02a_rnn_spacy_wikitext103_spec.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from fastai.vision.all import *\n",
    "from fastai.learner import *\n",
    "from fastai.data.all import *\n",
    "from fastai.callback.tracker import SaveModelCallback\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "from pathlib2 import Path\n",
    "import numpy as np\n",
    "import random\n",
    "from torch.nn import MSELoss\n",
    "from IPython.lib.display import Audio\n",
    "from python_speech_features.base import mfcc\n",
    "import librosa"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_train = pd.read_csv('data/wikitext-103/train.csv', header=None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(29539, 1)"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "texts = df_train[0].tolist()\n",
    "texts = texts[:2000]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/conda/lib/python3.7/site-packages/sklearn/utils/deprecation.py:143: FutureWarning: The sklearn.datasets.base module is  deprecated in version 0.22 and will be removed in version 0.24. The corresponding classes / functions should instead be imported from sklearn.datasets. Anything that cannot be imported from sklearn.datasets is now part of the private API.\n",
      "  warnings.warn(message, FutureWarning)\n"
     ]
    }
   ],
   "source": [
    "import logging\n",
    "from six import iteritems\n",
    "from web.datasets.similarity import fetch_MEN, fetch_WS353, fetch_SimLex999\n",
    "from web.embeddings import fetch_GloVe\n",
    "from web.evaluate import evaluate_similarity\n",
    "from web.embedding import Embedding, Vocabulary\n",
    "from gensim.models import Word2Vec\n",
    "from gensim.models import KeyedVectors"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "37622"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "speech2vec = KeyedVectors.load_word2vec_format('../speech2vec-pretrained-vectors/speech2vec/50.vec', binary=False) \n",
    "\n",
    "vocab = set(speech2vec.vocab.keys())\n",
    "len(vocab)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# mkdir -p data/audio"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# from time import sleep"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# %%time\n",
    "\n",
    "# from google.cloud import texttospeech\n",
    "\n",
    "# client = texttospeech.TextToSpeechClient()\n",
    "\n",
    "# for i, word in enumerate(vocab):\n",
    "#     if i % 300 == 0: sleep(60)\n",
    "\n",
    "#     synthesis_input = texttospeech.SynthesisInput(text=word)\n",
    "\n",
    "#     voice = texttospeech.VoiceSelectionParams(\n",
    "#         language_code=\"en-US\", ssml_gender=texttospeech.SsmlVoiceGender.NEUTRAL\n",
    "#     )\n",
    "\n",
    "#     audio_config = texttospeech.AudioConfig(\n",
    "#         audio_encoding=texttospeech.AudioEncoding.MP3\n",
    "#     )\n",
    "\n",
    "#     response = client.synthesize_speech(\n",
    "#         input=synthesis_input, voice=voice, audio_config=audio_config\n",
    "#     )\n",
    "\n",
    "#     # The response's audio_content is binary.\n",
    "#     with open(f\"data/audio/{word}.mp3\", \"wb\") as out:\n",
    "#         out.write(response.audio_content)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "word2ary = pd.read_pickle('word2ary.pkl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "SAMPLE_RATE = 16000\n",
    "\n",
    "def audio_to_spec(audio):\n",
    "    f, t, spec = scipy.signal.spectrogram(audio, fs=SAMPLE_RATE)#, nperseg=360)\n",
    "    spec = np.log10(spec.clip(1e-10))\n",
    "    return spec[10:100].T\n",
    "    spec = librosa.power_to_db(\n",
    "#         librosa.feature.melspectrogram(audio, sr=SAMPLE_RATE, fmin=20, fmax=16000, n_mels=224, hop_length=360)\n",
    "        librosa.feature.melspectrogram(audio, sr=SAMPLE_RATE, fmin=20, fmax=16000, n_mels=128)\n",
    "#         librosa.feature.melspectrogram(audio, sr=SAMPLE_RATE, hop_length=1255, fmin=20, fmax=16000)\n",
    "    )\n",
    "    return spec"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 21.6 s, sys: 620 ms, total: 22.2 s\n",
      "Wall time: 22.2 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "word2features = {}\n",
    "for k, v in word2ary.items():\n",
    "    word2features[k] = audio_to_spec(v)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "from joblib import Parallel, delayed\n",
    "from functools import partial\n",
    "from spacy.util import minibatch\n",
    "import spacy\n",
    "from spacy.attrs import ORTH, LEMMA\n",
    "\n",
    "nlp = spacy.load(\"en_core_web_sm\")\n",
    "\n",
    "nlp.tokenizer.add_special_case('<unk>', [{ORTH: \"<unk>\"}])\n",
    "nlp.tokenizer.add_special_case('<eos>', [{ORTH: \"<eos>\"}])\n",
    "\n",
    "batch_size= 200\n",
    "partitions = minibatch(texts, size=batch_size)\n",
    "\n",
    "executor = Parallel(n_jobs=10, backend=\"multiprocessing\", prefer=\"threads\")\n",
    "\n",
    "def transform_texts(nlp, vocab, text):\n",
    "    docs = nlp.tokenizer.pipe(text)\n",
    "    word_pairs = []\n",
    "    \n",
    "    for doc in docs:\n",
    "        words = [token.text.lower() for token in doc if token.is_alpha]\n",
    "        words = [w if w in vocab else '<UNK>' for w in words]\n",
    "        \n",
    "        for i, word in enumerate(words):\n",
    "            if word == '<UNK>': continue\n",
    "            for offset in [-2, -1, 1, 2]:\n",
    "                if i + offset < 0 or i + offset >= len(words): continue\n",
    "                target_word = words[i+offset]\n",
    "                if target_word == '<UNK>': continue\n",
    "                word_pairs.append([word, target_word])\n",
    "                 \n",
    "    return word_pairs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 17.4 s, sys: 3.9 s, total: 21.3 s\n",
      "Wall time: 26.2 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "do = delayed(partial(transform_texts, nlp, vocab))\n",
    "tasks = (do(batch) for batch in partitions)\n",
    "word_pairs_hierarchical = executor(tasks)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "word_pairs = []\n",
    "\n",
    "for pairs in word_pairs_hierarchical: word_pairs += pairs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[['is', 'a'],\n",
       " ['a', 'is'],\n",
       " ['a', 'genus'],\n",
       " ['genus', 'a'],\n",
       " ['genus', 'in'],\n",
       " ['genus', 'the'],\n",
       " ['in', 'genus'],\n",
       " ['in', 'the'],\n",
       " ['in', 'family'],\n",
       " ['the', 'genus']]"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "word_pairs[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "18441070"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(word_pairs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "18441070"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(word_pairs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "vocab = list(vocab)\n",
    "word2index = {w: i for i, w in enumerate(vocab)}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "features = []\n",
    "\n",
    "for k, v in word2features.items():\n",
    "    features.append(v)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "155"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "max(f.shape[0] for f in features)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "def prepare_features(word, pad_to=max(f.shape[0] for f in features), pad_left=False):\n",
    "    ary = word2features[word][:pad_to]\n",
    "    example = np.zeros((pad_to, 90))\n",
    "    if pad_left:\n",
    "        example[-ary.shape[0]:, :] = ary\n",
    "    else: example[:ary.shape[0], :] = ary\n",
    "    return example.astype(np.float32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset_mean = np.concatenate(features).mean()\n",
    "dataset_std = np.concatenate(features).std()\n",
    "\n",
    "def normalize_data(ary):\n",
    "    return (ary - dataset_mean) / dataset_std"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "fn2features = {}\n",
    "for word in vocab:\n",
    "    fn2features[word] = normalize_data(prepare_features(word))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Dataset():\n",
    "    def __init__(self, pairs, length=10_000_000):\n",
    "        self.pairs = pairs\n",
    "        self.len = length\n",
    "    def __len__(self):\n",
    "        return self.len\n",
    "    def __getitem__(self, idx):\n",
    "        source_word, target_word = self.pairs[np.random.randint(len(self.pairs))]\n",
    "        return fn2features[source_word], word2index[target_word]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "# 14 mln -> 180 mln"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "18441070"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(word_pairs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_ds = Dataset(word_pairs)\n",
    "valid_ds = Dataset(word_pairs, 1_000_000)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(10000000, 1000000)"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(train_ds), len(valid_ds)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "BS = 2048\n",
    "NUM_WORKERS = 8\n",
    "\n",
    "train_dl = DataLoader(train_ds, BS, NUM_WORKERS, shuffle=True)\n",
    "valid_dl = DataLoader(valid_ds, BS, NUM_WORKERS)\n",
    "\n",
    "dls = DataLoaders(train_dl, valid_dl)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Model(Module):\n",
    "    def __init__(self, hidden_size=25):\n",
    "        self.return_embeddings = False\n",
    "        self.hidden_size = hidden_size\n",
    "        \n",
    "        self.encoder= nn.LSTM(\n",
    "            input_size=90,\n",
    "            hidden_size=hidden_size,\n",
    "            num_layers=3,\n",
    "            batch_first=True,\n",
    "            dropout=0,\n",
    "            bidirectional=True\n",
    "        )\n",
    "        \n",
    "        self.linear = nn.Linear(2*hidden_size, len(vocab))\n",
    "            \n",
    "    def forward(self, source_features):\n",
    "        _, (embeddings, _) = self.encoder(source_features)        \n",
    "        embeddings = torch.cat((embeddings[-1], embeddings[-2]), 1)\n",
    "        \n",
    "        if self.return_embeddings: return embeddings\n",
    "\n",
    "        return self.linear(embeddings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn = Learner(\n",
    "    dls.cuda(),\n",
    "    Model().cuda(),\n",
    "    loss_func=CrossEntropyLossFlat(),\n",
    "    opt_func=Adam,\n",
    "    metrics=[accuracy]\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>6.932520</td>\n",
       "      <td>6.935555</td>\n",
       "      <td>0.094152</td>\n",
       "      <td>11:35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>6.888471</td>\n",
       "      <td>6.893290</td>\n",
       "      <td>0.095840</td>\n",
       "      <td>11:35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>6.854970</td>\n",
       "      <td>6.863441</td>\n",
       "      <td>0.096903</td>\n",
       "      <td>11:36</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>6.849358</td>\n",
       "      <td>6.846374</td>\n",
       "      <td>0.096313</td>\n",
       "      <td>11:34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>6.835235</td>\n",
       "      <td>6.836201</td>\n",
       "      <td>0.096961</td>\n",
       "      <td>11:34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>6.823748</td>\n",
       "      <td>6.821783</td>\n",
       "      <td>0.097352</td>\n",
       "      <td>11:35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>6.823327</td>\n",
       "      <td>6.812018</td>\n",
       "      <td>0.097267</td>\n",
       "      <td>11:34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>6.804551</td>\n",
       "      <td>6.802811</td>\n",
       "      <td>0.097990</td>\n",
       "      <td>11:33</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>6.808207</td>\n",
       "      <td>6.794943</td>\n",
       "      <td>0.097674</td>\n",
       "      <td>11:32</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>6.803380</td>\n",
       "      <td>6.793743</td>\n",
       "      <td>0.097508</td>\n",
       "      <td>11:34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>10</td>\n",
       "      <td>6.788463</td>\n",
       "      <td>6.792137</td>\n",
       "      <td>0.097623</td>\n",
       "      <td>11:34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>11</td>\n",
       "      <td>6.780608</td>\n",
       "      <td>6.781083</td>\n",
       "      <td>0.098097</td>\n",
       "      <td>11:39</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12</td>\n",
       "      <td>6.780528</td>\n",
       "      <td>6.776615</td>\n",
       "      <td>0.098284</td>\n",
       "      <td>11:39</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>13</td>\n",
       "      <td>6.778561</td>\n",
       "      <td>6.779403</td>\n",
       "      <td>0.097588</td>\n",
       "      <td>11:35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>14</td>\n",
       "      <td>6.786719</td>\n",
       "      <td>6.775402</td>\n",
       "      <td>0.097507</td>\n",
       "      <td>11:35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>15</td>\n",
       "      <td>6.769154</td>\n",
       "      <td>6.776482</td>\n",
       "      <td>0.097395</td>\n",
       "      <td>11:35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>16</td>\n",
       "      <td>6.768485</td>\n",
       "      <td>6.766014</td>\n",
       "      <td>0.097987</td>\n",
       "      <td>11:34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>17</td>\n",
       "      <td>6.772653</td>\n",
       "      <td>6.764089</td>\n",
       "      <td>0.097657</td>\n",
       "      <td>11:34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>18</td>\n",
       "      <td>6.749450</td>\n",
       "      <td>6.773117</td>\n",
       "      <td>0.097470</td>\n",
       "      <td>11:34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>19</td>\n",
       "      <td>6.772224</td>\n",
       "      <td>6.765487</td>\n",
       "      <td>0.097501</td>\n",
       "      <td>11:33</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>20</td>\n",
       "      <td>6.775717</td>\n",
       "      <td>6.770425</td>\n",
       "      <td>0.097256</td>\n",
       "      <td>11:36</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>21</td>\n",
       "      <td>6.764843</td>\n",
       "      <td>6.760496</td>\n",
       "      <td>0.097896</td>\n",
       "      <td>11:36</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>22</td>\n",
       "      <td>6.761556</td>\n",
       "      <td>6.756197</td>\n",
       "      <td>0.098036</td>\n",
       "      <td>11:36</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>23</td>\n",
       "      <td>6.765841</td>\n",
       "      <td>6.758107</td>\n",
       "      <td>0.098047</td>\n",
       "      <td>11:34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>24</td>\n",
       "      <td>6.756263</td>\n",
       "      <td>6.759020</td>\n",
       "      <td>0.097894</td>\n",
       "      <td>11:35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>25</td>\n",
       "      <td>6.755969</td>\n",
       "      <td>6.757759</td>\n",
       "      <td>0.097642</td>\n",
       "      <td>11:30</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>26</td>\n",
       "      <td>6.764309</td>\n",
       "      <td>6.766229</td>\n",
       "      <td>0.097557</td>\n",
       "      <td>11:44</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>27</td>\n",
       "      <td>6.746200</td>\n",
       "      <td>6.758292</td>\n",
       "      <td>0.097947</td>\n",
       "      <td>11:44</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>28</td>\n",
       "      <td>6.758583</td>\n",
       "      <td>6.752354</td>\n",
       "      <td>0.098481</td>\n",
       "      <td>11:41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>29</td>\n",
       "      <td>6.739727</td>\n",
       "      <td>6.749059</td>\n",
       "      <td>0.098315</td>\n",
       "      <td>11:51</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>30</td>\n",
       "      <td>6.747636</td>\n",
       "      <td>6.750079</td>\n",
       "      <td>0.097704</td>\n",
       "      <td>11:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>31</td>\n",
       "      <td>6.767912</td>\n",
       "      <td>6.748539</td>\n",
       "      <td>0.098389</td>\n",
       "      <td>11:48</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>32</td>\n",
       "      <td>6.743858</td>\n",
       "      <td>6.748840</td>\n",
       "      <td>0.097771</td>\n",
       "      <td>11:49</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>33</td>\n",
       "      <td>6.741860</td>\n",
       "      <td>6.742044</td>\n",
       "      <td>0.098541</td>\n",
       "      <td>11:51</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>34</td>\n",
       "      <td>6.734061</td>\n",
       "      <td>6.747488</td>\n",
       "      <td>0.098578</td>\n",
       "      <td>11:41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>35</td>\n",
       "      <td>6.737804</td>\n",
       "      <td>6.746786</td>\n",
       "      <td>0.098921</td>\n",
       "      <td>11:42</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>36</td>\n",
       "      <td>6.746954</td>\n",
       "      <td>6.742561</td>\n",
       "      <td>0.098155</td>\n",
       "      <td>11:44</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>37</td>\n",
       "      <td>6.749112</td>\n",
       "      <td>6.742377</td>\n",
       "      <td>0.098585</td>\n",
       "      <td>11:44</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>38</td>\n",
       "      <td>6.741895</td>\n",
       "      <td>6.739891</td>\n",
       "      <td>0.097790</td>\n",
       "      <td>11:43</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>39</td>\n",
       "      <td>6.736873</td>\n",
       "      <td>6.739872</td>\n",
       "      <td>0.098083</td>\n",
       "      <td>11:42</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>40</td>\n",
       "      <td>6.726790</td>\n",
       "      <td>6.736025</td>\n",
       "      <td>0.098904</td>\n",
       "      <td>11:39</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>41</td>\n",
       "      <td>6.744344</td>\n",
       "      <td>6.733075</td>\n",
       "      <td>0.098495</td>\n",
       "      <td>11:37</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>42</td>\n",
       "      <td>6.726522</td>\n",
       "      <td>6.736481</td>\n",
       "      <td>0.098574</td>\n",
       "      <td>11:37</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>43</td>\n",
       "      <td>6.735545</td>\n",
       "      <td>6.732143</td>\n",
       "      <td>0.098905</td>\n",
       "      <td>11:38</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>44</td>\n",
       "      <td>6.719336</td>\n",
       "      <td>6.736372</td>\n",
       "      <td>0.098005</td>\n",
       "      <td>11:37</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>45</td>\n",
       "      <td>6.728365</td>\n",
       "      <td>6.732464</td>\n",
       "      <td>0.098358</td>\n",
       "      <td>11:41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>46</td>\n",
       "      <td>6.727263</td>\n",
       "      <td>6.727840</td>\n",
       "      <td>0.098293</td>\n",
       "      <td>11:38</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>47</td>\n",
       "      <td>6.729787</td>\n",
       "      <td>6.736483</td>\n",
       "      <td>0.097817</td>\n",
       "      <td>11:39</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>48</td>\n",
       "      <td>6.730904</td>\n",
       "      <td>6.731863</td>\n",
       "      <td>0.098600</td>\n",
       "      <td>11:40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>49</td>\n",
       "      <td>6.727291</td>\n",
       "      <td>6.728088</td>\n",
       "      <td>0.098710</td>\n",
       "      <td>11:34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>50</td>\n",
       "      <td>6.740599</td>\n",
       "      <td>6.726965</td>\n",
       "      <td>0.098336</td>\n",
       "      <td>11:27</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>51</td>\n",
       "      <td>6.735195</td>\n",
       "      <td>6.723128</td>\n",
       "      <td>0.099148</td>\n",
       "      <td>11:24</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>52</td>\n",
       "      <td>6.730174</td>\n",
       "      <td>6.727301</td>\n",
       "      <td>0.098632</td>\n",
       "      <td>11:35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>53</td>\n",
       "      <td>6.733843</td>\n",
       "      <td>6.723759</td>\n",
       "      <td>0.098452</td>\n",
       "      <td>11:34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>54</td>\n",
       "      <td>6.722220</td>\n",
       "      <td>6.726471</td>\n",
       "      <td>0.098173</td>\n",
       "      <td>11:27</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>55</td>\n",
       "      <td>6.721390</td>\n",
       "      <td>6.732336</td>\n",
       "      <td>0.097903</td>\n",
       "      <td>11:30</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>56</td>\n",
       "      <td>6.725743</td>\n",
       "      <td>6.726742</td>\n",
       "      <td>0.097943</td>\n",
       "      <td>11:23</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>57</td>\n",
       "      <td>6.716660</td>\n",
       "      <td>6.726186</td>\n",
       "      <td>0.098407</td>\n",
       "      <td>11:44</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>58</td>\n",
       "      <td>6.725584</td>\n",
       "      <td>6.721332</td>\n",
       "      <td>0.098863</td>\n",
       "      <td>11:48</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>59</td>\n",
       "      <td>6.726845</td>\n",
       "      <td>6.721571</td>\n",
       "      <td>0.098515</td>\n",
       "      <td>11:41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>60</td>\n",
       "      <td>6.723654</td>\n",
       "      <td>6.720761</td>\n",
       "      <td>0.098863</td>\n",
       "      <td>11:40</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>61</td>\n",
       "      <td>6.720410</td>\n",
       "      <td>6.725514</td>\n",
       "      <td>0.098571</td>\n",
       "      <td>11:34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>62</td>\n",
       "      <td>6.729940</td>\n",
       "      <td>6.724699</td>\n",
       "      <td>0.098185</td>\n",
       "      <td>11:27</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>63</td>\n",
       "      <td>6.716588</td>\n",
       "      <td>6.718497</td>\n",
       "      <td>0.099015</td>\n",
       "      <td>11:26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>64</td>\n",
       "      <td>6.725656</td>\n",
       "      <td>6.783796</td>\n",
       "      <td>0.096920</td>\n",
       "      <td>11:24</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>65</td>\n",
       "      <td>6.722355</td>\n",
       "      <td>6.722536</td>\n",
       "      <td>0.098877</td>\n",
       "      <td>11:26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>66</td>\n",
       "      <td>6.726385</td>\n",
       "      <td>6.719967</td>\n",
       "      <td>0.098870</td>\n",
       "      <td>11:29</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>67</td>\n",
       "      <td>6.715254</td>\n",
       "      <td>6.713105</td>\n",
       "      <td>0.098874</td>\n",
       "      <td>11:28</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>68</td>\n",
       "      <td>6.713314</td>\n",
       "      <td>6.717456</td>\n",
       "      <td>0.098741</td>\n",
       "      <td>11:27</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>69</td>\n",
       "      <td>6.732337</td>\n",
       "      <td>6.732457</td>\n",
       "      <td>0.098332</td>\n",
       "      <td>11:27</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>70</td>\n",
       "      <td>6.713326</td>\n",
       "      <td>6.715087</td>\n",
       "      <td>0.099259</td>\n",
       "      <td>11:28</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>71</td>\n",
       "      <td>6.718836</td>\n",
       "      <td>6.719411</td>\n",
       "      <td>0.098556</td>\n",
       "      <td>11:28</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>72</td>\n",
       "      <td>6.707809</td>\n",
       "      <td>6.720122</td>\n",
       "      <td>0.098069</td>\n",
       "      <td>11:28</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>73</td>\n",
       "      <td>6.722353</td>\n",
       "      <td>6.722435</td>\n",
       "      <td>0.098818</td>\n",
       "      <td>11:28</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>74</td>\n",
       "      <td>6.709846</td>\n",
       "      <td>6.717856</td>\n",
       "      <td>0.098809</td>\n",
       "      <td>11:42</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>75</td>\n",
       "      <td>6.720128</td>\n",
       "      <td>6.713192</td>\n",
       "      <td>0.099200</td>\n",
       "      <td>11:54</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>76</td>\n",
       "      <td>6.718256</td>\n",
       "      <td>6.720899</td>\n",
       "      <td>0.097941</td>\n",
       "      <td>11:48</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>77</td>\n",
       "      <td>6.722522</td>\n",
       "      <td>6.722878</td>\n",
       "      <td>0.098244</td>\n",
       "      <td>11:48</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>78</td>\n",
       "      <td>6.716048</td>\n",
       "      <td>6.718272</td>\n",
       "      <td>0.098328</td>\n",
       "      <td>11:51</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>79</td>\n",
       "      <td>6.717943</td>\n",
       "      <td>6.716679</td>\n",
       "      <td>0.098380</td>\n",
       "      <td>11:52</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>80</td>\n",
       "      <td>6.707885</td>\n",
       "      <td>6.713239</td>\n",
       "      <td>0.098612</td>\n",
       "      <td>11:48</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>81</td>\n",
       "      <td>6.720417</td>\n",
       "      <td>6.711720</td>\n",
       "      <td>0.098956</td>\n",
       "      <td>11:51</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>82</td>\n",
       "      <td>6.713458</td>\n",
       "      <td>6.718053</td>\n",
       "      <td>0.099117</td>\n",
       "      <td>11:51</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>83</td>\n",
       "      <td>6.714202</td>\n",
       "      <td>6.714656</td>\n",
       "      <td>0.098900</td>\n",
       "      <td>11:51</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>84</td>\n",
       "      <td>6.718156</td>\n",
       "      <td>6.713133</td>\n",
       "      <td>0.098459</td>\n",
       "      <td>11:49</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>85</td>\n",
       "      <td>6.705763</td>\n",
       "      <td>6.715020</td>\n",
       "      <td>0.098303</td>\n",
       "      <td>11:47</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>86</td>\n",
       "      <td>6.705877</td>\n",
       "      <td>6.718959</td>\n",
       "      <td>0.098744</td>\n",
       "      <td>11:53</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>87</td>\n",
       "      <td>6.701169</td>\n",
       "      <td>6.712381</td>\n",
       "      <td>0.098708</td>\n",
       "      <td>11:41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>88</td>\n",
       "      <td>6.716999</td>\n",
       "      <td>6.710381</td>\n",
       "      <td>0.099041</td>\n",
       "      <td>11:51</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>89</td>\n",
       "      <td>6.705362</td>\n",
       "      <td>6.713930</td>\n",
       "      <td>0.098738</td>\n",
       "      <td>11:47</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>90</td>\n",
       "      <td>6.710818</td>\n",
       "      <td>6.714345</td>\n",
       "      <td>0.098353</td>\n",
       "      <td>11:47</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>91</td>\n",
       "      <td>6.704442</td>\n",
       "      <td>6.738797</td>\n",
       "      <td>0.097481</td>\n",
       "      <td>11:46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>92</td>\n",
       "      <td>6.724284</td>\n",
       "      <td>6.715572</td>\n",
       "      <td>0.098733</td>\n",
       "      <td>11:48</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>93</td>\n",
       "      <td>6.717839</td>\n",
       "      <td>6.719276</td>\n",
       "      <td>0.098530</td>\n",
       "      <td>11:47</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>94</td>\n",
       "      <td>6.711279</td>\n",
       "      <td>6.711872</td>\n",
       "      <td>0.098751</td>\n",
       "      <td>11:39</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>95</td>\n",
       "      <td>6.713540</td>\n",
       "      <td>6.712434</td>\n",
       "      <td>0.098706</td>\n",
       "      <td>11:41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>96</td>\n",
       "      <td>6.715127</td>\n",
       "      <td>6.709778</td>\n",
       "      <td>0.098659</td>\n",
       "      <td>11:44</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>97</td>\n",
       "      <td>6.704834</td>\n",
       "      <td>6.709046</td>\n",
       "      <td>0.098440</td>\n",
       "      <td>11:42</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>98</td>\n",
       "      <td>6.714187</td>\n",
       "      <td>6.713832</td>\n",
       "      <td>0.098918</td>\n",
       "      <td>11:41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>99</td>\n",
       "      <td>6.724922</td>\n",
       "      <td>6.710684</td>\n",
       "      <td>0.098360</td>\n",
       "      <td>11:38</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "NUM_EPOCHS = 100\n",
    "learn.fit(NUM_EPOCHS, lr=1e-3, cbs=SaveModelCallback(fname='rnn_encoder_spec', every_epoch=True))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Evalute embeddings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "from utils import Embeddings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "class DatasetPred():\n",
    "    def __len__(self):\n",
    "        return len(vocab)\n",
    "    def __getitem__(self, idx):\n",
    "        _, target_word = word_pairs[np.random.randint(len(word_pairs))]\n",
    "        source_word = vocab[idx]\n",
    "        return fn2features[source_word], word2index[target_word]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "pred_dl = DataLoader(DatasetPred(), BS, NUM_WORKERS)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 960 ms, sys: 1.71 s, total: 2.67 s\n",
      "Wall time: 3.17 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "learn.model.return_embeddings = True\n",
    "learn.model.eval()\n",
    "\n",
    "all_embeddings = []\n",
    "\n",
    "with torch.no_grad():    \n",
    "    for batch in pred_dl:\n",
    "        embeddings = learn.model(batch[0].cuda())\n",
    "        all_embeddings.append(embeddings.detach().cpu().squeeze(0))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "embeddings = torch.cat(all_embeddings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "e = Embeddings(\n",
    "    embeddings,\n",
    "    vocab\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for w in ['fast', 'lost', 'small', 'true', 'crazy', 'slow']:\n",
    "    print(f'{w}: {e.nn_words_to(e[w])}')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Evaluating embeddings using [word-embeddings-benchmarks](https://github.com/kudkudak/word-embeddings-benchmarks)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "import logging\n",
    "from six import iteritems\n",
    "from web.datasets.similarity import fetch_MEN, fetch_WS353, fetch_SimLex999\n",
    "from web.embeddings import fetch_GloVe\n",
    "from web.evaluate import evaluate_similarity\n",
    "from web.embedding import Embedding, Vocabulary\n",
    "from gensim.models import Word2Vec\n",
    "from gensim.models import KeyedVectors"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "tasks = {\n",
    "    \"MEN\": fetch_MEN(),\n",
    "    \"WS353\": fetch_WS353(),\n",
    "    \"SIMLEX999\": fetch_SimLex999()\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "our_embeddings = Embedding(\n",
    "    Vocabulary(vocab),\n",
    "    embeddings\n",
    ")\n",
    "\n",
    "speech2vec = KeyedVectors.load_word2vec_format('../speech2vec-pretrained-vectors/word2vec/50.vec', binary=False) \n",
    "speech2vec_embeddings = Embedding(Vocabulary(list(speech2vec.vocab.keys())), speech2vec.vectors)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Missing 392 words. Will replace them with mean vector\n",
      "/opt/conda/lib/python3.7/site-packages/web-0.0.1-py3.7.egg/web/evaluate.py:336: FutureWarning: arrays to stack must be passed as a \"sequence\" type such as list or tuple. Support for non-sequence iterables such as generators is deprecated as of NumPy 1.16 and will raise an error in the future.\n",
      "  A = np.vstack(w.get(word, mean_vector) for word in X[:, 0])\n",
      "/opt/conda/lib/python3.7/site-packages/web-0.0.1-py3.7.egg/web/evaluate.py:337: FutureWarning: arrays to stack must be passed as a \"sequence\" type such as list or tuple. Support for non-sequence iterables such as generators is deprecated as of NumPy 1.16 and will raise an error in the future.\n",
      "  B = np.vstack(w.get(word, mean_vector) for word in X[:, 1])\n",
      "Missing 61 words. Will replace them with mean vector\n",
      "Missing 24 words. Will replace them with mean vector\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Spearman correlation of scores on MEN -0.02711111464006222\n",
      "Spearman correlation of scores on WS353 0.15145529914207986\n",
      "Spearman correlation of scores on SIMLEX999 0.0293231777105999\n"
     ]
    }
   ],
   "source": [
    "for name, data in iteritems(tasks):\n",
    "    print(\"Spearman correlation of scores on {} {}\".format(name, evaluate_similarity(our_embeddings, data.X, data.y)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Missing 392 words. Will replace them with mean vector\n",
      "Missing 61 words. Will replace them with mean vector\n",
      "Missing 24 words. Will replace them with mean vector\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Spearman correlation of scores on MEN 0.6056592803599269\n",
      "Spearman correlation of scores on WS353 0.43349390636024643\n",
      "Spearman correlation of scores on SIMLEX999 0.25938770901422736\n"
     ]
    }
   ],
   "source": [
    "for name, data in iteritems(tasks):\n",
    "    print(\"Spearman correlation of scores on {} {}\".format(name, evaluate_similarity(speech2vec_embeddings, data.X, data.y)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Loss decrease and improvements on semantic tasks as training progresses"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%capture\n",
    "\n",
    "val_losses, accuracies, task_perf = [], [], []\n",
    "for i in range(NUM_EPOCHS):\n",
    "    learn.load(f'rnn_encoder_spec_{i}')\n",
    "#     loss, accuracy = learn.validate()\n",
    "#     val_losses.append(loss)\n",
    "#     accuracies.append(accuracy)\n",
    "    \n",
    "    learn.model.return_embeddings = True\n",
    "    learn.model.eval()\n",
    "\n",
    "    all_embeddings = []\n",
    "\n",
    "    with torch.no_grad():    \n",
    "        for batch in pred_dl:\n",
    "            embeddings = learn.model(batch[0].cuda())\n",
    "            all_embeddings.append(embeddings.detach().cpu().squeeze(0))\n",
    "\n",
    "    embeddings = torch.cat(all_embeddings)\n",
    "    our_embeddings = Embedding(\n",
    "        Vocabulary([w.lower() for w in vocab]),\n",
    "        embeddings\n",
    "    )\n",
    "\n",
    "    task_perf.append([evaluate_similarity(our_embeddings, data.X, data.y) for name, data in iteritems(tasks)])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "men, ws353, simlex999 = list(zip(*task_perf))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "No handles with labels found to put in legend.\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax1 = plt.subplots()\n",
    "\n",
    "# ax1.plot(val_losses, label='val loss')\n",
    "\n",
    "ax2 = ax1.twinx()\n",
    "ax2.plot(men, label='MEN', c='g')\n",
    "ax2.plot(ws353, label='WS353', c='m')\n",
    "ax2.plot(simlex999, label='SIMLEX999', c='y')\n",
    "\n",
    "ax1.legend(loc=[0.07, 0.9])\n",
    "ax2.legend(loc=[0.7, 0.15])\n",
    "\n",
    "ax1.set_xlabel('epochs');"
   ]
  }
 ],
 "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": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}