word2vec.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/anaconda3/lib/python3.6/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n"
     ]
    }
   ],
   "source": [
    "import urllib.request\n",
    "import collections\n",
    "import math\n",
    "import os\n",
    "import random\n",
    "import zipfile\n",
    "import datetime as dt\n",
    "\n",
    "import numpy as np\n",
    "import tensorflow as tf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def maybe_download(filename, url, expected_bytes):\n",
    "    \"\"\"Download a file if not present, and make sure it's the right size.\"\"\"\n",
    "    if not os.path.exists(filename):\n",
    "        filename, _ = urllib.request.urlretrieve(url + filename, filename)\n",
    "    statinfo = os.stat(filename)\n",
    "    if statinfo.st_size == expected_bytes:\n",
    "        print('Found and verified', filename)\n",
    "    else:\n",
    "        print(statinfo.st_size)\n",
    "        raise Exception(\n",
    "            'Failed to verify ' + filename + '. Can you get to it with a browser?')\n",
    "    return filename"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Read the data into a list of strings.\n",
    "def read_data(filename):\n",
    "    \"\"\"Extract the first file enclosed in a zip file as a list of words.\"\"\"\n",
    "    with zipfile.ZipFile(filename) as f:\n",
    "        data = tf.compat.as_str(f.read(f.namelist()[0])).split()\n",
    "    return data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def build_dataset(words, n_words):\n",
    "    \"\"\"Process raw inputs into a dataset.\"\"\"\n",
    "    count = [['UNK', -1]]\n",
    "    count.extend(collections.Counter(words).most_common(n_words - 1))\n",
    "    dictionary = dict()\n",
    "    for word, _ in count:\n",
    "        dictionary[word] = len(dictionary)\n",
    "    data = list()\n",
    "    unk_count = 0\n",
    "    for word in words:\n",
    "        if word in dictionary:\n",
    "            index = dictionary[word]\n",
    "        else:\n",
    "            index = 0  # dictionary['UNK']\n",
    "            unk_count += 1\n",
    "        data.append(index)\n",
    "    count[0][1] = unk_count\n",
    "    reversed_dictionary = dict(zip(dictionary.values(), dictionary.keys()))\n",
    "    return data, count, dictionary, reversed_dictionary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def collect_data(vocabulary_size=10000):\n",
    "    url = 'http://mattmahoney.net/dc/'\n",
    "    filename = maybe_download('text8.zip', url, 31344016)\n",
    "    vocabulary = read_data(filename)\n",
    "    print(vocabulary[:7])\n",
    "    data, count, dictionary, reverse_dictionary = build_dataset(vocabulary,\n",
    "                                                                vocabulary_size)\n",
    "    #del vocabulary  # Hint to reduce memory.\n",
    "    return data, count, dictionary, reverse_dictionary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_index = 0\n",
    "# generate batch data\n",
    "def generate_batch(data, batch_size, num_skips, skip_window):\n",
    "    global data_index\n",
    "    assert batch_size % num_skips == 0\n",
    "    assert num_skips <= 2 * skip_window\n",
    "    batch = np.ndarray(shape=(batch_size), dtype=np.int32)\n",
    "    context = np.ndarray(shape=(batch_size, 1), dtype=np.int32)\n",
    "    span = 2 * skip_window + 1  # [ skip_window input_word skip_window ]\n",
    "    buffer = collections.deque(maxlen=span)\n",
    "    for _ in range(span):\n",
    "        buffer.append(data[data_index])\n",
    "        data_index = (data_index + 1) % len(data)\n",
    "    for i in range(batch_size // num_skips):\n",
    "        target = skip_window  # input word at the center of the buffer\n",
    "        targets_to_avoid = [skip_window]\n",
    "        for j in range(num_skips):\n",
    "            while target in targets_to_avoid:\n",
    "                target = random.randint(0, span - 1)\n",
    "            targets_to_avoid.append(target)\n",
    "            batch[i * num_skips + j] = buffer[skip_window]  # this is the input word\n",
    "            context[i * num_skips + j, 0] = buffer[target]  # these are the context words\n",
    "        buffer.append(data[data_index])\n",
    "        data_index = (data_index + 1) % len(data)\n",
    "    # Backtrack a little bit to avoid skipping words in the end of a batch\n",
    "    data_index = (data_index + len(data) - span) % len(data)\n",
    "    return batch, context"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found and verified text8.zip\n",
      "['anarchism', 'originated', 'as', 'a', 'term', 'of', 'abuse']\n"
     ]
    }
   ],
   "source": [
    "vocabulary_size = 10000\n",
    "data, count, dictionary, reverse_dictionary = collect_data(vocabulary_size=vocabulary_size)\n",
    "\n",
    "batch_size = 128\n",
    "embedding_size = 300  # Dimension of the embedding vector.\n",
    "skip_window = 2       # How many words to consider left and right.\n",
    "num_skips = 2         # How many times to reuse an input to generate a label.\n",
    "\n",
    "# We pick a random validation set to sample nearest neighbors. Here we limit the\n",
    "# validation samples to the words that have a low numeric ID, which by\n",
    "# construction are also the most frequent.\n",
    "valid_size = 16     # Random set of words to evaluate similarity on.\n",
    "valid_window = 100  # Only pick dev samples in the head of the distribution.\n",
    "valid_examples = np.random.choice(valid_window, valid_size, replace=False)\n",
    "num_sampled = 64    # Number of negative examples to sample."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "ename": "IndexError",
     "evalue": "list index out of range",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mIndexError\u001b[0m                                Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-12-b53530325ad5>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mvocab\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mIndexError\u001b[0m: list index out of range"
     ]
    }
   ],
   "source": [
    "vocab[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "graph = tf.Graph()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From <ipython-input-9-4c48e404f389>:21: softmax_cross_entropy_with_logits (from tensorflow.python.ops.nn_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "\n",
      "Future major versions of TensorFlow will allow gradients to flow\n",
      "into the labels input on backprop by default.\n",
      "\n",
      "See @{tf.nn.softmax_cross_entropy_with_logits_v2}.\n",
      "\n",
      "WARNING:tensorflow:From <ipython-input-9-4c48e404f389>:27: calling reduce_sum (from tensorflow.python.ops.math_ops) with keep_dims is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "keep_dims is deprecated, use keepdims instead\n"
     ]
    }
   ],
   "source": [
    "with graph.as_default():\n",
    "    # Input data.\n",
    "    train_inputs = tf.placeholder(tf.int32, shape=[batch_size])\n",
    "    train_context = tf.placeholder(tf.int32, shape=[batch_size, 1])\n",
    "    valid_dataset = tf.constant(valid_examples, dtype=tf.int32)\n",
    "\n",
    "    # Look up embeddings for inputs.\n",
    "    embeddings = tf.Variable(\n",
    "        tf.random_uniform([vocabulary_size, embedding_size], -1.0, 1.0))\n",
    "    embed = tf.nn.embedding_lookup(embeddings, train_inputs)\n",
    "\n",
    "    # Construct the variables for the softmax\n",
    "    weights = tf.Variable(\n",
    "        tf.truncated_normal([embedding_size, vocabulary_size], stddev=1.0 / math.sqrt(embedding_size)))\n",
    "    biases = tf.Variable(tf.zeros([vocabulary_size]))\n",
    "    hidden_out = tf.transpose(tf.matmul(tf.transpose(weights), tf.transpose(embed))) + biases\n",
    "\n",
    "    # convert train_context to a one-hot format\n",
    "    train_one_hot = tf.one_hot(train_context, vocabulary_size)\n",
    "\n",
    "    cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=hidden_out, labels=train_one_hot))\n",
    "\n",
    "    # Construct the SGD optimizer using a learning rate of 1.0.\n",
    "    optimizer = tf.train.GradientDescentOptimizer(1.0).minimize(cross_entropy)\n",
    "\n",
    "    # Compute the cosine similarity between minibatch examples and all embeddings.\n",
    "    norm = tf.sqrt(tf.reduce_sum(tf.square(embeddings), 1, keep_dims=True))\n",
    "    normalized_embeddings = embeddings / norm\n",
    "    valid_embeddings = tf.nn.embedding_lookup(normalized_embeddings, valid_dataset)\n",
    "    similarity = tf.matmul(valid_embeddings, normalized_embeddings, transpose_b=True)\n",
    "\n",
    "    # Add variable initializer.\n",
    "    init = tf.global_variables_initializer()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def run(graph, num_steps):\n",
    "    with tf.Session(graph=graph) as session:\n",
    "        # We must initialize all variables before we use them.\n",
    "        init.run()\n",
    "        print('Initialized')\n",
    "\n",
    "        average_loss = 0\n",
    "        for step in range(num_steps):\n",
    "            batch_inputs, batch_context = generate_batch(data, batch_size, num_skips, skip_window)\n",
    "            feed_dict = {train_inputs: batch_inputs, train_context: batch_context}\n",
    "\n",
    "            # We perform one update step by evaluating the optimizer op (including it\n",
    "            # in the list of returned values for session.run()\n",
    "            _, loss_val = session.run([optimizer, cross_entropy], feed_dict=feed_dict)\n",
    "            average_loss += loss_val\n",
    "\n",
    "            if step % 2000 == 0:\n",
    "                if step > 0:\n",
    "                    average_loss /= 2000\n",
    "                # The average loss is an estimate of the loss over the last 2000 batches.\n",
    "                print('Average loss at step ', step, ': ', average_loss)\n",
    "                average_loss = 0\n",
    "\n",
    "            # Note that this is expensive (~20% slowdown if computed every 500 steps)\n",
    "            if step % 10000 == 0:\n",
    "                sim = similarity.eval()\n",
    "                for i in range(valid_size):\n",
    "                    valid_word = reverse_dictionary[valid_examples[i]]\n",
    "                    top_k = 8  # number of nearest neighbors\n",
    "                    nearest = (-sim[i, :]).argsort()[1:top_k + 1]\n",
    "                    log_str = 'Nearest to %s:' % valid_word\n",
    "                    for k in range(top_k):\n",
    "                        close_word = reverse_dictionary[nearest[k]]\n",
    "                        log_str = '%s %s,' % (log_str, close_word)\n",
    "                    print(log_str)\n",
    "        final_embeddings = normalized_embeddings.eval()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initialized\n",
      "Average loss at step  0 :  9.278728485107422\n",
      "Nearest to often: tape, octavian, promising, peru, risks, ever, manpower, grandson,\n",
      "Nearest to however: remote, guests, avoided, amplitude, higher, pradesh, fatty, amos,\n",
      "Nearest to been: admitted, anxiety, glass, limiting, program, consist, eucharist, bce,\n",
      "Nearest to the: fresh, modules, temples, keeping, theories, optimal, word, fires,\n",
      "Nearest to about: attacking, lifestyle, gifts, lighting, injury, strain, makers, rice,\n",
      "Nearest to this: johann, budget, revolutions, focuses, generations, vampire, south, madrid,\n",
      "Nearest to if: alfred, duration, worship, symphony, hearts, adults, openly, compiler,\n",
      "Nearest to people: ash, colors, fiscal, accounts, examination, descendant, landing, hotels,\n",
      "Nearest to in: missile, defeated, outdoor, losses, behind, observable, gel, done,\n",
      "Nearest to which: sri, issued, bytes, posthumous, demon, squadron, journey, contributions,\n",
      "Nearest to into: puzzle, proof, toronto, legislative, real, beautiful, wake, centred,\n",
      "Nearest to for: compared, leo, electricity, memories, anarcho, projection, elector, employed,\n",
      "Nearest to all: his, after, latest, ryan, quixote, kentucky, wings, wholly,\n",
      "Nearest to be: mercy, set, appear, vary, translate, difficult, invariably, gas,\n",
      "Nearest to UNK: academics, mutations, parish, replaced, cole, bbc, healthy, intermediate,\n",
      "Nearest to d: happens, jump, eiffel, image, ago, airways, game, fernando,\n",
      "Softmax method took 13.843023 seconds to run 100 iterations\n"
     ]
    }
   ],
   "source": [
    "num_steps = 100\n",
    "softmax_start_time = dt.datetime.now()\n",
    "run(graph, num_steps=num_steps)\n",
    "softmax_end_time = dt.datetime.now()\n",
    "print(\"Softmax method took {} seconds to run {} iterations\".format((softmax_end_time-softmax_start_time).total_seconds(),num_steps))\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "with graph.as_default():\n",
    "\n",
    "    # Construct the variables for the NCE loss\n",
    "    nce_weights = tf.Variable(\n",
    "        tf.truncated_normal([vocabulary_size, embedding_size],\n",
    "                            stddev=1.0 / math.sqrt(embedding_size)))\n",
    "    nce_biases = tf.Variable(tf.zeros([vocabulary_size]))\n",
    "\n",
    "    nce_loss = tf.reduce_mean(\n",
    "        tf.nn.nce_loss(weights=nce_weights,\n",
    "                       biases=nce_biases,\n",
    "                       labels=train_context,\n",
    "                       inputs=embed,\n",
    "                       num_sampled=num_sampled,\n",
    "                       num_classes=vocabulary_size))\n",
    "\n",
    "    optimizer = tf.train.GradientDescentOptimizer(1.0).minimize(nce_loss)\n",
    "\n",
    "    # Add variable initializer.\n",
    "    init = tf.global_variables_initializer()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initialized\n",
      "Average loss at step  0 :  9.40816879272461\n",
      "Nearest to often: camera, habsburg, frame, plane, effectively, earth, elements, jewish,\n",
      "Nearest to however: moroccan, invariant, medicine, frisian, championships, lady, exploration, membership,\n",
      "Nearest to been: networking, quantum, alphabet, waters, starred, ahmad, immense, colour,\n",
      "Nearest to the: broadway, paintings, strange, luke, waste, descended, agreed, ear,\n",
      "Nearest to about: derivative, boeing, claim, resurrection, print, feminine, velocity, eggs,\n",
      "Nearest to this: therapy, gulf, surrounding, myth, dependence, booth, partition, timor,\n",
      "Nearest to if: investigation, countryside, elephants, copenhagen, preserve, strategic, wanted, devices,\n",
      "Nearest to people: balls, lovers, nazis, didn, capitalist, inc, offshore, town,\n",
      "Nearest to in: tennis, heroin, afro, outward, feels, wolf, identity, must,\n",
      "Nearest to which: johnny, unite, dot, lady, ratio, highways, july, notes,\n",
      "Nearest to into: alter, correct, chandler, writings, neck, denial, disorders, ph,\n",
      "Nearest to for: engaged, knee, delivery, indonesia, stones, air, forming, color,\n",
      "Nearest to all: rn, inaccurate, government, neutron, disambiguation, gen, significance, ii,\n",
      "Nearest to be: pakistan, service, protests, kart, jump, sphere, ashes, age,\n",
      "Nearest to UNK: films, whenever, happy, barriers, containing, regulated, mathbf, genres,\n",
      "Nearest to d: four, claudius, atlanta, contain, rally, economist, fantasy, deaf,\n",
      "Average loss at step  2000 :  6.763666571855545\n",
      "Average loss at step  4000 :  6.322544091641903\n",
      "Average loss at step  6000 :  6.308463817000389\n",
      "Average loss at step  8000 :  6.23149065643549\n",
      "Average loss at step  10000 :  6.057535208702087\n",
      "Nearest to often: elements, camera, frame, habsburg, jewish, plane, effectively, burnt,\n",
      "Nearest to however: moroccan, invariant, championships, medicine, lady, frisian, riel, exploration,\n",
      "Nearest to been: networking, quantum, alphabet, ahmad, immense, waters, homepage, starred,\n",
      "Nearest to the: a, UNK, an, its, his, luke, daughter, graph,\n",
      "Nearest to about: boeing, derivative, print, velocity, bay, du, legs, motors,\n",
      "Nearest to this: gulf, therapy, surrounding, others, partition, dependence, laurent, funk,\n",
      "Nearest to if: investigation, copenhagen, countryside, preserve, elephants, devices, assets, strategic,\n",
      "Nearest to people: balls, lovers, area, didn, barely, capitalist, inc, nazis,\n",
      "Nearest to in: UNK, of, tennis, heroin, and, afro, feels, must,\n",
      "Nearest to which: johnny, dot, unite, lady, notes, honor, originally, ratio,\n",
      "Nearest to into: alter, correct, neck, writings, chandler, denial, rhythm, disorders,\n",
      "Nearest to for: lighter, taxation, knee, engaged, terrible, color, ash, held,\n",
      "Nearest to all: rn, government, neutron, ii, inaccurate, disambiguation, significance, dos,\n",
      "Nearest to be: service, protests, ashes, pakistan, jump, angry, kart, incorrect,\n",
      "Nearest to UNK: the, in, one, a, and, s, na, an,\n",
      "Nearest to d: four, claudius, atlanta, contain, rally, fantasy, external, economist,\n",
      "Average loss at step  12000 :  6.11270499509573\n",
      "Average loss at step  14000 :  6.022344716906548\n",
      "Average loss at step  16000 :  5.726902204543352\n",
      "Average loss at step  18000 :  5.910292772531509\n",
      "Average loss at step  20000 :  6.115611336708069\n",
      "Nearest to often: elements, frame, camera, habsburg, burnt, jewish, effectively, valves,\n",
      "Nearest to however: invariant, moroccan, championships, medicine, riel, lady, lines, bankruptcy,\n",
      "Nearest to been: networking, quantum, homepage, alphabet, rotation, waters, ahmad, immense,\n",
      "Nearest to the: a, UNK, one, its, his, an, luke, this,\n",
      "Nearest to about: print, boeing, derivative, motors, velocity, bay, consecutive, legs,\n",
      "Nearest to this: therapy, gulf, surrounding, the, others, partition, laurent, dependence,\n",
      "Nearest to if: investigation, copenhagen, countryside, preserve, wanted, devices, assets, strategic,\n",
      "Nearest to people: balls, town, lovers, didn, nazis, barely, capitalist, inc,\n",
      "Nearest to in: of, and, on, must, heroin, tennis, lawyers, battery,\n",
      "Nearest to which: johnny, dot, unite, lady, originally, boiling, honor, configuration,\n",
      "Nearest to into: alter, correct, neck, writings, rhythm, denial, chandler, disorders,\n",
      "Nearest to for: of, lighter, montenegro, taxation, forming, controversial, knee, engaged,\n",
      "Nearest to all: rn, government, disambiguation, neutron, ii, ham, image, abstract,\n",
      "Nearest to be: protests, service, ashes, jump, pakistan, kart, angry, follow,\n",
      "Nearest to UNK: the, one, a, s, module, apollo, accompanying, component,\n",
      "Nearest to d: b, claudius, atlanta, four, rally, contain, external, scored,\n",
      "Average loss at step  22000 :  6.084851184368134\n",
      "Average loss at step  24000 :  6.08506400346756\n",
      "Average loss at step  26000 :  6.007789438843727\n",
      "Average loss at step  28000 :  6.086256842017174\n",
      "Average loss at step  30000 :  5.9308228093385695\n",
      "Nearest to often: elements, burnt, frame, camera, jewish, effectively, plane, habsburg,\n",
      "Nearest to however: invariant, moroccan, riel, championships, medicine, and, bankruptcy, lines,\n",
      "Nearest to been: networking, quantum, homepage, rotation, cobain, waters, distinguished, immense,\n",
      "Nearest to the: a, its, an, UNK, this, one, focused, road,\n",
      "Nearest to about: motors, print, derivative, velocity, boeing, dc, consecutive, bay,\n",
      "Nearest to this: the, therapy, surrounding, gulf, others, partition, laurent, dependence,\n",
      "Nearest to if: investigation, preserve, countryside, copenhagen, devices, wanted, strategic, suffered,\n",
      "Nearest to people: balls, town, nazis, barely, didn, jim, lovers, capitalist,\n",
      "Nearest to in: of, on, lawyers, from, heroin, and, at, by,\n",
      "Nearest to which: johnny, dot, originally, cycles, lady, unite, that, boiling,\n",
      "Nearest to into: alter, correct, writings, denial, neck, rhythm, chandler, disorders,\n",
      "Nearest to for: of, controversial, montenegro, thai, lighter, to, indonesia, and,\n",
      "Nearest to all: rn, disambiguation, government, neutron, image, ii, ham, abstract,\n",
      "Nearest to be: is, protests, were, service, jump, follow, ashes, kart,\n",
      "Nearest to UNK: titles, the, it, jerome, isn, india, severe, declaration,\n",
      "Nearest to d: b, claudius, atlanta, four, rally, contain, external, deaf,\n",
      "Average loss at step  32000 :  6.203061672747135\n",
      "Average loss at step  34000 :  6.091743029475212\n",
      "Average loss at step  36000 :  6.132061564326286\n",
      "Average loss at step  38000 :  6.0893978720903394\n",
      "Average loss at step  40000 :  6.067196767568588\n",
      "Nearest to often: elements, burnt, camera, frame, effectively, jewish, valves, plane,\n",
      "Nearest to however: invariant, moroccan, riel, bankruptcy, championships, enormous, prominently, medicine,\n",
      "Nearest to been: networking, quantum, homepage, cobain, waters, distinguished, rotation, immense,\n",
      "Nearest to the: a, UNK, its, his, their, an, one, this,\n",
      "Nearest to about: motors, print, derivative, boeing, dc, velocity, du, bay,\n",
      "Nearest to this: surrounding, that, the, therapy, laurent, gulf, else, booth,\n",
      "Nearest to if: investigation, preserve, countryside, devices, wanted, copenhagen, passes, dome,\n",
      "Nearest to people: balls, nazis, barely, town, didn, lovers, jim, glass,\n",
      "Nearest to in: of, on, and, from, lawyers, at, tennis, by,\n",
      "Nearest to which: that, johnny, dot, lady, originally, cycles, boiling, also,\n",
      "Nearest to into: alter, correct, writings, denial, rhythm, neck, chandler, disorders,\n",
      "Nearest to for: of, indonesia, montenegro, to, thai, controversial, lighter, terrible,\n",
      "Nearest to all: disambiguation, rn, ham, pole, image, neutron, government, abstract,\n",
      "Nearest to be: is, were, protests, follow, refer, jump, have, ashes,\n",
      "Nearest to UNK: the, pressed, horizon, ep, seventh, communicate, synonymous, normal,\n",
      "Nearest to d: b, claudius, atlanta, rally, contain, external, deaf, scored,\n",
      "Average loss at step  42000 :  6.126774079203606\n",
      "Average loss at step  44000 :  6.101098069548607\n",
      "Average loss at step  46000 :  6.144429542601109\n",
      "Average loss at step  48000 :  6.105939457416534\n",
      "Softmax method took 13.843023 seconds to run 50000 iterations\n"
     ]
    }
   ],
   "source": [
    "num_steps = 50000\n",
    "nce_start_time = dt.datetime.now()\n",
    "run(graph, num_steps)\n",
    "nce_end_time = dt.datetime.now()\n",
    "print(\"Softmax method took {} seconds to run {} iterations\".format((softmax_end_time-softmax_start_time).total_seconds(),num_steps))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}