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 [23]:

    
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 [24]:

    
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 [25]:

    
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 [26]:

    
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 [27]:

    
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 [28]:

    
## Your code here
from collections import Counter
import random

counter = Counter(int_words)
total_count = len(int_words)
word_freq = {word: counter[word]/total_count for word in counter}
t = 1e-5
P = lambda word: 1 - np.sqrt(t/word_freq[word])
train_words = [word for word in int_words if P(word) < random.random()]  # The final subsampled word list

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 [29]:

    
def get_target(words, idx, window_size=5):
    ''' Get a list of words in a window around an index. '''

    # Your code here
    R = random.randint(1, window_size+1)
    before_words = words[max(0, idx-R): idx]
    after_words = words[idx+1:min(len(words), idx+1+R)]
    return list(set(before_words + after_words))

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



In [30]:

    
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 [31]:

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

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 [32]:

    
n_vocab = len(int_to_vocab)
n_embedding = 300 # Number of embedding features 
with train_graph.as_default():
    embedding = tf.Variable(tf.random_uniform([n_vocab, n_embedding], -1.0, 1.0))  # create embedding weight matrix here
    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 [33]:

    
# 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 [34]:

    
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 [35]:

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









    



mkdir: cannot create directory ‘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 [36]:

    
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: 5.7846 0.1337 sec/batch
Epoch 1/10 Iteration: 200 Avg. Training loss: 5.7257 0.1339 sec/batch
Epoch 1/10 Iteration: 300 Avg. Training loss: 5.6063 0.1340 sec/batch
Epoch 1/10 Iteration: 400 Avg. Training loss: 5.7121 0.1338 sec/batch
Epoch 1/10 Iteration: 500 Avg. Training loss: 5.6427 0.1323 sec/batch
Epoch 1/10 Iteration: 600 Avg. Training loss: 5.6568 0.1325 sec/batch
Epoch 1/10 Iteration: 700 Avg. Training loss: 5.6675 0.1340 sec/batch
Epoch 1/10 Iteration: 800 Avg. Training loss: 5.6584 0.1382 sec/batch
Epoch 1/10 Iteration: 900 Avg. Training loss: 5.5502 0.1442 sec/batch
Epoch 1/10 Iteration: 1000 Avg. Training loss: 5.5248 0.1477 sec/batch
Nearest to or: terrain, trade, mastered, neverwinter, combatted, armoured, afghani, trolleys,
Nearest to than: taste, lawrencium, masses, monos, undermine, fortified, countess, taping,
Nearest to d: raffarin, mwi, martinique, niggers, bundeswehr, zoroastrianism, alomar, sumptuous,
Nearest to about: round, dubbed, enforcement, triglyceride, pwnage, jake, kazan, conditions,
Nearest to zero: shale, completion, tasman, wp, earned, sleep, orozco, invincibility,
Nearest to five: carnivorous, cappadocian, thelema, elite, bid, shut, lawrence, reunion,
Nearest to th: troubles, newswire, boulevard, fainted, defender, recommendation, botulism, hijacking,
Nearest to more: funk, ismaili, pt, ditko, clary, scummvm, stanis, lott,
Nearest to alternative: grammarians, concisely, genetics, asean, blish, deprives, intelligence, theatrum,
Nearest to prince: presuming, pelts, interchange, believers, prd, iditarod, vaporized, bukhari,
Nearest to police: motif, xxxv, doris, outweigh, bevel, lichfield, sidereal, iasi,
Nearest to professional: students, quo, ugaritic, oscillates, smale, riebeeck, melancholic, stitch,
Nearest to construction: linseed, schoolmates, hephaestion, bluegrass, diophantus, lasker, bulky, walesa,
Nearest to http: nicholls, reiterated, zoom, disenchanted, tangle, topless, christchurch, scathing,
Nearest to issue: hittites, haine, primigenius, album, frightened, sackville, incongruous, resonances,
Nearest to behind: sending, sieur, demonstrating, brewery, felicity, forefront, ceases, breech,
Epoch 1/10 Iteration: 1100 Avg. Training loss: 5.5731 0.1519 sec/batch
Epoch 1/10 Iteration: 1200 Avg. Training loss: 5.4393 0.1533 sec/batch
Epoch 1/10 Iteration: 1300 Avg. Training loss: 5.3907 0.1562 sec/batch
Epoch 1/10 Iteration: 1400 Avg. Training loss: 5.2820 0.1580 sec/batch
Epoch 1/10 Iteration: 1500 Avg. Training loss: 5.2168 0.1566 sec/batch
Epoch 1/10 Iteration: 1600 Avg. Training loss: 5.2383 0.1587 sec/batch
Epoch 1/10 Iteration: 1700 Avg. Training loss: 5.1674 0.1598 sec/batch
Epoch 1/10 Iteration: 1800 Avg. Training loss: 5.1338 0.1604 sec/batch
Epoch 1/10 Iteration: 1900 Avg. Training loss: 5.0735 0.1590 sec/batch
Epoch 1/10 Iteration: 2000 Avg. Training loss: 5.0355 0.1614 sec/batch
Nearest to or: terrain, trade, mastered, their, time, cubic, traits, armoured,
Nearest to than: case, taste, extensively, masses, fortified, undermine, pillars, assistants,
Nearest to d: martinique, u, raffarin, mwi, similar, lake, somali, alomar,
Nearest to about: round, expected, dubbed, enforcement, conditions, cold, get, sq,
Nearest to zero: completion, shale, earned, ginger, tasman, sleep, pearl, wp,
Nearest to five: shut, cappadocian, elite, lawrence, junior, carnivorous, reunion, bid,
Nearest to th: troubles, recommendation, boulevard, defender, hijacking, newswire, depart, tutorial,
Nearest to more: funk, capital, ismaili, manner, pt, ditko, cycles, braithwaite,
Nearest to alternative: concisely, intelligence, genetics, grammarians, isotope, asean, cylinders, maintains,
Nearest to prince: presuming, pelts, believers, interchange, prd, ec, defenseless, american,
Nearest to police: motif, doris, xxxv, resistance, outweigh, lichfield, king, bevel,
Nearest to professional: students, quo, melancholic, ugaritic, animals, smale, manifested, oscillates,
Nearest to construction: linseed, bluegrass, schoolmates, bulky, diophantus, helped, lasker, contradicts,
Nearest to http: nicholls, disenchanted, reiterated, elm, zoom, tangle, topless, differently,
Nearest to issue: hittites, making, album, would, name, haine, research, surround,
Nearest to behind: sending, demonstrating, ceases, forefront, sieur, scope, meters, erica,
Epoch 1/10 Iteration: 2100 Avg. Training loss: 4.9588 0.1615 sec/batch
Epoch 1/10 Iteration: 2200 Avg. Training loss: 4.9363 0.1610 sec/batch
Epoch 1/10 Iteration: 2300 Avg. Training loss: 4.9147 0.1617 sec/batch
Epoch 1/10 Iteration: 2400 Avg. Training loss: 4.9129 0.1620 sec/batch
Epoch 1/10 Iteration: 2500 Avg. Training loss: 4.8855 0.1640 sec/batch
Epoch 1/10 Iteration: 2600 Avg. Training loss: 4.8564 0.1636 sec/batch
Epoch 1/10 Iteration: 2700 Avg. Training loss: 4.8693 0.1637 sec/batch
Epoch 1/10 Iteration: 2800 Avg. Training loss: 4.8441 0.1624 sec/batch
Epoch 1/10 Iteration: 2900 Avg. Training loss: 4.8319 0.1628 sec/batch
Epoch 1/10 Iteration: 3000 Avg. Training loss: 4.8231 0.1647 sec/batch
Nearest to or: terrain, mastered, trade, cubic, hacker, heron, armoured, afghani,
Nearest to than: case, taste, fortified, masses, extensively, undermine, pillars, taping,
Nearest to d: martinique, mwi, raffarin, u, composer, evelyn, malthus, lake,
Nearest to about: expected, round, dubbed, conditions, enforcement, triglyceride, cold, beauty,
Nearest to zero: shale, completion, pearl, tasman, ginger, russia, beth, austrians,
Nearest to five: shut, lawrence, bid, cappadocian, junior, carnivorous, earthly, elite,
Nearest to th: troubles, recommendation, defender, boulevard, hijacking, temper, newswire, fainted,
Nearest to more: funk, pt, capital, lott, ismaili, plunge, microscope, defenders,
Nearest to alternative: concisely, grammarians, genetics, isotope, intelligence, asean, cylinders, rgen,
Nearest to prince: pelts, believers, interchange, presuming, prd, ec, american, projecting,
Nearest to police: motif, doris, resistance, concludes, outweigh, sidereal, xxxv, lichfield,
Nearest to professional: students, melancholic, quo, oscillates, ugaritic, manifested, cumbersome, smale,
Nearest to construction: linseed, schoolmates, diophantus, bulky, technologically, bluegrass, barberini, commenting,
Nearest to http: reiterated, nicholls, disenchanted, topless, tangle, zoom, elm, christchurch,
Nearest to issue: hittites, frightened, haine, would, name, filled, making, victims,
Nearest to behind: sending, demonstrating, forefront, ceases, erica, sieur, scope, felicity,
Epoch 1/10 Iteration: 3100 Avg. Training loss: 4.8334 0.1654 sec/batch
Epoch 1/10 Iteration: 3200 Avg. Training loss: 4.8227 0.1662 sec/batch
Epoch 1/10 Iteration: 3300 Avg. Training loss: 4.7699 0.1675 sec/batch
Epoch 1/10 Iteration: 3400 Avg. Training loss: 4.7423 0.1655 sec/batch
Epoch 1/10 Iteration: 3500 Avg. Training loss: 4.7680 0.1656 sec/batch
Epoch 1/10 Iteration: 3600 Avg. Training loss: 4.7376 0.1650 sec/batch
Epoch 1/10 Iteration: 3700 Avg. Training loss: 4.7684 0.1662 sec/batch
Epoch 1/10 Iteration: 3800 Avg. Training loss: 4.7564 0.1653 sec/batch
Epoch 1/10 Iteration: 3900 Avg. Training loss: 4.7310 0.1655 sec/batch
Epoch 1/10 Iteration: 4000 Avg. Training loss: 4.6904 0.1654 sec/batch
Nearest to or: terrain, mastered, combatted, hacker, cubic, armoured, neverwinter, heron,
Nearest to than: case, taste, fortified, extensively, masses, taping, undermine, booty,
Nearest to d: composer, martinique, mwi, raffarin, evelyn, antoine, u, alighieri,
Nearest to about: expected, round, dubbed, conditions, enforcement, sq, triglyceride, cold,
Nearest to zero: shale, completion, pearl, tasman, austrians, beth, russia, ginger,
Nearest to five: lawrence, shut, junior, runciman, axel, bid, cappadocian, thelema,
Nearest to th: troubles, recommendation, boulevard, newswire, defender, hijacking, fainted, exaggeration,
Nearest to more: funk, lott, pt, plunge, defenders, ditko, capital, microscope,
Nearest to alternative: concisely, grammarians, genetics, intelligence, isotope, deprives, asean, ocs,
Nearest to prince: pelts, believers, american, prd, presuming, bout, interchange, iditarod,
Nearest to police: motif, resistance, doris, pde, lichfield, xxxv, concludes, outweigh,
Nearest to professional: students, melancholic, oscillates, quo, ugaritic, gx, grounding, smale,
Nearest to construction: linseed, schoolmates, diophantus, barberini, technologically, bulky, bluegrass, bitstream,
Nearest to http: reiterated, nicholls, zoom, tangle, topless, disenchanted, elm, nashville,
Nearest to issue: hittites, frightened, sackville, haine, mbox, incongruous, worshipers, would,
Nearest to behind: sending, ceases, demonstrating, forefront, sieur, erica, roaming, scope,
Epoch 1/10 Iteration: 4100 Avg. Training loss: 4.6906 0.1656 sec/batch
Epoch 1/10 Iteration: 4200 Avg. Training loss: 4.7067 0.1655 sec/batch
Epoch 1/10 Iteration: 4300 Avg. Training loss: 4.6458 0.1663 sec/batch
Epoch 1/10 Iteration: 4400 Avg. Training loss: 4.6478 0.1664 sec/batch
Epoch 1/10 Iteration: 4500 Avg. Training loss: 4.6397 0.1657 sec/batch
Epoch 1/10 Iteration: 4600 Avg. Training loss: 4.6579 0.1664 sec/batch
Epoch 2/10 Iteration: 4700 Avg. Training loss: 4.6257 0.1210 sec/batch
Epoch 2/10 Iteration: 4800 Avg. Training loss: 4.5422 0.1652 sec/batch
Epoch 2/10 Iteration: 4900 Avg. Training loss: 4.5418 0.1637 sec/batch
Epoch 2/10 Iteration: 5000 Avg. Training loss: 4.4879 0.1639 sec/batch
Nearest to or: terrain, mastered, combatted, hacker, armoured, cubic, affect, afghani,
Nearest to than: case, taste, masses, taping, extensively, fortified, booty, assistants,
Nearest to d: composer, antoine, martinique, evelyn, mwi, raffarin, biologist, british,
Nearest to about: expected, round, dubbed, conditions, triglyceride, sq, zawahiri, cold,
Nearest to zero: shale, completion, tasman, pearl, orozco, two, beth, invincibility,
Nearest to five: lawrence, hectare, runciman, reunion, junior, bid, axel, deaths,
Nearest to th: troubles, recommendation, boulevard, newswire, macrostate, exaggeration, fainted, hijacking,
Nearest to more: funk, lott, microscope, spreads, defenders, plunge, gonzales, capital,
Nearest to alternative: concisely, grammarians, genetics, intelligence, deprives, ocs, asean, belted,
Nearest to prince: iditarod, pelts, american, william, presuming, prd, believers, bestsellers,
Nearest to police: motif, resistance, pde, doris, reverence, bevel, lichfield, legal,
Nearest to professional: students, melancholic, oscillates, smale, gx, ugaritic, cumbersome, manifested,
Nearest to construction: linseed, schoolmates, diophantus, bulky, barberini, bluegrass, bitstream, draining,
Nearest to http: zoom, nicholls, reiterated, org, tangle, elm, topless, hilltop,
Nearest to issue: hittites, frightened, sackville, would, haine, mbox, incongruous, worshipers,
Nearest to behind: sending, ceases, demonstrating, sieur, forefront, erica, roaming, felicity,
Epoch 2/10 Iteration: 5100 Avg. Training loss: 4.5004 0.1649 sec/batch
Epoch 2/10 Iteration: 5200 Avg. Training loss: 4.4737 0.1652 sec/batch
Epoch 2/10 Iteration: 5300 Avg. Training loss: 4.4453 0.1648 sec/batch
Epoch 2/10 Iteration: 5400 Avg. Training loss: 4.5270 0.1641 sec/batch
Epoch 2/10 Iteration: 5500 Avg. Training loss: 4.4642 0.1637 sec/batch
Epoch 2/10 Iteration: 5600 Avg. Training loss: 4.4795 0.1674 sec/batch
Epoch 2/10 Iteration: 5700 Avg. Training loss: 4.4480 0.1653 sec/batch
Epoch 2/10 Iteration: 5800 Avg. Training loss: 4.3978 0.1670 sec/batch
Epoch 2/10 Iteration: 5900 Avg. Training loss: 4.4072 0.1687 sec/batch
Epoch 2/10 Iteration: 6000 Avg. Training loss: 4.4158 0.1681 sec/batch
Nearest to or: terrain, affect, combatted, mastered, thread, hacker, armoured, afghani,
Nearest to than: case, taste, masses, taping, grazing, fortified, earner, extensively,
Nearest to d: composer, raffarin, martinique, antoine, mwi, evelyn, biologist, herndon,
Nearest to about: expected, round, dubbed, conditions, sq, triglyceride, cold, zawahiri,
Nearest to zero: tasman, pearl, shale, completion, two, agricultural, six, invincibility,
Nearest to five: lawrence, hectare, reunion, runciman, whist, trucks, bid, junior,
Nearest to th: troubles, boulevard, recommendation, newswire, fainted, macrostate, exaggeration, tagging,
Nearest to more: funk, lott, nodes, spot, eml, plunge, microscope, vulcan,
Nearest to alternative: concisely, grammarians, genetics, intelligence, deprives, belted, ocs, isotope,
Nearest to prince: pelts, iditarod, presuming, mehmed, bout, salian, prd, tricked,
Nearest to police: resistance, pde, legal, motif, pastrana, bevel, doris, lichfield,
Nearest to professional: students, smale, melancholic, oscillates, gx, manifested, ugaritic, steaming,
Nearest to construction: linseed, schoolmates, diophantus, bulky, draining, technologically, barberini, bitstream,
Nearest to http: org, nicholls, zoom, reiterated, tangle, elm, hilltop, nashville,
Nearest to issue: hittites, sackville, would, frightened, primigenius, haine, hasdai, worshipers,
Nearest to behind: sending, ceases, sieur, demonstrating, forefront, erica, felicity, breech,
Epoch 2/10 Iteration: 6100 Avg. Training loss: 4.4217 0.1672 sec/batch
Epoch 2/10 Iteration: 6200 Avg. Training loss: 4.4199 0.1668 sec/batch
Epoch 2/10 Iteration: 6300 Avg. Training loss: 4.4504 0.1678 sec/batch
Epoch 2/10 Iteration: 6400 Avg. Training loss: 4.3734 0.1674 sec/batch
Epoch 2/10 Iteration: 6500 Avg. Training loss: 4.4253 0.1663 sec/batch
Epoch 2/10 Iteration: 6600 Avg. Training loss: 4.4455 0.1662 sec/batch
Epoch 2/10 Iteration: 6700 Avg. Training loss: 4.3729 0.1674 sec/batch
Epoch 2/10 Iteration: 6800 Avg. Training loss: 4.3579 0.1655 sec/batch
Epoch 2/10 Iteration: 6900 Avg. Training loss: 4.3906 0.1660 sec/batch
Epoch 2/10 Iteration: 7000 Avg. Training loss: 4.3643 0.1669 sec/batch
Nearest to or: mastered, their, terrain, combatted, thread, affect, afghani, heron,
Nearest to than: case, taste, masses, taping, fortified, candidates, grazing, extensively,
Nearest to d: composer, raffarin, antoine, mwi, evelyn, martinique, geneticist, slovenian,
Nearest to about: expected, dubbed, round, conditions, triglyceride, democide, zawahiri, cold,
Nearest to zero: two, eight, shale, pearl, six, tasman, completion, agricultural,
Nearest to five: hectare, lawrence, marquess, runciman, whist, reunion, three, seater,
Nearest to th: boulevard, troubles, fainted, newswire, recommendation, macrostate, dolly, exaggeration,
Nearest to more: funk, nodes, plunge, lott, eml, spreads, pt, gakuin,
Nearest to alternative: concisely, grammarians, genetics, intelligence, deprives, belted, perineum, cognition,
Nearest to prince: iditarod, pelts, bestsellers, reigning, mehmed, tricked, bukhari, salian,
Nearest to police: resistance, legal, pde, pastrana, bevel, referendum, doris, patrolled,
Nearest to professional: students, smale, gx, melancholic, oscillates, steaming, fas, rec,
Nearest to construction: linseed, bulky, schoolmates, diophantus, draining, barberini, medway, technologically,
Nearest to http: org, nicholls, zoom, edu, reiterated, html, tangle, https,
Nearest to issue: hittites, would, hasdai, worshipers, ftc, haine, primigenius, hindus,
Nearest to behind: sending, ceases, demonstrating, breech, roaming, sieur, felicity, cobra,
Epoch 2/10 Iteration: 7100 Avg. Training loss: 4.3615 0.1665 sec/batch
Epoch 2/10 Iteration: 7200 Avg. Training loss: 4.3963 0.1654 sec/batch
Epoch 2/10 Iteration: 7300 Avg. Training loss: 4.3590 0.1655 sec/batch
Epoch 2/10 Iteration: 7400 Avg. Training loss: 4.3842 0.1639 sec/batch
Epoch 2/10 Iteration: 7500 Avg. Training loss: 4.4091 0.1642 sec/batch
Epoch 2/10 Iteration: 7600 Avg. Training loss: 4.3692 0.1670 sec/batch
Epoch 2/10 Iteration: 7700 Avg. Training loss: 4.3410 0.1645 sec/batch
Epoch 2/10 Iteration: 7800 Avg. Training loss: 4.3443 0.1682 sec/batch
Epoch 2/10 Iteration: 7900 Avg. Training loss: 4.3055 0.1682 sec/batch
Epoch 2/10 Iteration: 8000 Avg. Training loss: 4.3199 0.1657 sec/batch
Nearest to or: terrain, mastered, thread, combatted, heron, their, candles, traits,
Nearest to than: case, taste, fortified, masses, taping, grazing, amon, candidates,
Nearest to d: composer, raffarin, antoine, mwi, h, martinique, bundeswehr, dazzling,
Nearest to about: expected, dubbed, democide, conditions, round, sq, cold, reston,
Nearest to zero: two, eight, six, nine, three, pearl, invincibility, shale,
Nearest to five: three, hectare, lawrence, scepter, whist, marquess, runciman, expressed,
Nearest to th: boulevard, troubles, fainted, macrostate, bearded, monastic, recommendation, newswire,
Nearest to more: funk, lott, plunge, satellites, gonzales, capital, gakuin, nodes,
Nearest to alternative: concisely, grammarians, genetics, belted, intelligence, deprives, isotope, rgen,
Nearest to prince: salian, pelts, iditarod, bestsellers, reigning, mehmed, quell, bukhari,
Nearest to police: resistance, legal, pastrana, patrolled, pde, bevel, referendum, doris,
Nearest to professional: students, smale, rec, gx, fas, melancholic, oscillates, klansmen,
Nearest to construction: linseed, bulky, schoolmates, diophantus, draining, barberini, medway, gerrymander,
Nearest to http: org, edu, html, nicholls, reiterated, zoom, com, https,
Nearest to issue: hittites, would, ftc, worshipers, hasdai, hindus, victims, sackville,
Nearest to behind: sending, ceases, breech, demonstrating, roaming, sieur, forefront, felicity,
Epoch 2/10 Iteration: 8100 Avg. Training loss: 4.3289 0.1675 sec/batch
Epoch 2/10 Iteration: 8200 Avg. Training loss: 4.2894 0.1668 sec/batch
Epoch 2/10 Iteration: 8300 Avg. Training loss: 4.3812 0.1672 sec/batch
Epoch 2/10 Iteration: 8400 Avg. Training loss: 4.3637 0.1658 sec/batch
Epoch 2/10 Iteration: 8500 Avg. Training loss: 4.4112 0.1656 sec/batch
Epoch 2/10 Iteration: 8600 Avg. Training loss: 4.3069 0.1665 sec/batch
Epoch 2/10 Iteration: 8700 Avg. Training loss: 4.2785 0.1649 sec/batch
Epoch 2/10 Iteration: 8800 Avg. Training loss: 4.3040 0.1642 sec/batch
Epoch 2/10 Iteration: 8900 Avg. Training loss: 4.2047 0.1703 sec/batch
Epoch 2/10 Iteration: 9000 Avg. Training loss: 4.2878 0.1663 sec/batch
Nearest to or: terrain, mastered, thread, combatted, heron, their, sympathisers, traits,
Nearest to than: masses, taste, case, grazing, fortified, amon, taping, monos,
Nearest to d: composer, antoine, raffarin, slovenian, geneticist, biologist, martinique, u,
Nearest to about: expected, dubbed, sq, democide, conditions, showering, triglyceride, reston,
Nearest to zero: two, six, eight, nine, est, one, rate, five,
Nearest to five: zero, three, six, seven, four, marquess, lawrence, hectare,
Nearest to th: boulevard, macrostate, fainted, kupa, bearded, troubles, newswire, implements,
Nearest to more: funk, capital, satellites, nodes, gonzales, eml, lott, plunge,
Nearest to alternative: concisely, grammarians, genetics, intelligence, belted, deprives, seaborg, creel,
Nearest to prince: pelts, salian, iditarod, reigning, surratt, mehmed, quell, bukhari,
Nearest to police: resistance, patrolled, pastrana, bevel, legal, pde, legality, doris,
Nearest to professional: students, klansmen, rec, melancholic, smale, fas, oncology, gx,
Nearest to construction: linseed, bulky, schoolmates, diophantus, medway, draining, barberini, gerrymander,
Nearest to http: org, edu, html, htm, nicholls, www, com, https,
Nearest to issue: hittites, ftc, sackville, would, worshipers, hasdai, primigenius, greens,
Nearest to behind: sending, ceases, breech, demonstrating, erica, roaming, felicity, forefront,
Epoch 2/10 Iteration: 9100 Avg. Training loss: 4.2884 0.1664 sec/batch
Epoch 2/10 Iteration: 9200 Avg. Training loss: 4.3003 0.1654 sec/batch
Epoch 3/10 Iteration: 9300 Avg. Training loss: 4.3148 0.0767 sec/batch
Epoch 3/10 Iteration: 9400 Avg. Training loss: 4.2662 0.1656 sec/batch
Epoch 3/10 Iteration: 9500 Avg. Training loss: 4.1981 0.1651 sec/batch
Epoch 3/10 Iteration: 9600 Avg. Training loss: 4.2021 0.1653 sec/batch
Epoch 3/10 Iteration: 9700 Avg. Training loss: 4.2070 0.1639 sec/batch
Epoch 3/10 Iteration: 9800 Avg. Training loss: 4.1912 0.1636 sec/batch
Epoch 3/10 Iteration: 9900 Avg. Training loss: 4.1921 0.1658 sec/batch
Epoch 3/10 Iteration: 10000 Avg. Training loss: 4.1366 0.1653 sec/batch
Nearest to or: terrain, mastered, heron, combatted, candles, footpaths, thread, sympathisers,
Nearest to than: masses, taste, fortified, case, amon, grazing, monos, taping,
Nearest to d: composer, raffarin, antoine, geneticist, slovenian, martinique, biologist, evelyn,
Nearest to about: expected, sq, conditions, democide, waters, showering, dubbed, triglyceride,
Nearest to zero: two, six, eight, nine, three, one, five, est,
Nearest to five: three, zero, hectare, seven, four, six, marquess, birth,
Nearest to th: boulevard, macrostate, kupa, fainted, newswire, implements, bearded, tagging,
Nearest to more: capital, satellites, funk, eml, lott, nodes, gakuin, gonzales,
Nearest to alternative: genetics, concisely, grammarians, intelligence, deprives, cognition, belted, bjorn,
Nearest to prince: salian, iditarod, pelts, bukhari, surratt, mehmed, reigning, alcibiades,
Nearest to police: patrolled, pastrana, resistance, bevel, legality, pde, legal, substandard,
Nearest to professional: students, rec, gx, smale, melancholic, klansmen, oncology, cleaveland,
Nearest to construction: linseed, schoolmates, bulky, barberini, medway, diophantus, draining, road,
Nearest to http: org, html, edu, htm, nicholls, https, www, com,
Nearest to issue: hittites, ftc, would, sackville, hasdai, amount, worshipers, nausica,
Nearest to behind: ceases, sending, breech, roaming, demonstrating, erica, felicity, berchtesgaden,
Epoch 3/10 Iteration: 10100 Avg. Training loss: 4.2462 0.1658 sec/batch
Epoch 3/10 Iteration: 10200 Avg. Training loss: 4.1824 0.1668 sec/batch
Epoch 3/10 Iteration: 10300 Avg. Training loss: 4.2007 0.1662 sec/batch
Epoch 3/10 Iteration: 10400 Avg. Training loss: 4.0900 0.1692 sec/batch
Epoch 3/10 Iteration: 10500 Avg. Training loss: 4.1620 0.1687 sec/batch
Epoch 3/10 Iteration: 10600 Avg. Training loss: 4.1261 0.1669 sec/batch
Epoch 3/10 Iteration: 10700 Avg. Training loss: 4.1527 0.1680 sec/batch
Epoch 3/10 Iteration: 10800 Avg. Training loss: 4.1842 0.1669 sec/batch
Epoch 3/10 Iteration: 10900 Avg. Training loss: 4.1695 0.1661 sec/batch
Epoch 3/10 Iteration: 11000 Avg. Training loss: 4.1361 0.1661 sec/batch
Nearest to or: mastered, candles, hacker, heron, cepheus, thread, dressers, footpaths,
Nearest to than: amon, case, masses, taping, monos, fortified, taste, desqview,
Nearest to d: raffarin, composer, martinique, slovenian, antoine, u, h, mwi,
Nearest to about: expected, sq, democide, reston, conditions, triglyceride, showering, dubbed,
Nearest to zero: two, six, eight, nine, one, five, three, est,
Nearest to five: three, zero, four, hectare, seven, six, birth, expressed,
Nearest to th: boulevard, fainted, macrostate, tagging, newswire, inches, kupa, implements,
Nearest to more: funk, eml, nodes, capital, manner, sputtering, spreads, rephlex,
Nearest to alternative: concisely, grammarians, genetics, intelligence, deprives, rgen, belted, ecclesiastics,
Nearest to prince: pelts, salian, iditarod, mehmed, surratt, reigning, bout, defenseless,
Nearest to police: pastrana, patrolled, resistance, bevel, legality, legal, pde, substandard,
Nearest to professional: students, rec, oncology, melancholic, smale, soccer, cleaveland, klansmen,
Nearest to construction: linseed, schoolmates, bulky, barberini, medway, technologically, diophantus, fergus,
Nearest to http: org, html, edu, www, htm, https, reiterated, com,
Nearest to issue: ftc, hittites, hasdai, worshipers, sackville, amount, haine, would,
Nearest to behind: ceases, breech, sending, roaming, demonstrating, erica, ride, felicity,
Epoch 3/10 Iteration: 11100 Avg. Training loss: 4.1523 0.1678 sec/batch
Epoch 3/10 Iteration: 11200 Avg. Training loss: 4.2139 0.1653 sec/batch
Epoch 3/10 Iteration: 11300 Avg. Training loss: 4.1557 0.1642 sec/batch
Epoch 3/10 Iteration: 11400 Avg. Training loss: 4.1228 0.1654 sec/batch
Epoch 3/10 Iteration: 11500 Avg. Training loss: 4.1846 0.1650 sec/batch
Epoch 3/10 Iteration: 11600 Avg. Training loss: 4.1762 0.1681 sec/batch
Epoch 3/10 Iteration: 11700 Avg. Training loss: 4.1709 0.1647 sec/batch
Epoch 3/10 Iteration: 11800 Avg. Training loss: 4.1376 0.1662 sec/batch
Epoch 3/10 Iteration: 11900 Avg. Training loss: 4.1291 0.1650 sec/batch
Epoch 3/10 Iteration: 12000 Avg. Training loss: 4.1700 0.1665 sec/batch
Nearest to or: candles, cepheus, waxy, mastered, combatted, heron, thread, footpaths,
Nearest to than: masses, amon, case, taping, fortified, monos, booty, contracting,
Nearest to d: raffarin, composer, antoine, slovenian, geneticist, variste, u, martinique,
Nearest to about: expected, democide, dubbed, showering, reston, conditions, sq, gd,
Nearest to zero: two, six, eight, nine, three, one, five, est,
Nearest to five: three, zero, four, six, seven, hectare, two, birth,
Nearest to th: fainted, boulevard, newswire, kupa, tagging, bearded, century, cadmus,
Nearest to more: funk, capital, satellites, eml, manner, rephlex, plunge, spreads,
Nearest to alternative: concisely, grammarians, genetics, intelligence, rgen, deprives, belted, asean,
Nearest to prince: salian, pelts, reigning, iditarod, mehmed, surratt, bukhari, alcibiades,
Nearest to police: patrolled, pastrana, resistance, bevel, legality, pde, substandard, legal,
Nearest to professional: students, rec, oncology, melancholic, smale, soccer, fas, klansmen,
Nearest to construction: linseed, bulky, barberini, schoolmates, medway, fergus, booming, technologically,
Nearest to http: org, html, edu, www, htm, https, com, reiterated,
Nearest to issue: ftc, hittites, would, worshipers, sackville, hasdai, primigenius, nausica,
Nearest to behind: ceases, breech, sending, travels, felicity, ride, roaming, erica,
Epoch 3/10 Iteration: 12100 Avg. Training loss: 4.1744 0.1676 sec/batch
Epoch 3/10 Iteration: 12200 Avg. Training loss: 4.1539 0.1656 sec/batch
Epoch 3/10 Iteration: 12300 Avg. Training loss: 4.1631 0.1645 sec/batch
Epoch 3/10 Iteration: 12400 Avg. Training loss: 4.1871 0.1665 sec/batch
Epoch 3/10 Iteration: 12500 Avg. Training loss: 4.1239 0.1656 sec/batch
Epoch 3/10 Iteration: 12600 Avg. Training loss: 4.1187 0.1661 sec/batch
Epoch 3/10 Iteration: 12700 Avg. Training loss: 4.1576 0.1653 sec/batch
Epoch 3/10 Iteration: 12800 Avg. Training loss: 4.1125 0.1660 sec/batch
Epoch 3/10 Iteration: 12900 Avg. Training loss: 4.1847 0.1674 sec/batch
Epoch 3/10 Iteration: 13000 Avg. Training loss: 4.2080 0.1663 sec/batch
Nearest to or: candles, cepheus, sympathisers, heron, mastered, dressers, thread, hacker,
Nearest to than: amon, masses, case, monos, fortified, karabakh, contracting, taping,
Nearest to d: raffarin, composer, antoine, geneticist, slovenian, martinique, biologist, u,
Nearest to about: expected, democide, sq, dubbed, monti, cold, showering, reston,
Nearest to zero: two, six, eight, nine, one, three, five, four,
Nearest to five: three, zero, four, six, seven, two, one, marquess,
Nearest to th: century, troubles, fainted, kupa, bearded, macrostate, boulevard, deportivo,
Nearest to more: funk, capital, satellites, eml, rephlex, gonzales, manner, karamchand,
Nearest to alternative: concisely, grammarians, asean, intelligence, rgen, genetics, deprives, streetcar,
Nearest to prince: salian, pelts, reigning, iditarod, surratt, surmise, mehmed, bout,
Nearest to police: patrolled, resistance, pastrana, bevel, legality, pde, substandard, iasi,
Nearest to professional: students, rec, melancholic, soccer, smale, cleaveland, oncology, klansmen,
Nearest to construction: linseed, schoolmates, barberini, medway, bulky, fergus, diophantus, booming,
Nearest to http: org, html, htm, edu, www, com, https, reiterated,
Nearest to issue: ftc, hasdai, would, sackville, hittites, primigenius, examines, nausica,
Nearest to behind: breech, sending, ceases, travels, ride, roaming, demonstrating, felicity,
Epoch 3/10 Iteration: 13100 Avg. Training loss: 4.2094 0.1667 sec/batch
Epoch 3/10 Iteration: 13200 Avg. Training loss: 4.1201 0.1647 sec/batch
Epoch 3/10 Iteration: 13300 Avg. Training loss: 4.1336 0.1661 sec/batch
Epoch 3/10 Iteration: 13400 Avg. Training loss: 4.1386 0.1648 sec/batch
Epoch 3/10 Iteration: 13500 Avg. Training loss: 4.0279 0.1646 sec/batch
Epoch 3/10 Iteration: 13600 Avg. Training loss: 4.1525 0.1664 sec/batch
Epoch 3/10 Iteration: 13700 Avg. Training loss: 4.1113 0.1649 sec/batch
Epoch 3/10 Iteration: 13800 Avg. Training loss: 4.1229 0.1640 sec/batch
Epoch 4/10 Iteration: 13900 Avg. Training loss: 4.2066 0.0317 sec/batch
Epoch 4/10 Iteration: 14000 Avg. Training loss: 4.1026 0.1645 sec/batch
Nearest to or: candles, sympathisers, waxy, cepheus, heron, thread, mastered, footpaths,
Nearest to than: masses, case, amon, desqview, monos, amending, contracting, adjusted,
Nearest to d: raffarin, composer, antoine, geneticist, slovenian, martinique, variste, u,
Nearest to about: expected, sq, democide, waters, showering, dubbed, monti, gd,
Nearest to zero: two, six, eight, nine, one, five, three, four,
Nearest to five: zero, three, four, two, six, seven, one, hectare,
Nearest to th: century, pentameter, nouakchott, fainted, boulevard, macrostate, cadmus, deportivo,
Nearest to more: capital, satellites, funk, eml, rephlex, unemployment, nodes, manner,
Nearest to alternative: concisely, grammarians, asean, intelligence, deprives, genetics, streetcar, rgen,
Nearest to prince: salian, pelts, iditarod, reigning, surratt, mehmed, surmise, bukhari,
Nearest to police: patrolled, resistance, pastrana, bevel, legality, strangle, substandard, adultery,
Nearest to professional: students, rec, soccer, manifested, oncology, smale, melancholic, cleaveland,
Nearest to construction: linseed, schoolmates, bulky, medway, barberini, fergus, booming, diophantus,
Nearest to http: org, html, htm, www, edu, com, https, nicholls,
Nearest to issue: ftc, hasdai, sackville, hittites, greens, would, kadena, primigenius,
Nearest to behind: breech, sending, ceases, travels, demonstrating, felicity, ride, roaming,
Epoch 4/10 Iteration: 14100 Avg. Training loss: 4.0532 0.1672 sec/batch
Epoch 4/10 Iteration: 14200 Avg. Training loss: 4.0855 0.1643 sec/batch
Epoch 4/10 Iteration: 14300 Avg. Training loss: 4.0696 0.1625 sec/batch
Epoch 4/10 Iteration: 14400 Avg. Training loss: 4.0485 0.1642 sec/batch
Epoch 4/10 Iteration: 14500 Avg. Training loss: 4.0751 0.1658 sec/batch
Epoch 4/10 Iteration: 14600 Avg. Training loss: 4.0239 0.1640 sec/batch
Epoch 4/10 Iteration: 14700 Avg. Training loss: 4.0627 0.1652 sec/batch
Epoch 4/10 Iteration: 14800 Avg. Training loss: 4.0496 0.1653 sec/batch
Epoch 4/10 Iteration: 14900 Avg. Training loss: 4.1154 0.1678 sec/batch
Epoch 4/10 Iteration: 15000 Avg. Training loss: 3.9827 0.1670 sec/batch
Nearest to or: waxy, candles, footpaths, heron, mastered, summoner, thread, cepheus,
Nearest to than: masses, case, amon, amending, monos, desqview, grazing, karabakh,
Nearest to d: raffarin, composer, variste, martinique, geneticist, antoine, slovenian, cardiology,
Nearest to about: sq, expected, waters, gd, democide, showering, round, rahula,
Nearest to zero: two, six, eight, nine, one, five, three, four,
Nearest to five: three, zero, four, two, six, seven, one, nine,
Nearest to th: century, pentameter, tagging, nd, troubles, nouakchott, fainted, kupa,
Nearest to more: eml, capital, funk, ability, rephlex, vulcan, ditko, satellites,
Nearest to alternative: grammarians, concisely, genetics, asean, intelligence, deprives, rgen, bjorn,
Nearest to prince: pelts, salian, reigning, iditarod, surratt, surmise, alcibiades, vcr,
Nearest to police: patrolled, resistance, pastrana, bevel, legality, strangle, substandard, pde,
Nearest to professional: students, rec, soccer, manifested, smale, oncology, cleaveland, melancholic,
Nearest to construction: linseed, bulky, medway, schoolmates, barberini, fergus, saratoga, laplace,
Nearest to http: org, html, www, htm, edu, https, com, nicholls,
Nearest to issue: ftc, hasdai, hittites, sackville, greens, primigenius, examines, kadena,
Nearest to behind: breech, sending, ceases, travels, demonstrating, felicity, ride, roaming,
Epoch 4/10 Iteration: 15100 Avg. Training loss: 4.0192 0.1701 sec/batch
Epoch 4/10 Iteration: 15200 Avg. Training loss: 4.0333 0.1693 sec/batch
Epoch 4/10 Iteration: 15300 Avg. Training loss: 3.9777 0.1698 sec/batch
Epoch 4/10 Iteration: 15400 Avg. Training loss: 4.0340 0.1669 sec/batch
Epoch 4/10 Iteration: 15500 Avg. Training loss: 4.0524 0.1680 sec/batch
Epoch 4/10 Iteration: 15600 Avg. Training loss: 4.0325 0.1678 sec/batch
Epoch 4/10 Iteration: 15700 Avg. Training loss: 4.0630 0.1661 sec/batch
Epoch 4/10 Iteration: 15800 Avg. Training loss: 4.0939 0.1646 sec/batch
Epoch 4/10 Iteration: 15900 Avg. Training loss: 4.0735 0.1661 sec/batch
Epoch 4/10 Iteration: 16000 Avg. Training loss: 4.0252 0.1661 sec/batch
Nearest to or: waxy, candles, are, cepheus, thread, footpaths, heron, minorities,
Nearest to than: masses, amon, case, desqview, amending, grazing, monos, taping,
Nearest to d: raffarin, variste, composer, geneticist, martinique, u, slovenian, cardiology,
Nearest to about: sq, democide, expected, gd, showering, round, reston, rahula,
Nearest to zero: two, six, eight, three, five, one, nine, four,
Nearest to five: three, zero, four, six, two, one, seven, nine,
Nearest to th: century, nd, pentameter, nouakchott, rd, deportivo, pawe, cadmus,
Nearest to more: capital, eml, ability, rephlex, funk, tagging, manner, nodes,
Nearest to alternative: grammarians, concisely, asean, rgen, genetics, intelligence, deprives, scarcity,
Nearest to prince: salian, pelts, reigning, iditarod, surratt, vcr, starred, surmise,
Nearest to police: patrolled, resistance, pastrana, bevel, legality, pde, strangle, postpartum,
Nearest to professional: students, rec, manifested, soccer, oncology, smale, cleaveland, aircrew,
Nearest to construction: linseed, bulky, barberini, schoolmates, medway, bab, fergus, laplace,
Nearest to http: org, html, www, htm, edu, https, com, nicholls,
Nearest to issue: ftc, hasdai, examines, hittites, would, sackville, greens, filled,
Nearest to behind: breech, travels, ceases, sending, ride, felicity, demonstrating, berchtesgaden,
Epoch 4/10 Iteration: 16100 Avg. Training loss: 4.0414 0.1669 sec/batch
Epoch 4/10 Iteration: 16200 Avg. Training loss: 4.0277 0.1640 sec/batch
Epoch 4/10 Iteration: 16300 Avg. Training loss: 4.0512 0.1643 sec/batch
Epoch 4/10 Iteration: 16400 Avg. Training loss: 4.0113 0.1640 sec/batch
Epoch 4/10 Iteration: 16500 Avg. Training loss: 4.0605 0.1636 sec/batch
Epoch 4/10 Iteration: 16600 Avg. Training loss: 4.0548 0.1666 sec/batch
Epoch 4/10 Iteration: 16700 Avg. Training loss: 4.0399 0.1646 sec/batch
Epoch 4/10 Iteration: 16800 Avg. Training loss: 4.0446 0.1655 sec/batch
Epoch 4/10 Iteration: 16900 Avg. Training loss: 4.0548 0.1645 sec/batch
Epoch 4/10 Iteration: 17000 Avg. Training loss: 4.0534 0.1647 sec/batch
Nearest to or: candles, waxy, are, footpaths, cepheus, thread, heron, praecox,
Nearest to than: case, masses, amon, grazing, taping, karabakh, amending, desqview,
Nearest to d: raffarin, composer, variste, martinique, fibonacci, geneticist, antoine, cardiology,
Nearest to about: democide, sq, round, expected, showering, gd, reston, rahula,
Nearest to zero: two, six, eight, nine, three, five, one, four,
Nearest to five: three, zero, four, six, two, seven, one, nine,
Nearest to th: century, pentameter, nd, nouakchott, rd, andhras, monastic, troubles,
Nearest to more: capital, tagging, rephlex, funk, sway, manner, ability, plunge,
Nearest to alternative: grammarians, asean, rgen, deprives, genetics, concisely, streetcar, intelligence,
Nearest to prince: salian, reigning, pelts, surmise, surratt, iditarod, conventual, lorraine,
Nearest to police: patrolled, resistance, pastrana, bevel, pde, paramilitary, strangle, legality,
Nearest to professional: students, rec, manifested, smale, soccer, oncology, cleaveland, quo,
Nearest to construction: bulky, medway, linseed, schoolmates, barberini, bab, fergus, booming,
Nearest to http: html, org, htm, www, edu, com, reiterated, nicholls,
Nearest to issue: ftc, hasdai, hittites, examines, would, sackville, newsgroup, kadena,
Nearest to behind: breech, ceases, travels, sending, felicity, demonstrating, ride, erica,
Epoch 4/10 Iteration: 17100 Avg. Training loss: 4.0394 0.1682 sec/batch
Epoch 4/10 Iteration: 17200 Avg. Training loss: 4.0128 0.1672 sec/batch
Epoch 4/10 Iteration: 17300 Avg. Training loss: 3.9895 0.1661 sec/batch
Epoch 4/10 Iteration: 17400 Avg. Training loss: 4.0679 0.1672 sec/batch
Epoch 4/10 Iteration: 17500 Avg. Training loss: 4.0597 0.1664 sec/batch
Epoch 4/10 Iteration: 17600 Avg. Training loss: 4.1177 0.1681 sec/batch
Epoch 4/10 Iteration: 17700 Avg. Training loss: 4.1148 0.1658 sec/batch
Epoch 4/10 Iteration: 17800 Avg. Training loss: 4.0549 0.1657 sec/batch
Epoch 4/10 Iteration: 17900 Avg. Training loss: 4.0219 0.1652 sec/batch
Epoch 4/10 Iteration: 18000 Avg. Training loss: 4.0268 0.1654 sec/batch
Nearest to or: waxy, candles, are, cepheus, is, thread, footpaths, heron,
Nearest to than: masses, amon, case, karabakh, amending, grazing, adjusted, candidates,
Nearest to d: raffarin, composer, variste, slovenian, antoine, cardiology, geneticist, martinique,
Nearest to about: democide, sq, expected, reston, round, showering, gd, rahula,
Nearest to zero: two, six, eight, three, nine, five, one, four,
Nearest to five: zero, three, four, two, six, seven, one, nine,
Nearest to th: century, pentameter, nouakchott, andhras, monastic, nd, vishnu, rd,
Nearest to more: rephlex, tagging, funk, ability, eml, capital, clary, manner,
Nearest to alternative: grammarians, asean, rgen, genetics, streetcar, concisely, deprives, intelligence,
Nearest to prince: salian, pelts, reigning, surratt, surmise, iditarod, quell, bout,
Nearest to police: patrolled, resistance, pastrana, bevel, strangle, legality, comanche, paramilitary,
Nearest to professional: students, rec, manifested, oncology, smale, soccer, melancholic, quo,
Nearest to construction: medway, linseed, bulky, barberini, schoolmates, fergus, bab, booming,
Nearest to http: html, htm, www, org, edu, com, https, reiterated,
Nearest to issue: ftc, examines, hittites, hasdai, would, newsgroup, sackville, carangi,
Nearest to behind: breech, ceases, travels, felicity, demonstrating, sending, erica, ride,
Epoch 4/10 Iteration: 18100 Avg. Training loss: 3.9534 0.1689 sec/batch
Epoch 4/10 Iteration: 18200 Avg. Training loss: 4.0143 0.1676 sec/batch
Epoch 4/10 Iteration: 18300 Avg. Training loss: 4.0435 0.1677 sec/batch
Epoch 4/10 Iteration: 18400 Avg. Training loss: 4.0052 0.1648 sec/batch
Epoch 4/10 Iteration: 18500 Avg. Training loss: 4.0754 0.1654 sec/batch
Epoch 5/10 Iteration: 18600 Avg. Training loss: 4.0234 0.1522 sec/batch
Epoch 5/10 Iteration: 18700 Avg. Training loss: 3.9865 0.1658 sec/batch
Epoch 5/10 Iteration: 18800 Avg. Training loss: 3.9661 0.1646 sec/batch
Epoch 5/10 Iteration: 18900 Avg. Training loss: 3.9981 0.1640 sec/batch
Epoch 5/10 Iteration: 19000 Avg. Training loss: 3.9626 0.1636 sec/batch
Nearest to or: candles, waxy, are, cepheus, is, footpaths, sympathisers, praecox,
Nearest to than: masses, amon, case, karabakh, amending, monos, grazing, percent,
Nearest to d: composer, raffarin, variste, geneticist, slovenian, cardiology, antoine, martinique,
Nearest to about: sq, democide, reston, round, expected, vicious, gd, waters,
Nearest to zero: two, six, eight, three, nine, five, one, four,
Nearest to five: three, four, zero, two, six, one, seven, nine,
Nearest to th: century, nd, pentameter, rd, andhras, monastic, nouakchott, ghilzai,
Nearest to more: ability, tagging, frequent, rephlex, eml, funk, capital, satellites,
Nearest to alternative: grammarians, asean, genetics, deprives, streetcar, concisely, intelligence, scarcity,
Nearest to prince: salian, reigning, pelts, surratt, surmise, quell, conventual, iditarod,
Nearest to police: patrolled, resistance, pastrana, bevel, strangle, paramilitary, legality, concealed,
Nearest to professional: rec, students, soccer, manifested, oncology, smale, kay, wrestling,
Nearest to construction: bulky, medway, linseed, bab, fergus, barberini, schoolmates, laplace,
Nearest to http: htm, html, www, org, edu, com, https, reiterated,
Nearest to issue: hasdai, ftc, examines, hittites, carangi, newsgroup, amount, kadena,
Nearest to behind: breech, ceases, travels, felicity, sending, demonstrating, maximizing, erica,
Epoch 5/10 Iteration: 19100 Avg. Training loss: 3.9709 0.1648 sec/batch
Epoch 5/10 Iteration: 19200 Avg. Training loss: 3.9071 0.1652 sec/batch
Epoch 5/10 Iteration: 19300 Avg. Training loss: 4.0014 0.1660 sec/batch
Epoch 5/10 Iteration: 19400 Avg. Training loss: 3.9900 0.1664 sec/batch
Epoch 5/10 Iteration: 19500 Avg. Training loss: 3.9945 0.1657 sec/batch
Epoch 5/10 Iteration: 19600 Avg. Training loss: 3.9980 0.1688 sec/batch
Epoch 5/10 Iteration: 19700 Avg. Training loss: 3.8867 0.1673 sec/batch
Epoch 5/10 Iteration: 19800 Avg. Training loss: 3.9488 0.1684 sec/batch
Epoch 5/10 Iteration: 19900 Avg. Training loss: 3.9010 0.1682 sec/batch
Epoch 5/10 Iteration: 20000 Avg. Training loss: 3.9439 0.1675 sec/batch
Nearest to or: are, waxy, candles, is, footpaths, cepheus, pocketing, heron,
Nearest to than: case, masses, amon, amending, karabakh, monos, grazing, five,
Nearest to d: raffarin, variste, composer, martinique, cardiology, slovenian, mwi, geneticist,
Nearest to about: sq, round, reston, waters, rahula, gd, democide, vicious,
Nearest to zero: two, six, five, nine, three, eight, four, one,
Nearest to five: three, four, zero, two, six, seven, one, nine,
Nearest to th: century, nd, rd, pentameter, deportivo, ghilzai, tagging, divisional,
Nearest to more: tagging, funk, frequent, ability, rephlex, eml, capital, manner,
Nearest to alternative: grammarians, asean, genetics, streetcar, deprives, concisely, rgen, albums,
Nearest to prince: salian, pelts, reigning, surratt, surmise, vcr, conventual, iditarod,
Nearest to police: patrolled, resistance, pastrana, bevel, paramilitary, strangle, legality, pde,
Nearest to professional: students, soccer, manifested, rec, oncology, gaining, cleaveland, wrestling,
Nearest to construction: bulky, bab, schoolmates, barberini, medway, linseed, fergus, conservatory,
Nearest to http: html, htm, org, www, edu, com, reiterated, https,
Nearest to issue: hittites, hasdai, ftc, amount, examines, way, carangi, newsgroup,
Nearest to behind: breech, ceases, travels, felicity, sending, maximizing, clips, left,
Epoch 5/10 Iteration: 20100 Avg. Training loss: 3.9634 0.1670 sec/batch
Epoch 5/10 Iteration: 20200 Avg. Training loss: 3.9934 0.1670 sec/batch
Epoch 5/10 Iteration: 20300 Avg. Training loss: 3.9507 0.1668 sec/batch
Epoch 5/10 Iteration: 20400 Avg. Training loss: 3.9882 0.1656 sec/batch
Epoch 5/10 Iteration: 20500 Avg. Training loss: 4.0209 0.1662 sec/batch
Epoch 5/10 Iteration: 20600 Avg. Training loss: 3.9292 0.1679 sec/batch
Epoch 5/10 Iteration: 20700 Avg. Training loss: 3.9963 0.1665 sec/batch
Epoch 5/10 Iteration: 20800 Avg. Training loss: 3.9700 0.1652 sec/batch
Epoch 5/10 Iteration: 20900 Avg. Training loss: 3.9677 0.1653 sec/batch
Epoch 5/10 Iteration: 21000 Avg. Training loss: 4.0103 0.1644 sec/batch
Nearest to or: are, waxy, candles, is, pocketing, footpaths, be, cepheus,
Nearest to than: masses, case, amon, karabakh, amending, candidates, monos, analyzes,
Nearest to d: raffarin, composer, variste, antoine, geneticist, martinique, cardiology, slovenian,
Nearest to about: sq, round, democide, reston, gd, vicious, expected, rahula,
Nearest to zero: two, six, three, eight, one, nine, five, four,
Nearest to five: three, zero, four, two, one, six, nine, seven,
Nearest to th: century, nd, rd, pentameter, andhras, deportivo, shaded, nouakchott,
Nearest to more: tagging, funk, frequent, rephlex, capital, consecutive, mandolinists, ability,
Nearest to alternative: grammarians, asean, genetics, streetcar, rgen, intelligence, deprives, seaborg,
Nearest to prince: salian, reigning, pelts, surratt, surmise, conventual, quell, elizabeth,
Nearest to police: patrolled, resistance, pastrana, bevel, strangle, paramilitary, postpartum, legality,
Nearest to professional: students, soccer, manifested, rec, oncology, wrestling, quo, smale,
Nearest to construction: bulky, bab, barberini, schoolmates, linseed, medway, fergus, conservatory,
Nearest to http: www, htm, html, org, edu, com, reiterated, https,
Nearest to issue: examines, hasdai, hittites, amount, ftc, way, carangi, filled,
Nearest to behind: breech, ceases, travels, felicity, sending, maximizing, clips, erica,
Epoch 5/10 Iteration: 21100 Avg. Training loss: 3.9841 0.1680 sec/batch
Epoch 5/10 Iteration: 21200 Avg. Training loss: 3.9803 0.1671 sec/batch
Epoch 5/10 Iteration: 21300 Avg. Training loss: 3.9734 0.1634 sec/batch
Epoch 5/10 Iteration: 21400 Avg. Training loss: 3.9673 0.1653 sec/batch
Epoch 5/10 Iteration: 21500 Avg. Training loss: 3.9937 0.1655 sec/batch
Epoch 5/10 Iteration: 21600 Avg. Training loss: 3.9838 0.1635 sec/batch
Epoch 5/10 Iteration: 21700 Avg. Training loss: 3.9711 0.1676 sec/batch
Epoch 5/10 Iteration: 21800 Avg. Training loss: 3.9308 0.1675 sec/batch
Epoch 5/10 Iteration: 21900 Avg. Training loss: 3.9585 0.1679 sec/batch
Epoch 5/10 Iteration: 22000 Avg. Training loss: 4.0277 0.1683 sec/batch
Nearest to or: are, candles, waxy, is, cepheus, usually, footpaths, praecox,
Nearest to than: masses, case, karabakh, amon, grazing, amending, percent, monos,
Nearest to d: raffarin, composer, variste, martinique, antoine, geneticist, cardiology, slovenian,
Nearest to about: sq, round, democide, vicious, reston, showering, rahula, expected,
Nearest to zero: two, six, three, five, one, nine, eight, four,
Nearest to five: three, zero, four, two, six, one, seven, nine,
Nearest to th: century, nd, rd, andhras, pentameter, ghilzai, nouakchott, shaded,
Nearest to more: tagging, frequent, rephlex, gonzales, imo, ability, funk, consecutive,
Nearest to alternative: asean, grammarians, streetcar, genetics, rgen, deprives, seaborg, scarcity,
Nearest to prince: reigning, salian, pelts, surratt, surmise, keflav, conventual, projecting,
Nearest to police: patrolled, resistance, pastrana, paramilitary, bevel, legality, pde, comanche,
Nearest to professional: students, soccer, wrestling, manifested, rec, quo, cleaveland, oncology,
Nearest to construction: barberini, bulky, bab, fergus, schoolmates, medway, linseed, conservatory,
Nearest to http: html, www, htm, org, edu, com, reiterated, https,
Nearest to issue: ftc, examines, hittites, hasdai, carangi, amount, newsgroup, way,
Nearest to behind: breech, felicity, travels, ceases, sending, maximizing, berchtesgaden, erica,
Epoch 5/10 Iteration: 22100 Avg. Training loss: 3.9688 0.1697 sec/batch
Epoch 5/10 Iteration: 22200 Avg. Training loss: 4.0446 0.1663 sec/batch
Epoch 5/10 Iteration: 22300 Avg. Training loss: 4.0235 0.1674 sec/batch
Epoch 5/10 Iteration: 22400 Avg. Training loss: 4.0339 0.1654 sec/batch
Epoch 5/10 Iteration: 22500 Avg. Training loss: 3.9288 0.1662 sec/batch
Epoch 5/10 Iteration: 22600 Avg. Training loss: 3.9474 0.1647 sec/batch
Epoch 5/10 Iteration: 22700 Avg. Training loss: 3.9592 0.1654 sec/batch
Epoch 5/10 Iteration: 22800 Avg. Training loss: 3.8828 0.1667 sec/batch
Epoch 5/10 Iteration: 22900 Avg. Training loss: 3.9972 0.1655 sec/batch
Epoch 5/10 Iteration: 23000 Avg. Training loss: 3.9827 0.1650 sec/batch
Nearest to or: are, waxy, is, candles, such, thread, praecox, cepheus,
Nearest to than: masses, case, karabakh, amon, monos, grazing, intervals, percent,
Nearest to d: composer, variste, raffarin, u, martinique, cardiology, slovenian, biologist,
Nearest to about: sq, round, gd, democide, vicious, reston, expected, showering,
Nearest to zero: two, six, three, five, nine, one, four, eight,
Nearest to five: three, zero, four, two, one, six, seven, nine,
Nearest to th: century, nd, rd, andhras, deportivo, ghilzai, pentameter, shaded,
Nearest to more: tagging, frequent, rephlex, capital, funk, harsh, consecutive, nodes,
Nearest to alternative: asean, grammarians, seaborg, streetcar, genetics, rgen, creel, scarcity,
Nearest to prince: reigning, salian, pelts, surratt, keflav, surmise, duchess, conventual,
Nearest to police: resistance, patrolled, pastrana, bevel, paramilitary, comanche, pde, postpartum,
Nearest to professional: soccer, students, manifested, wrestling, rec, smale, cleaveland, quo,
Nearest to construction: barberini, bab, bulky, fergus, medway, schoolmates, abroad, conservatory,
Nearest to http: html, www, htm, org, edu, com, nicholls, reiterated,
Nearest to issue: ftc, hittites, hasdai, examines, carangi, amount, kadena, sackville,
Nearest to behind: breech, travels, felicity, ceases, rube, maximizing, clips, erica,
Epoch 5/10 Iteration: 23100 Avg. Training loss: 3.9661 0.1650 sec/batch
Epoch 6/10 Iteration: 23200 Avg. Training loss: 4.0038 0.1082 sec/batch
Epoch 6/10 Iteration: 23300 Avg. Training loss: 3.9635 0.1654 sec/batch
Epoch 6/10 Iteration: 23400 Avg. Training loss: 3.9050 0.1635 sec/batch
Epoch 6/10 Iteration: 23500 Avg. Training loss: 3.9726 0.1623 sec/batch
Epoch 6/10 Iteration: 23600 Avg. Training loss: 3.8918 0.1639 sec/batch
Epoch 6/10 Iteration: 23700 Avg. Training loss: 3.9344 0.1642 sec/batch
Epoch 6/10 Iteration: 23800 Avg. Training loss: 3.9117 0.1648 sec/batch
Epoch 6/10 Iteration: 23900 Avg. Training loss: 3.9127 0.1649 sec/batch
Epoch 6/10 Iteration: 24000 Avg. Training loss: 3.9356 0.1640 sec/batch
Nearest to or: are, waxy, is, of, candles, called, be, and,
Nearest to than: case, karabakh, masses, amon, five, desqview, cotta, although,
Nearest to d: variste, raffarin, composer, martinique, cardiology, u, macdowell, mwi,
Nearest to about: sq, round, gd, democide, waters, rahula, vicious, showering,
Nearest to zero: two, six, three, four, five, eight, one, nine,
Nearest to five: three, four, two, zero, six, one, seven, nine,
Nearest to th: century, nd, rd, ghilzai, deportivo, andhras, pentameter, nouakchott,
Nearest to more: tagging, frequent, eml, intertoto, consecutive, capital, twice, niro,
Nearest to alternative: asean, grammarians, creel, genetics, rgen, streetcar, seaborg, deprives,
Nearest to prince: reigning, salian, pelts, surratt, surmise, projecting, refitted, gage,
Nearest to police: patrolled, resistance, pastrana, paramilitary, bevel, comanche, postpartum, pde,
Nearest to professional: soccer, manifested, rec, students, wrestling, cleaveland, kay, quo,
Nearest to construction: barberini, bab, bulky, fergus, medway, schoolmates, linseed, capo,
Nearest to http: html, www, htm, org, edu, com, https, reiterated,
Nearest to issue: ftc, hasdai, examines, amount, hittites, carangi, kadena, merges,
Nearest to behind: breech, travels, ceases, felicity, maximizing, sending, screams, rube,
Epoch 6/10 Iteration: 24100 Avg. Training loss: 3.9074 0.1670 sec/batch
Epoch 6/10 Iteration: 24200 Avg. Training loss: 3.9648 0.1675 sec/batch
Epoch 6/10 Iteration: 24300 Avg. Training loss: 3.8488 0.1685 sec/batch
Epoch 6/10 Iteration: 24400 Avg. Training loss: 3.9387 0.1684 sec/batch
Epoch 6/10 Iteration: 24500 Avg. Training loss: 3.9057 0.1675 sec/batch
Epoch 6/10 Iteration: 24600 Avg. Training loss: 3.8843 0.1661 sec/batch
Epoch 6/10 Iteration: 24700 Avg. Training loss: 3.8970 0.1674 sec/batch
Epoch 6/10 Iteration: 24800 Avg. Training loss: 3.9591 0.1686 sec/batch
Epoch 6/10 Iteration: 24900 Avg. Training loss: 3.9121 0.1671 sec/batch
Epoch 6/10 Iteration: 25000 Avg. Training loss: 3.8910 0.1647 sec/batch
Nearest to or: are, is, waxy, be, of, usually, candles, cepheus,
Nearest to than: case, amon, masses, much, karabakh, desqview, five, firmware,
Nearest to d: variste, composer, raffarin, cardiology, martinique, u, fibonacci, wrapper,
Nearest to about: sq, round, democide, rahula, gd, vicious, reston, triglyceride,
Nearest to zero: two, six, three, five, one, four, eight, nine,
Nearest to five: three, four, zero, two, one, seven, six, nine,
Nearest to th: century, nd, rd, deportivo, ghilzai, andhras, shaded, nouakchott,
Nearest to more: tagging, frequent, twice, rephlex, consecutive, eml, ability, capital,
Nearest to alternative: asean, grammarians, genetics, streetcar, rgen, creel, albums, extroverted,
Nearest to prince: reigning, salian, pelts, surratt, surmise, projecting, conventual, keflav,
Nearest to police: patrolled, resistance, paramilitary, pastrana, pde, bevel, comanche, postpartum,
Nearest to professional: soccer, students, manifested, wrestling, profession, aircrew, cleaveland, rec,
Nearest to construction: barberini, bab, bulky, medway, fergus, linseed, capo, abroad,
Nearest to http: www, html, htm, org, edu, com, reiterated, https,
Nearest to issue: ftc, examines, amount, hasdai, hittites, carangi, sackville, kadena,
Nearest to behind: breech, travels, maximizing, ceases, felicity, clips, sending, communicative,
Epoch 6/10 Iteration: 25100 Avg. Training loss: 3.9932 0.1682 sec/batch
Epoch 6/10 Iteration: 25200 Avg. Training loss: 3.9089 0.1658 sec/batch
Epoch 6/10 Iteration: 25300 Avg. Training loss: 3.8960 0.1671 sec/batch
Epoch 6/10 Iteration: 25400 Avg. Training loss: 3.9427 0.1665 sec/batch
Epoch 6/10 Iteration: 25500 Avg. Training loss: 3.9301 0.1642 sec/batch
Epoch 6/10 Iteration: 25600 Avg. Training loss: 3.9379 0.1644 sec/batch
Epoch 6/10 Iteration: 25700 Avg. Training loss: 3.9504 0.1644 sec/batch
Epoch 6/10 Iteration: 25800 Avg. Training loss: 3.9364 0.1633 sec/batch
Epoch 6/10 Iteration: 25900 Avg. Training loss: 3.9154 0.1645 sec/batch
Epoch 6/10 Iteration: 26000 Avg. Training loss: 3.9435 0.1636 sec/batch
Nearest to or: are, waxy, is, candles, cepheus, usually, of, afghani,
Nearest to than: masses, karabakh, case, amon, five, much, july, generally,
Nearest to d: variste, raffarin, composer, cardiology, martinique, fibonacci, malthus, macdowell,
Nearest to about: sq, round, democide, reston, vicious, rahula, gd, showering,
Nearest to zero: two, six, three, five, one, nine, four, eight,
Nearest to five: four, three, zero, two, six, one, seven, nine,
Nearest to th: century, nd, rd, andhras, ghilzai, deportivo, pentameter, nouakchott,
Nearest to more: frequent, tagging, rephlex, twice, scot, capital, satellites, ability,
Nearest to alternative: asean, grammarians, streetcar, genetics, seaborg, rgen, albums, kj,
Nearest to prince: reigning, salian, pelts, surmise, emperor, bout, kabbalistic, projecting,
Nearest to police: resistance, patrolled, paramilitary, pastrana, pde, postpartum, bevel, comanche,
Nearest to professional: soccer, students, profession, manifested, wrestling, cleaveland, rec, grappling,
Nearest to construction: bab, barberini, bulky, fergus, linseed, capo, medway, bogdanovich,
Nearest to http: www, html, htm, org, edu, com, https, reiterated,
Nearest to issue: ftc, hasdai, examines, amount, hittites, carangi, greens, filled,
Nearest to behind: breech, travels, ceases, maximizing, felicity, communicative, screams, chihuahua,
Epoch 6/10 Iteration: 26100 Avg. Training loss: 3.9121 0.1667 sec/batch
Epoch 6/10 Iteration: 26200 Avg. Training loss: 3.9082 0.1656 sec/batch
Epoch 6/10 Iteration: 26300 Avg. Training loss: 3.9518 0.1671 sec/batch
Epoch 6/10 Iteration: 26400 Avg. Training loss: 3.9011 0.1689 sec/batch
Epoch 6/10 Iteration: 26500 Avg. Training loss: 3.9222 0.1685 sec/batch
Epoch 6/10 Iteration: 26600 Avg. Training loss: 3.9419 0.1693 sec/batch
Epoch 6/10 Iteration: 26700 Avg. Training loss: 3.9349 0.1679 sec/batch
Epoch 6/10 Iteration: 26800 Avg. Training loss: 4.0291 0.1655 sec/batch
Epoch 6/10 Iteration: 26900 Avg. Training loss: 3.9783 0.1669 sec/batch
Epoch 6/10 Iteration: 27000 Avg. Training loss: 3.9745 0.1665 sec/batch
Nearest to or: are, is, of, usually, waxy, such, cepheus, be,
Nearest to than: five, amon, case, karabakh, masses, tallied, much, amending,
Nearest to d: composer, variste, raffarin, antoine, cardiology, neander, slovenian, macdowell,
Nearest to about: sq, democide, round, rahula, reston, vicious, reaffirms, gd,
Nearest to zero: two, six, three, one, five, four, nine, eight,
Nearest to five: four, three, zero, two, one, six, seven, nine,
Nearest to th: century, nd, rd, andhras, ghilzai, deportivo, zero, nouakchott,
Nearest to more: frequent, tagging, twice, satellites, rephlex, harsh, intertoto, gonzales,
Nearest to alternative: asean, grammarians, streetcar, genetics, albums, seaborg, extroverted, kj,
Nearest to prince: reigning, salian, pelts, bout, surratt, surmise, duchess, starred,
Nearest to police: resistance, patrolled, paramilitary, pastrana, comanche, bevel, tallest, pde,
Nearest to professional: soccer, students, wrestling, profession, manifested, cleaveland, aircrew, exopterygota,
Nearest to construction: bab, barberini, fergus, capo, bulky, abroad, medway, conservatory,
Nearest to http: html, htm, www, org, edu, com, https, reiterated,
Nearest to issue: ftc, examines, hasdai, carangi, amount, way, hittites, kadena,
Nearest to behind: breech, travels, ceases, maximizing, screams, chihuahua, felicity, communicative,
Epoch 6/10 Iteration: 27100 Avg. Training loss: 3.9026 0.1675 sec/batch
Epoch 6/10 Iteration: 27200 Avg. Training loss: 3.8631 0.1671 sec/batch
Epoch 6/10 Iteration: 27300 Avg. Training loss: 3.9202 0.1655 sec/batch
Epoch 6/10 Iteration: 27400 Avg. Training loss: 3.8360 0.1660 sec/batch
Epoch 6/10 Iteration: 27500 Avg. Training loss: 3.9472 0.1656 sec/batch
Epoch 6/10 Iteration: 27600 Avg. Training loss: 3.9156 0.1650 sec/batch
Epoch 6/10 Iteration: 27700 Avg. Training loss: 3.9224 0.1661 sec/batch
Epoch 7/10 Iteration: 27800 Avg. Training loss: 4.0107 0.0635 sec/batch
Epoch 7/10 Iteration: 27900 Avg. Training loss: 3.8910 0.1648 sec/batch
Epoch 7/10 Iteration: 28000 Avg. Training loss: 3.9190 0.1663 sec/batch
Nearest to or: are, is, of, waxy, usually, such, be, summoner,
Nearest to than: masses, case, tallied, amon, believed, higher, amending, karabakh,
Nearest to d: composer, variste, raffarin, cardiology, einar, slovenian, antoine, macdowell,
Nearest to about: sq, round, vicious, reston, democide, gd, rahula, showering,
Nearest to zero: two, six, three, one, five, four, nine, eight,
Nearest to five: four, three, zero, two, one, six, seven, nine,
Nearest to th: century, nd, rd, andhras, zero, ghilzai, deportivo, nouakchott,
Nearest to more: frequent, twice, tagging, satellites, sway, rephlex, scot, frowned,
Nearest to alternative: asean, grammarians, albums, streetcar, gigabytes, seaborg, creel, kj,
Nearest to prince: reigning, salian, pelts, bout, keflav, gage, vanier, starred,
Nearest to police: patrolled, resistance, paramilitary, pastrana, comanche, bevel, infringe, pde,
Nearest to professional: profession, manifested, soccer, students, aircrew, wrestling, cleaveland, kindergarten,
Nearest to construction: bab, capo, fergus, bulky, barberini, abroad, medway, bogdanovich,
Nearest to http: html, www, htm, org, edu, com, reiterated, https,
Nearest to issue: ftc, hasdai, examines, carangi, way, amount, hittites, greens,
Nearest to behind: breech, travels, ceases, maximizing, screams, chihuahua, veyron, rube,
Epoch 7/10 Iteration: 28100 Avg. Training loss: 3.8762 0.1651 sec/batch
Epoch 7/10 Iteration: 28200 Avg. Training loss: 3.8957 0.1635 sec/batch
Epoch 7/10 Iteration: 28300 Avg. Training loss: 3.8758 0.1628 sec/batch
Epoch 7/10 Iteration: 28400 Avg. Training loss: 3.8911 0.1643 sec/batch
Epoch 7/10 Iteration: 28500 Avg. Training loss: 3.8613 0.1657 sec/batch
Epoch 7/10 Iteration: 28600 Avg. Training loss: 3.9204 0.1647 sec/batch
Epoch 7/10 Iteration: 28700 Avg. Training loss: 3.8907 0.1653 sec/batch
Epoch 7/10 Iteration: 28800 Avg. Training loss: 3.9277 0.1672 sec/batch
Epoch 7/10 Iteration: 28900 Avg. Training loss: 3.8104 0.1660 sec/batch
Epoch 7/10 Iteration: 29000 Avg. Training loss: 3.8523 0.1681 sec/batch
Nearest to or: are, is, waxy, of, footpaths, usually, such, called,
Nearest to than: much, case, masses, five, higher, amon, percent, tallied,
Nearest to d: variste, composer, raffarin, cardiology, martinique, einar, nous, scratchpad,
Nearest to about: sq, round, rahula, reston, vicious, gd, democide, showering,
Nearest to zero: two, six, three, five, four, one, nine, eight,
Nearest to five: four, three, zero, two, six, one, seven, nine,
Nearest to th: century, nd, rd, zero, andhras, deportivo, shaded, ghilzai,
Nearest to more: frequent, tagging, twice, eml, krill, ability, pamir, uniquely,
Nearest to alternative: asean, grammarians, streetcar, albums, creel, seaborg, aegeas, extroverted,
Nearest to prince: reigning, pelts, salian, bout, keflav, surmise, ecologically, gage,
Nearest to police: patrolled, resistance, paramilitary, pastrana, comanche, bevel, pde, postpartum,
Nearest to professional: soccer, manifested, students, profession, cleaveland, aircrew, wrestling, dienes,
Nearest to construction: bab, fergus, storey, bulky, barberini, abroad, bogdanovich, capo,
Nearest to http: www, html, htm, org, edu, com, reiterated, asp,
Nearest to issue: ftc, hasdai, examines, amount, way, greens, hittites, carangi,
Nearest to behind: breech, ceases, travels, maximizing, chihuahua, formation, felicity, heathrow,
Epoch 7/10 Iteration: 29100 Avg. Training loss: 3.9038 0.1698 sec/batch
Epoch 7/10 Iteration: 29200 Avg. Training loss: 3.7889 0.1675 sec/batch
Epoch 7/10 Iteration: 29300 Avg. Training loss: 3.8658 0.1662 sec/batch
Epoch 7/10 Iteration: 29400 Avg. Training loss: 3.8894 0.1668 sec/batch
Epoch 7/10 Iteration: 29500 Avg. Training loss: 3.8909 0.1667 sec/batch
Epoch 7/10 Iteration: 29600 Avg. Training loss: 3.9012 0.1668 sec/batch
Epoch 7/10 Iteration: 29700 Avg. Training loss: 3.9424 0.1642 sec/batch
Epoch 7/10 Iteration: 29800 Avg. Training loss: 3.9012 0.1654 sec/batch
Epoch 7/10 Iteration: 29900 Avg. Training loss: 3.8822 0.1651 sec/batch
Epoch 7/10 Iteration: 30000 Avg. Training loss: 3.8747 0.1667 sec/batch
Nearest to or: are, is, of, usually, such, be, waxy, footpaths,
Nearest to than: much, masses, amon, five, case, tallied, believed, higher,
Nearest to d: variste, cardiology, composer, scratchpad, mwi, macdowell, fibonacci, raffarin,
Nearest to about: sq, round, vicious, reston, democide, rahula, gd, zero,
Nearest to zero: two, six, three, one, five, four, nine, eight,
Nearest to five: four, three, zero, two, one, six, seven, nine,
Nearest to th: century, rd, nd, andhras, deportivo, shaded, zero, ghilzai,
Nearest to more: twice, frequent, tagging, uniquely, krill, pamir, frowned, consecutive,
Nearest to alternative: asean, grammarians, streetcar, albums, kj, genetics, rgen, seaborg,
Nearest to prince: salian, reigning, pelts, duchess, gage, hedwig, starred, bout,
Nearest to police: patrolled, resistance, paramilitary, pastrana, comanche, bevel, pde, postpartum,
Nearest to professional: profession, soccer, students, manifested, aircrew, cleaveland, successful, wrestling,
Nearest to construction: bab, bogdanovich, barberini, bulky, capo, storey, abroad, fergus,
Nearest to http: www, html, htm, org, edu, com, reiterated, asp,
Nearest to issue: ftc, hasdai, examines, way, amount, greens, carangi, ceos,
Nearest to behind: breech, travels, ceases, maximizing, heathrow, screams, chihuahua, berchtesgaden,
Epoch 7/10 Iteration: 30100 Avg. Training loss: 3.9128 0.1661 sec/batch
Epoch 7/10 Iteration: 30200 Avg. Training loss: 3.9227 0.1651 sec/batch
Epoch 7/10 Iteration: 30300 Avg. Training loss: 3.8799 0.1648 sec/batch
Epoch 7/10 Iteration: 30400 Avg. Training loss: 3.8698 0.1652 sec/batch
Epoch 7/10 Iteration: 30500 Avg. Training loss: 3.9119 0.1648 sec/batch
Epoch 7/10 Iteration: 30600 Avg. Training loss: 3.9138 0.1645 sec/batch
Epoch 7/10 Iteration: 30700 Avg. Training loss: 3.8967 0.1643 sec/batch
Epoch 7/10 Iteration: 30800 Avg. Training loss: 3.8873 0.1649 sec/batch
Epoch 7/10 Iteration: 30900 Avg. Training loss: 3.8959 0.1653 sec/batch
Epoch 7/10 Iteration: 31000 Avg. Training loss: 3.8544 0.1683 sec/batch
Nearest to or: are, is, of, such, waxy, usually, called, be,
Nearest to than: five, case, much, tallied, percent, masses, higher, amon,
Nearest to d: variste, composer, cardiology, brigade, mwi, raffarin, scratchpad, wrapper,
Nearest to about: sq, round, democide, sophists, rahula, reston, zero, vicious,
Nearest to zero: two, three, six, five, four, one, nine, eight,
Nearest to five: four, three, zero, two, one, six, seven, nine,
Nearest to th: century, rd, nd, andhras, zero, involuntarily, ghilzai, shaded,
Nearest to more: twice, frequent, tagging, krill, scot, imo, gonzales, frowned,
Nearest to alternative: asean, streetcar, grammarians, albums, rgen, genetics, seaborg, creel,
Nearest to prince: salian, reigning, pelts, bout, gage, duchess, treaty, keflav,
Nearest to police: patrolled, resistance, paramilitary, comanche, bevel, civilian, pastrana, postpartum,
Nearest to professional: soccer, profession, students, wrestling, manifested, cleaveland, dienes, thai,
Nearest to construction: storey, barberini, bab, bogdanovich, bulky, fergus, abroad, medway,
Nearest to http: www, htm, html, org, edu, com, reiterated, asp,
Nearest to issue: hasdai, examines, ftc, way, carangi, greens, hittites, filled,
Nearest to behind: breech, travels, ceases, screams, heathrow, maximizing, chihuahua, communicative,
Epoch 7/10 Iteration: 31100 Avg. Training loss: 3.8832 0.1729 sec/batch
Epoch 7/10 Iteration: 31200 Avg. Training loss: 3.8909 0.1689 sec/batch
Epoch 7/10 Iteration: 31300 Avg. Training loss: 3.8885 0.1686 sec/batch
Epoch 7/10 Iteration: 31400 Avg. Training loss: 3.9230 0.1651 sec/batch
Epoch 7/10 Iteration: 31500 Avg. Training loss: 4.0045 0.1656 sec/batch
Epoch 7/10 Iteration: 31600 Avg. Training loss: 3.9298 0.1664 sec/batch
Epoch 7/10 Iteration: 31700 Avg. Training loss: 3.9178 0.1660 sec/batch
Epoch 7/10 Iteration: 31800 Avg. Training loss: 3.8338 0.1659 sec/batch
Epoch 7/10 Iteration: 31900 Avg. Training loss: 3.8861 0.1651 sec/batch
Epoch 7/10 Iteration: 32000 Avg. Training loss: 3.7745 0.1674 sec/batch
Nearest to or: are, is, of, such, four, usually, waxy, and,
Nearest to than: five, case, percent, much, higher, tallied, july, amon,
Nearest to d: variste, composer, cardiology, brigade, mwi, nausea, einar, scratchpad,
Nearest to about: zero, sq, round, on, two, reston, showering, democide,
Nearest to zero: two, one, six, three, five, nine, four, eight,
Nearest to five: four, zero, three, two, one, six, seven, nine,
Nearest to th: century, rd, nd, andhras, zero, involuntarily, deportivo, nouakchott,
Nearest to more: twice, frequent, five, tagging, they, so, daugavpils, pamir,
Nearest to alternative: asean, grammarians, streetcar, creel, rgen, albums, azam, genetics,
Nearest to prince: reigning, salian, pelts, treaty, gage, bout, quell, keflav,
Nearest to police: patrolled, resistance, paramilitary, civilian, pastrana, comanche, bevel, infringe,
Nearest to professional: profession, wrestling, soccer, manifested, students, echelons, dienes, thai,
Nearest to construction: barberini, bogdanovich, bab, abroad, storey, bulky, fergus, capo,
Nearest to http: www, htm, html, org, edu, com, asp, https,
Nearest to issue: ftc, way, examines, hasdai, carangi, hittites, bench, greens,
Nearest to behind: breech, travels, ceases, maximizing, screams, chihuahua, communicative, erica,
Epoch 7/10 Iteration: 32100 Avg. Training loss: 3.9319 0.1690 sec/batch
Epoch 7/10 Iteration: 32200 Avg. Training loss: 3.8936 0.1659 sec/batch
Epoch 7/10 Iteration: 32300 Avg. Training loss: 3.8697 0.1654 sec/batch
Epoch 8/10 Iteration: 32400 Avg. Training loss: 3.9422 0.0187 sec/batch
Epoch 8/10 Iteration: 32500 Avg. Training loss: 3.9000 0.1666 sec/batch
Epoch 8/10 Iteration: 32600 Avg. Training loss: 3.8744 0.1657 sec/batch
Epoch 8/10 Iteration: 32700 Avg. Training loss: 3.8718 0.1649 sec/batch
Epoch 8/10 Iteration: 32800 Avg. Training loss: 3.8989 0.1640 sec/batch
Epoch 8/10 Iteration: 32900 Avg. Training loss: 3.8290 0.1637 sec/batch
Epoch 8/10 Iteration: 33000 Avg. Training loss: 3.8627 0.1654 sec/batch
Nearest to or: are, is, of, such, usually, waxy, be, four,
Nearest to than: much, higher, five, case, percent, amon, only, firmware,
Nearest to d: composer, variste, australopithecus, cardiology, neander, brigade, nausea, einar,
Nearest to about: round, on, sq, reston, zero, vicious, democide, surrey,
Nearest to zero: two, six, three, one, five, four, nine, eight,
Nearest to five: four, three, one, two, zero, six, seven, nine,
Nearest to th: century, nd, rd, andhras, zero, divisional, deportivo, involuntarily,
Nearest to more: twice, frequent, they, tagging, ability, krill, satellites, scot,
Nearest to alternative: asean, grammarians, streetcar, creel, albums, azam, rgen, los,
Nearest to prince: reigning, salian, pelts, bout, kabbalistic, treaty, keflav, surmise,
Nearest to police: patrolled, resistance, paramilitary, ermey, civilian, comanche, bevel, balzac,
Nearest to professional: profession, soccer, wrestling, echelons, successful, thai, aircrew, cleaveland,
Nearest to construction: barberini, bogdanovich, bab, storey, capo, fergus, abroad, strands,
Nearest to http: www, htm, html, org, com, edu, asp, reiterated,
Nearest to issue: way, examines, carangi, hasdai, ftc, hittites, greens, amount,
Nearest to behind: breech, travels, ceases, maximizing, chihuahua, screams, heathrow, shed,
Epoch 8/10 Iteration: 33100 Avg. Training loss: 3.8290 0.1659 sec/batch
Epoch 8/10 Iteration: 33200 Avg. Training loss: 3.8496 0.1634 sec/batch
Epoch 8/10 Iteration: 33300 Avg. Training loss: 3.8745 0.1636 sec/batch
Epoch 8/10 Iteration: 33400 Avg. Training loss: 3.9227 0.1665 sec/batch
Epoch 8/10 Iteration: 33500 Avg. Training loss: 3.8508 0.1679 sec/batch
Epoch 8/10 Iteration: 33600 Avg. Training loss: 3.7979 0.1661 sec/batch
Epoch 8/10 Iteration: 33700 Avg. Training loss: 3.8140 0.1683 sec/batch
Epoch 8/10 Iteration: 33800 Avg. Training loss: 3.7758 0.1704 sec/batch
Epoch 8/10 Iteration: 33900 Avg. Training loss: 3.8384 0.1679 sec/batch
Epoch 8/10 Iteration: 34000 Avg. Training loss: 3.8556 0.1671 sec/batch
Nearest to or: are, is, of, such, be, usually, waxy, and,
Nearest to than: much, five, as, only, amon, higher, at, tallied,
Nearest to d: variste, australopithecus, cardiology, nausea, composer, scratchpad, mwi, conflate,
Nearest to about: round, reston, sq, zero, on, vicious, democide, two,
Nearest to zero: two, one, three, six, five, four, nine, eight,
Nearest to five: four, three, two, one, zero, seven, six, nine,
Nearest to th: century, nd, rd, andhras, zero, involuntarily, shaded, ghilzai,
Nearest to more: twice, are, frequent, tagging, they, to, ability, so,
Nearest to alternative: asean, grammarians, streetcar, creel, rgen, extroverted, azam, albums,
Nearest to prince: pelts, reigning, keflav, salian, treaty, kabbalistic, refitted, surmise,
Nearest to police: patrolled, resistance, paramilitary, compliant, ermey, civilian, bevel, pastrana,
Nearest to professional: profession, soccer, echelons, wrestling, dienes, thai, aircrew, cleaveland,
Nearest to construction: barberini, bogdanovich, bab, storey, abroad, capo, strands, fergus,
Nearest to http: www, htm, html, org, com, edu, reiterated, asp,
Nearest to issue: way, examines, hasdai, ftc, amount, hittites, filled, carangi,
Nearest to behind: breech, travels, ceases, left, maximizing, heathrow, screams, chihuahua,
Epoch 8/10 Iteration: 34100 Avg. Training loss: 3.8617 0.1690 sec/batch
Epoch 8/10 Iteration: 34200 Avg. Training loss: 3.8235 0.1661 sec/batch
Epoch 8/10 Iteration: 34300 Avg. Training loss: 3.9085 0.1657 sec/batch
Epoch 8/10 Iteration: 34400 Avg. Training loss: 3.8860 0.1637 sec/batch
Epoch 8/10 Iteration: 34500 Avg. Training loss: 3.8408 0.1657 sec/batch
Epoch 8/10 Iteration: 34600 Avg. Training loss: 3.8638 0.1667 sec/batch
Epoch 8/10 Iteration: 34700 Avg. Training loss: 3.8842 0.1653 sec/batch
Epoch 8/10 Iteration: 34800 Avg. Training loss: 3.8799 0.1656 sec/batch
Epoch 8/10 Iteration: 34900 Avg. Training loss: 3.8598 0.1651 sec/batch
Epoch 8/10 Iteration: 35000 Avg. Training loss: 3.9081 0.1659 sec/batch
Nearest to or: are, is, of, such, usually, be, four, and,
Nearest to than: much, tallied, amon, higher, less, suited, case, lower,
Nearest to d: variste, composer, nausea, australopithecus, cardiology, malthus, brigade, neander,
Nearest to about: round, zero, reston, democide, on, vicious, herg, sq,
Nearest to zero: two, three, one, four, six, five, nine, eight,
Nearest to five: four, three, one, zero, two, seven, six, nine,
Nearest to th: century, nd, rd, andhras, zero, involuntarily, shaded, birthday,
Nearest to more: twice, tagging, frequent, scot, they, ability, still, daugavpils,
Nearest to alternative: asean, grammarians, streetcar, extroverted, rgen, azam, albums, gigabytes,
Nearest to prince: reigning, salian, kabbalistic, pelts, treaty, bout, surmise, keflav,
Nearest to police: patrolled, paramilitary, resistance, pastrana, civilian, compliant, ermey, bevel,
Nearest to professional: profession, echelons, wrestling, successful, dienes, soccer, cleaveland, thai,
Nearest to construction: barberini, bab, storey, bogdanovich, strands, abroad, bulky, capo,
Nearest to http: www, htm, html, org, com, edu, asp, reiterated,
Nearest to issue: hasdai, way, ftc, examines, carangi, ross, filled, bench,
Nearest to behind: breech, chihuahua, ceases, travels, maximizing, heathrow, left, screams,
Epoch 8/10 Iteration: 35100 Avg. Training loss: 3.8566 0.1665 sec/batch
Epoch 8/10 Iteration: 35200 Avg. Training loss: 3.8923 0.1643 sec/batch
Epoch 8/10 Iteration: 35300 Avg. Training loss: 3.8839 0.1652 sec/batch
Epoch 8/10 Iteration: 35400 Avg. Training loss: 3.8926 0.1654 sec/batch
Epoch 8/10 Iteration: 35500 Avg. Training loss: 3.8711 0.1640 sec/batch
Epoch 8/10 Iteration: 35600 Avg. Training loss: 3.8635 0.1657 sec/batch
Epoch 8/10 Iteration: 35700 Avg. Training loss: 3.8642 0.1674 sec/batch
Epoch 8/10 Iteration: 35800 Avg. Training loss: 3.8453 0.1687 sec/batch
Epoch 8/10 Iteration: 35900 Avg. Training loss: 3.9592 0.1701 sec/batch
Epoch 8/10 Iteration: 36000 Avg. Training loss: 3.8580 0.1666 sec/batch
Nearest to or: are, is, such, of, usually, and, any, waxy,
Nearest to than: much, tallied, amon, less, lower, higher, case, percent,
Nearest to d: composer, one, variste, malthus, neander, slovenian, brigade, australopithecus,
Nearest to about: on, reston, democide, round, zero, vicious, surrey, herg,
Nearest to zero: two, three, one, five, six, nine, four, eight,
Nearest to five: one, three, four, two, zero, nine, six, seven,
Nearest to th: century, rd, nd, andhras, zero, involuntarily, ghilzai, shaded,
Nearest to more: tagging, twice, frequent, scot, still, imo, strive, daugavpils,
Nearest to alternative: asean, streetcar, grammarians, azam, albums, extroverted, rgen, gigabytes,
Nearest to prince: reigning, treaty, pelts, salian, duchess, kabbalistic, bout, refitted,
Nearest to police: paramilitary, resistance, patrolled, comanche, civilian, pastrana, bevel, ermey,
Nearest to professional: wrestling, profession, soccer, echelons, cleaveland, successful, thai, aircrew,
Nearest to construction: storey, bogdanovich, barberini, bab, abroad, capo, strands, dam,
Nearest to http: www, htm, html, org, edu, com, nicholls, reiterated,
Nearest to issue: way, ftc, hasdai, bench, examines, ross, hittites, carangi,
Nearest to behind: breech, chihuahua, ceases, travels, heathrow, screams, guthrie, maximizing,
Epoch 8/10 Iteration: 36100 Avg. Training loss: 3.9552 0.1681 sec/batch
Epoch 8/10 Iteration: 36200 Avg. Training loss: 3.9452 0.1656 sec/batch
Epoch 8/10 Iteration: 36300 Avg. Training loss: 3.8963 0.1665 sec/batch
Epoch 8/10 Iteration: 36400 Avg. Training loss: 3.8457 0.1661 sec/batch
Epoch 8/10 Iteration: 36500 Avg. Training loss: 3.8664 0.1668 sec/batch
Epoch 8/10 Iteration: 36600 Avg. Training loss: 3.8205 0.1661 sec/batch
Epoch 8/10 Iteration: 36700 Avg. Training loss: 3.8471 0.1650 sec/batch
Epoch 8/10 Iteration: 36800 Avg. Training loss: 3.8736 0.1648 sec/batch
Epoch 8/10 Iteration: 36900 Avg. Training loss: 3.8503 0.1653 sec/batch
Epoch 8/10 Iteration: 37000 Avg. Training loss: 3.8977 0.1663 sec/batch
Nearest to or: are, is, such, of, usually, and, sometimes, as,
Nearest to than: tallied, less, much, higher, lower, suited, percent, only,
Nearest to d: composer, variste, cardiology, neander, australopithecus, einar, slovenian, brigade,
Nearest to about: zero, on, reston, round, two, democide, vicious, gd,
Nearest to zero: two, three, five, one, six, nine, four, eight,
Nearest to five: two, zero, one, three, four, six, nine, seven,
Nearest to th: century, nd, rd, andhras, involuntarily, shaded, deportivo, zero,
Nearest to more: tagging, still, twice, scot, so, frequent, daugavpils, ability,
Nearest to alternative: streetcar, asean, gigabytes, azam, grammarians, albums, extroverted, creel,
Nearest to prince: reigning, treaty, pelts, bout, keflav, salian, kabbalistic, starred,
Nearest to police: patrolled, resistance, paramilitary, ermey, comanche, pastrana, civilian, pistol,
Nearest to professional: profession, wrestling, echelons, thai, successful, soccer, aircrew, dienes,
Nearest to construction: storey, bogdanovich, barberini, capo, dam, strands, abroad, bab,
Nearest to http: www, htm, html, org, edu, com, nicholls, asp,
Nearest to issue: bench, ftc, way, filled, hasdai, carangi, ross, examines,
Nearest to behind: breech, chihuahua, travels, ceases, maximizing, guthrie, heathrow, left,
Epoch 9/10 Iteration: 37100 Avg. Training loss: 3.8873 0.1392 sec/batch
Epoch 9/10 Iteration: 37200 Avg. Training loss: 3.8327 0.1663 sec/batch
Epoch 9/10 Iteration: 37300 Avg. Training loss: 3.8294 0.1645 sec/batch
Epoch 9/10 Iteration: 37400 Avg. Training loss: 3.8595 0.1625 sec/batch
Epoch 9/10 Iteration: 37500 Avg. Training loss: 3.8014 0.1639 sec/batch
Epoch 9/10 Iteration: 37600 Avg. Training loss: 3.8732 0.1650 sec/batch
Epoch 9/10 Iteration: 37700 Avg. Training loss: 3.7956 0.1649 sec/batch
Epoch 9/10 Iteration: 37800 Avg. Training loss: 3.8288 0.1669 sec/batch
Epoch 9/10 Iteration: 37900 Avg. Training loss: 3.8680 0.1655 sec/batch
Epoch 9/10 Iteration: 38000 Avg. Training loss: 3.8588 0.1690 sec/batch
Nearest to or: are, is, of, such, usually, called, be, sometimes,
Nearest to than: much, higher, tallied, less, amon, five, only, lower,
Nearest to d: variste, cardiology, australopithecus, nausea, composer, scratchpad, kerk, brigade,
Nearest to about: zero, on, reston, round, two, gd, surrey, vicious,
Nearest to zero: two, three, one, six, four, five, nine, eight,
Nearest to five: four, two, three, one, zero, six, seven, nine,
Nearest to th: century, rd, nd, andhras, involuntarily, deportivo, divisional, zero,
Nearest to more: twice, tagging, so, ability, they, still, intertoto, frequent,
Nearest to alternative: azam, streetcar, creel, asean, albums, extroverted, gigabytes, grammarians,
Nearest to prince: reigning, treaty, pelts, kabbalistic, keflav, starred, achromatism, bout,
Nearest to police: patrolled, paramilitary, resistance, ermey, compliant, comanche, civilian, bevel,
Nearest to professional: soccer, wrestling, successful, echelons, profession, aircrew, thai, showmanship,
Nearest to construction: storey, bogdanovich, barberini, capo, bab, bulky, dam, abroad,
Nearest to http: www, htm, html, org, edu, com, nicholls, reiterated,
Nearest to issue: bench, hasdai, ftc, way, examines, ross, greens, hittites,
Nearest to behind: breech, maximizing, heathrow, chihuahua, travels, ceases, team, left,
Epoch 9/10 Iteration: 38100 Avg. Training loss: 3.8603 0.1683 sec/batch
Epoch 9/10 Iteration: 38200 Avg. Training loss: 3.7573 0.1683 sec/batch
Epoch 9/10 Iteration: 38300 Avg. Training loss: 3.8269 0.1692 sec/batch
Epoch 9/10 Iteration: 38400 Avg. Training loss: 3.7998 0.1683 sec/batch
Epoch 9/10 Iteration: 38500 Avg. Training loss: 3.8143 0.1661 sec/batch
Epoch 9/10 Iteration: 38600 Avg. Training loss: 3.8226 0.1672 sec/batch
Epoch 9/10 Iteration: 38700 Avg. Training loss: 3.8363 0.1664 sec/batch
Epoch 9/10 Iteration: 38800 Avg. Training loss: 3.8183 0.1684 sec/batch
Epoch 9/10 Iteration: 38900 Avg. Training loss: 3.8406 0.1643 sec/batch
Epoch 9/10 Iteration: 39000 Avg. Training loss: 3.8992 0.1656 sec/batch
Nearest to or: are, is, of, such, be, any, as, usually,
Nearest to than: much, less, tallied, higher, only, as, lower, amon,
Nearest to d: variste, cardiology, australopithecus, composer, nausea, brigade, scratchpad, nous,
Nearest to about: zero, on, reston, round, two, vicious, democide, surrey,
Nearest to zero: two, three, five, four, one, six, nine, eight,
Nearest to five: four, three, two, one, zero, six, nine, seven,
Nearest to th: century, nd, rd, andhras, involuntarily, shaded, deportivo, divisional,
Nearest to more: twice, are, tagging, they, so, ability, still, such,
Nearest to alternative: streetcar, creel, azam, asean, extroverted, gigabytes, albums, grammarians,
Nearest to prince: treaty, reigning, kabbalistic, pelts, keflav, starred, achromatism, ecologically,
Nearest to police: patrolled, paramilitary, resistance, ermey, civilian, compliant, comanche, pistol,
Nearest to professional: profession, wrestling, echelons, aircrew, successful, soccer, kindergarten, students,
Nearest to construction: bogdanovich, storey, bab, barberini, capo, strands, abroad, dam,
Nearest to http: www, htm, html, org, edu, com, nicholls, asp,
Nearest to issue: bench, ftc, way, examines, timely, ross, hasdai, greens,
Nearest to behind: breech, maximizing, chihuahua, heathrow, left, right, travels, team,
Epoch 9/10 Iteration: 39100 Avg. Training loss: 3.8537 0.1673 sec/batch
Epoch 9/10 Iteration: 39200 Avg. Training loss: 3.8213 0.1673 sec/batch
Epoch 9/10 Iteration: 39300 Avg. Training loss: 3.8471 0.1654 sec/batch
Epoch 9/10 Iteration: 39400 Avg. Training loss: 3.8389 0.1663 sec/batch
Epoch 9/10 Iteration: 39500 Avg. Training loss: 3.8537 0.1667 sec/batch
Epoch 9/10 Iteration: 39600 Avg. Training loss: 3.8823 0.1645 sec/batch
Epoch 9/10 Iteration: 39700 Avg. Training loss: 3.8654 0.1646 sec/batch
Epoch 9/10 Iteration: 39800 Avg. Training loss: 3.8460 0.1638 sec/batch
Epoch 9/10 Iteration: 39900 Avg. Training loss: 3.8720 0.1656 sec/batch
Epoch 9/10 Iteration: 40000 Avg. Training loss: 3.8430 0.1624 sec/batch
Nearest to or: are, is, of, such, usually, as, any, sometimes,
Nearest to than: much, less, lower, only, tallied, as, higher, suited,
Nearest to d: variste, australopithecus, nausea, cardiology, nous, brigade, malthus, scratchpad,
Nearest to about: on, zero, democide, two, round, reston, of, herg,
Nearest to zero: two, three, five, six, four, nine, one, eight,
Nearest to five: four, two, three, one, zero, six, nine, seven,
Nearest to th: century, nd, rd, andhras, involuntarily, shaded, ningbo, monastic,
Nearest to more: tagging, twice, are, so, frequent, they, still, than,
Nearest to alternative: streetcar, azam, kj, extroverted, asean, albums, creel, gigabytes,
Nearest to prince: reigning, treaty, kabbalistic, gage, salian, achromatism, keflav, emperor,
Nearest to police: paramilitary, patrolled, resistance, ermey, civilian, pistol, compliant, comanche,
Nearest to professional: profession, wrestling, kindergarten, soccer, aircrew, successful, echelons, thai,
Nearest to construction: storey, bogdanovich, barberini, bab, capo, strands, dam, bulky,
Nearest to http: www, htm, html, org, edu, com, maths, nicholls,
Nearest to issue: ftc, bench, way, examines, timely, hasdai, ross, greens,
Nearest to behind: breech, chihuahua, maximizing, heathrow, right, formation, taller, ceases,
Epoch 9/10 Iteration: 40100 Avg. Training loss: 3.8501 0.1679 sec/batch
Epoch 9/10 Iteration: 40200 Avg. Training loss: 3.8549 0.1667 sec/batch
Epoch 9/10 Iteration: 40300 Avg. Training loss: 3.8157 0.1671 sec/batch
Epoch 9/10 Iteration: 40400 Avg. Training loss: 3.8016 0.1664 sec/batch
Epoch 9/10 Iteration: 40500 Avg. Training loss: 3.8865 0.1681 sec/batch
Epoch 9/10 Iteration: 40600 Avg. Training loss: 3.8392 0.1668 sec/batch
Epoch 9/10 Iteration: 40700 Avg. Training loss: 3.9547 0.1663 sec/batch
Epoch 9/10 Iteration: 40800 Avg. Training loss: 3.8887 0.1649 sec/batch
Epoch 9/10 Iteration: 40900 Avg. Training loss: 3.8857 0.1659 sec/batch
Epoch 9/10 Iteration: 41000 Avg. Training loss: 3.8087 0.1666 sec/batch
Nearest to or: is, are, such, of, usually, be, sometimes, as,
Nearest to than: less, much, lower, only, five, tallied, more, higher,
Nearest to d: variste, composer, cardiology, australopithecus, brigade, scratchpad, nous, nausea,
Nearest to about: zero, on, two, democide, reston, herg, round, vicious,
Nearest to zero: two, four, three, five, six, one, nine, eight,
Nearest to five: two, one, four, three, zero, six, nine, seven,
Nearest to th: century, nd, rd, andhras, zero, involuntarily, nouakchott, shaded,
Nearest to more: tagging, twice, of, than, still, they, daugavpils, so,
Nearest to alternative: streetcar, azam, kj, albums, creel, extroverted, grammarians, gigabytes,
Nearest to prince: reigning, treaty, keflav, kabbalistic, ecologically, refitted, starred, duchess,
Nearest to police: paramilitary, patrolled, resistance, ermey, civilian, comanche, pistol, balzac,
Nearest to professional: wrestling, profession, aircrew, soccer, successful, kindergarten, thai, saunderson,
Nearest to construction: bogdanovich, storey, barberini, capo, bab, dam, strands, completed,
Nearest to http: www, htm, html, org, com, edu, maths, nicholls,
Nearest to issue: ftc, bench, way, negotiated, hasdai, timely, examines, greens,
Nearest to behind: breech, chihuahua, pollinators, maximizing, heathrow, right, formation, taller,
Epoch 9/10 Iteration: 41100 Avg. Training loss: 3.7792 0.1673 sec/batch
Epoch 9/10 Iteration: 41200 Avg. Training loss: 3.8722 0.1647 sec/batch
Epoch 9/10 Iteration: 41300 Avg. Training loss: 3.7804 0.1664 sec/batch
Epoch 9/10 Iteration: 41400 Avg. Training loss: 3.8669 0.1666 sec/batch
Epoch 9/10 Iteration: 41500 Avg. Training loss: 3.8325 0.1653 sec/batch
Epoch 9/10 Iteration: 41600 Avg. Training loss: 3.8489 0.1646 sec/batch
Epoch 10/10 Iteration: 41700 Avg. Training loss: 3.8758 0.0937 sec/batch
Epoch 10/10 Iteration: 41800 Avg. Training loss: 3.7930 0.1659 sec/batch
Epoch 10/10 Iteration: 41900 Avg. Training loss: 3.8616 0.1649 sec/batch
Epoch 10/10 Iteration: 42000 Avg. Training loss: 3.8441 0.1630 sec/batch
Nearest to or: is, are, of, such, as, and, usually, be,
Nearest to than: less, much, lower, higher, tallied, only, at, more,
Nearest to d: one, composer, variste, neander, cardiology, australopithecus, brigade, scratchpad,
Nearest to about: on, zero, two, reston, democide, of, round, herg,
Nearest to zero: two, six, three, one, nine, five, four, eight,
Nearest to five: one, two, four, three, six, zero, nine, seven,
Nearest to th: century, nd, rd, andhras, zero, involuntarily, ningbo, sinhala,
Nearest to more: tagging, twice, than, they, still, are, such, nearly,
Nearest to alternative: kj, streetcar, creel, used, albums, azam, gigabytes, asean,
Nearest to prince: treaty, reigning, gage, keflav, starred, kabbalistic, salian, achromatism,
Nearest to police: patrolled, paramilitary, resistance, ermey, compliant, comanche, balzac, offences,
Nearest to professional: wrestling, aircrew, profession, thai, successful, manifested, kindergarten, saunderson,
Nearest to construction: bogdanovich, storey, capo, bab, strands, barberini, dam, bulky,
Nearest to http: www, htm, html, org, com, edu, maths, nicholls,
Nearest to issue: ftc, way, bench, hasdai, carangi, greens, timely, scenario,
Nearest to behind: breech, chihuahua, maximizing, heathrow, right, pollinators, formation, taller,
Epoch 10/10 Iteration: 42100 Avg. Training loss: 3.8093 0.1641 sec/batch
Epoch 10/10 Iteration: 42200 Avg. Training loss: 3.8078 0.1646 sec/batch
Epoch 10/10 Iteration: 42300 Avg. Training loss: 3.8105 0.1636 sec/batch
Epoch 10/10 Iteration: 42400 Avg. Training loss: 3.7910 0.1649 sec/batch
Epoch 10/10 Iteration: 42500 Avg. Training loss: 3.8614 0.1645 sec/batch
Epoch 10/10 Iteration: 42600 Avg. Training loss: 3.8066 0.1661 sec/batch
Epoch 10/10 Iteration: 42700 Avg. Training loss: 3.8759 0.1677 sec/batch
Epoch 10/10 Iteration: 42800 Avg. Training loss: 3.7590 0.1672 sec/batch
Epoch 10/10 Iteration: 42900 Avg. Training loss: 3.7911 0.1681 sec/batch
Epoch 10/10 Iteration: 43000 Avg. Training loss: 3.7670 0.1671 sec/batch
Nearest to or: is, are, such, of, be, as, other, and,
Nearest to than: much, less, more, only, lower, higher, five, four,
Nearest to d: australopithecus, variste, cardiology, nausea, scratchpad, martinique, kerk, nous,
Nearest to about: zero, two, of, on, herg, reston, vicious, surrey,
Nearest to zero: two, three, six, five, four, one, nine, eight,
Nearest to five: two, four, three, one, zero, six, seven, nine,
Nearest to th: century, rd, nd, involuntarily, andhras, zero, the, shaded,
Nearest to more: than, twice, of, tagging, to, are, such, still,
Nearest to alternative: albums, streetcar, azam, kj, creel, extroverted, asean, gigabytes,
Nearest to prince: treaty, starred, keflav, achromatism, ecologically, kabbalistic, vaporized, reigning,
Nearest to police: paramilitary, patrolled, resistance, ermey, compliant, comanche, balzac, pistol,
Nearest to professional: aircrew, profession, wrestling, kindergarten, successful, echelons, soccer, cleaveland,
Nearest to construction: bogdanovich, bab, storey, capo, barberini, dam, abroad, ostensibly,
Nearest to http: www, htm, html, org, com, edu, maths, nicholls,
Nearest to issue: bench, ftc, greens, timely, way, kadena, ross, hasdai,
Nearest to behind: breech, right, heathrow, chihuahua, left, maximizing, formation, taller,
Epoch 10/10 Iteration: 43100 Avg. Training loss: 3.7412 0.1698 sec/batch
Epoch 10/10 Iteration: 43200 Avg. Training loss: 3.8093 0.1678 sec/batch
Epoch 10/10 Iteration: 43300 Avg. Training loss: 3.8249 0.1669 sec/batch
Epoch 10/10 Iteration: 43400 Avg. Training loss: 3.8164 0.1679 sec/batch
Epoch 10/10 Iteration: 43500 Avg. Training loss: 3.8062 0.1645 sec/batch
Epoch 10/10 Iteration: 43600 Avg. Training loss: 3.8863 0.1652 sec/batch
Epoch 10/10 Iteration: 43700 Avg. Training loss: 3.8485 0.1669 sec/batch
Epoch 10/10 Iteration: 43800 Avg. Training loss: 3.8310 0.1667 sec/batch
Epoch 10/10 Iteration: 43900 Avg. Training loss: 3.8076 0.1660 sec/batch
Epoch 10/10 Iteration: 44000 Avg. Training loss: 3.8441 0.1668 sec/batch
Nearest to or: are, is, of, other, such, as, be, usually,
Nearest to than: much, less, more, percent, lower, as, suited, higher,
Nearest to d: australopithecus, variste, cardiology, nausea, nous, scratchpad, donn, kerk,
Nearest to about: zero, two, herg, reston, on, vicious, of, round,
Nearest to zero: two, three, four, one, five, six, nine, eight,
Nearest to five: four, two, three, one, zero, six, nine, eight,
Nearest to th: century, nd, rd, andhras, birthday, involuntarily, ningbo, until,
Nearest to more: than, of, tagging, twice, still, so, nearly, other,
Nearest to alternative: albums, streetcar, kj, azam, extroverted, gigabytes, asean, grammarians,
Nearest to prince: reigning, vaporized, treaty, kabbalistic, starred, gage, achromatism, projecting,
Nearest to police: paramilitary, patrolled, resistance, ermey, comanche, compliant, civilian, balzac,
Nearest to professional: profession, aircrew, wrestling, kindergarten, certification, echelons, soccer, successful,
Nearest to construction: bogdanovich, storey, bab, capo, ostensibly, dam, abroad, strands,
Nearest to http: www, htm, html, org, edu, com, maths, nicholls,
Nearest to issue: bench, ftc, greens, timely, way, hasdai, negotiated, ross,
Nearest to behind: breech, right, chihuahua, left, heathrow, pollinators, maximizing, formation,
Epoch 10/10 Iteration: 44100 Avg. Training loss: 3.8531 0.1660 sec/batch
Epoch 10/10 Iteration: 44200 Avg. Training loss: 3.8652 0.1655 sec/batch
Epoch 10/10 Iteration: 44300 Avg. Training loss: 3.8126 0.1655 sec/batch
Epoch 10/10 Iteration: 44400 Avg. Training loss: 3.7984 0.1649 sec/batch
Epoch 10/10 Iteration: 44500 Avg. Training loss: 3.8485 0.1648 sec/batch
Epoch 10/10 Iteration: 44600 Avg. Training loss: 3.8352 0.1653 sec/batch
Epoch 10/10 Iteration: 44700 Avg. Training loss: 3.8115 0.1661 sec/batch
Epoch 10/10 Iteration: 44800 Avg. Training loss: 3.8389 0.1685 sec/batch
Epoch 10/10 Iteration: 44900 Avg. Training loss: 3.7902 0.1680 sec/batch
Epoch 10/10 Iteration: 45000 Avg. Training loss: 3.8437 0.1659 sec/batch
Nearest to or: is, are, of, such, as, be, other, to,
Nearest to than: less, much, more, five, in, but, four, percent,
Nearest to d: australopithecus, nausea, variste, cardiology, scratchpad, nous, before, mwi,
Nearest to about: two, zero, on, interview, herg, of, democide, reston,
Nearest to zero: two, three, one, six, five, four, nine, eight,
Nearest to five: four, one, two, three, zero, six, nine, eight,
Nearest to th: century, nd, rd, the, andhras, involuntarily, centuries, in,
Nearest to more: tagging, than, still, of, so, twice, nearly, to,
Nearest to alternative: streetcar, used, extroverted, gigabytes, albums, kj, azam, asean,
Nearest to prince: reigning, treaty, gage, vaporized, starred, achromatism, keflav, kabbalistic,
Nearest to police: paramilitary, patrolled, resistance, ermey, compliant, comanche, balzac, civilian,
Nearest to professional: profession, wrestling, aircrew, kindergarten, certification, cleaveland, soccer, echelons,
Nearest to construction: bogdanovich, storey, bab, barberini, dam, capo, strands, completed,
Nearest to http: www, htm, html, org, edu, com, maths, nicholls,
Nearest to issue: bench, ftc, way, negotiated, timely, greens, scenario, hasdai,
Nearest to behind: breech, right, chihuahua, heathrow, left, pollinators, taller, formation,
Epoch 10/10 Iteration: 45100 Avg. Training loss: 3.8516 0.1675 sec/batch
Epoch 10/10 Iteration: 45200 Avg. Training loss: 3.8501 0.1665 sec/batch
Epoch 10/10 Iteration: 45300 Avg. Training loss: 3.9262 0.1658 sec/batch
Epoch 10/10 Iteration: 45400 Avg. Training loss: 3.8868 0.1662 sec/batch
Epoch 10/10 Iteration: 45500 Avg. Training loss: 3.8828 0.1643 sec/batch
Epoch 10/10 Iteration: 45600 Avg. Training loss: 3.8363 0.1677 sec/batch
Epoch 10/10 Iteration: 45700 Avg. Training loss: 3.7541 0.1666 sec/batch
Epoch 10/10 Iteration: 45800 Avg. Training loss: 3.8410 0.1674 sec/batch
Epoch 10/10 Iteration: 45900 Avg. Training loss: 3.7449 0.1671 sec/batch
Epoch 10/10 Iteration: 46000 Avg. Training loss: 3.8450 0.1678 sec/batch
Nearest to or: is, are, of, such, as, other, and, to,
Nearest to than: less, much, more, percent, higher, but, lower, suited,
Nearest to d: one, variste, composer, australopithecus, brigade, cardiology, nausea, donn,
Nearest to about: on, zero, two, of, interview, herg, democide, reston,
Nearest to zero: two, five, nine, one, three, six, four, eight,
Nearest to five: one, four, two, zero, three, six, nine, eight,
Nearest to th: century, nd, rd, andhras, compiled, zero, the, days,
Nearest to more: tagging, than, of, still, s, twice, nearly, to,
Nearest to alternative: streetcar, gigabytes, creel, asean, kj, used, albums, garofalo,
Nearest to prince: treaty, reigning, gage, achromatism, keflav, vaporized, ecologically, affaires,
Nearest to police: paramilitary, patrolled, resistance, ermey, compliant, comanche, pistol, ex,
Nearest to professional: wrestling, profession, aircrew, soccer, kindergarten, certification, saunderson, echelons,
Nearest to construction: bogdanovich, storey, bab, strands, dam, capo, barberini, ostensibly,
Nearest to http: www, htm, html, org, edu, com, maths, asp,
Nearest to issue: bench, ftc, greens, negotiated, scenario, hasdai, way, timely,
Nearest to behind: breech, right, left, pollinators, chihuahua, deposition, taller, formation,
Epoch 10/10 Iteration: 46100 Avg. Training loss: 3.8408 0.1667 sec/batch
Epoch 10/10 Iteration: 46200 Avg. Training loss: 3.8104 0.1664 sec/batch

Restore the trained network if you need to:



In [37]:

    
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 [38]:

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

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



In [39]:

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



In [40]:

    
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 [ ]: