearest to i: myself, me, ve, want, my, saying, windy, rhymes,
Nearest to have: manichean, inspires, hesitates, merisi, dwells, rakhine, rossby, rica,
Nearest to bible: swann, endnotes, tiberian, homogenization, ve, loading, clavichord, colossians,
Nearest to troops: army, celtic, ashburton, soldiers, sacu, attacked, paleography, forearms,
Nearest to articles: newswire, pages, wikipedia, federations, belgaum, covenants, membership, encyclop,
Nearest to additional: straddle, kosinski, trumbull, clause, adsorption, winner, hardwoods, pickford,
Nearest to shown: block, tetrahedral, hashem, ostensibly, asymmetric, below, alleviation, incredulity,
Nearest to heavy: vangelis, organise, garage, monopolist, orbison, umlaut, bagger, eretz,
Nearest to professional: wrestling, graduates, baseball, wrestler, pianist, amateur, association, negro,
Nearest to recorded: collingridge, recordings, song, compilation, coltrane, flavian, phenomenological, recording,
Epoch 8/10 Iteration: 36100 Avg. Training loss: 3.6206 0.4849 sec/batch
Epoch 8/10 Iteration: 36200 Avg. Training loss: 3.6082 0.5152 sec/batch
Epoch 8/10 Iteration: 36300 Avg. Training loss: 3.5219 0.5560 sec/batch
Epoch 8/10 Iteration: 36400 Avg. Training loss: 3.4747 0.5197 sec/batch
Epoch 8/10 Iteration: 36500 Avg. Training loss: 3.5132 0.4878 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 [ ]:

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

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



In [ ]:

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



In [ ]:

    
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 [ ]: