Comparing TF and PT models.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Comparing TensorFlow (original) and PyTorch models\n",
    "\n",
    "We use this small notebook to test the conversion of the model's weights and to make sure both the TensorFlow and PyTorch are coherent. In particular, we compare the weights of the last layer on a simple example (in `input.txt`).\n",
    "\n",
    "To run this notebook, please make sure that your Python environment has both TensorFlow and PyTorch.\n",
    "You should follow the instructions in the `README.md` and make sure that you have:\n",
    "- the original TensorFlow implementation\n",
    "- the `BERT-base, Uncased` model\n",
    "- run the script `convert_tf_checkpoint_to_pytorch.py` to convert the weights to PyTorch\n",
    "\n",
    "Please modify the relative paths accordingly (at the beggining of Sections 1 and 2)."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1/ TensorFlow code"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "original_tf_inplem_dir = \"../bert/\"\n",
    "model_dir = \"../uncased_L-12_H-768_A-12/\"\n",
    "\n",
    "vocab_file = model_dir + \"vocab.txt\"\n",
    "bert_config_file = model_dir + \"bert_config.json\"\n",
    "init_checkpoint = model_dir + \"bert_model.ckpt\"\n",
    "\n",
    "input_file = \"input.txt\"\n",
    "max_seq_length = 128"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:09:37.498678Z",
     "start_time": "2018-11-03T02:09:36.366672Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.6/site-packages/h5py/__init__.py:34: 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 sys\n",
    "sys.path.append(original_tf_inplem_dir)\n",
    "\n",
    "from extract_features import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:09:37.621865Z",
     "start_time": "2018-11-03T02:09:37.500988Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:*** Example ***\n",
      "INFO:tensorflow:unique_id: 0\n",
      "INFO:tensorflow:tokens: [CLS] who was jim henson ? [SEP] jim henson was a puppet ##eer [SEP]\n",
      "INFO:tensorflow:input_ids: 101 2040 2001 3958 27227 1029 102 3958 27227 2001 1037 13997 11510 102 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n",
      "INFO:tensorflow:input_mask: 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n",
      "INFO:tensorflow:input_type_ids: 0 0 0 0 0 0 0 1 1 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0\n"
     ]
    }
   ],
   "source": [
    "layer_indexes = list(range(12))\n",
    "bert_config = modeling.BertConfig.from_json_file(bert_config_file)\n",
    "tokenizer = tokenization.FullTokenizer(\n",
    "    vocab_file=vocab_file, do_lower_case=True)\n",
    "examples = read_examples(input_file)\n",
    "\n",
    "features = convert_examples_to_features(\n",
    "    examples=examples, seq_length=max_seq_length, tokenizer=tokenizer)\n",
    "unique_id_to_feature = {}\n",
    "for feature in features:\n",
    "    unique_id_to_feature[feature.unique_id] = feature"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:09:40.831618Z",
     "start_time": "2018-11-03T02:09:37.624063Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:Estimator's model_fn (<function model_fn_builder.<locals>.model_fn at 0x1289c1a60>) includes params argument, but params are not passed to Estimator.\n",
      "WARNING:tensorflow:Using temporary folder as model directory: /var/folders/y2/py87pn6115bdsdftbc6394nh0000gn/T/tmpmcfk2tyr\n",
      "INFO:tensorflow:Using config: {'_model_dir': '/var/folders/y2/py87pn6115bdsdftbc6394nh0000gn/T/tmpmcfk2tyr', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_session_config': allow_soft_placement: true\n",
      "graph_options {\n",
      "  rewrite_options {\n",
      "    meta_optimizer_iterations: ONE\n",
      "  }\n",
      "}\n",
      ", '_keep_checkpoint_max': 5, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': None, '_train_distribute': None, '_device_fn': None, '_protocol': None, '_eval_distribute': None, '_experimental_distribute': None, '_service': None, '_cluster_spec': <tensorflow.python.training.server_lib.ClusterSpec object at 0x12c242470>, '_task_type': 'worker', '_task_id': 0, '_global_id_in_cluster': 0, '_master': '', '_evaluation_master': '', '_is_chief': True, '_num_ps_replicas': 0, '_num_worker_replicas': 1, '_tpu_config': TPUConfig(iterations_per_loop=2, num_shards=1, num_cores_per_replica=None, per_host_input_for_training=3, tpu_job_name=None, initial_infeed_sleep_secs=None, input_partition_dims=None), '_cluster': None}\n",
      "WARNING:tensorflow:Setting TPUConfig.num_shards==1 is an unsupported behavior. Please fix as soon as possible (leaving num_shards as None.\n",
      "INFO:tensorflow:_TPUContext: eval_on_tpu True\n",
      "WARNING:tensorflow:eval_on_tpu ignored because use_tpu is False.\n"
     ]
    }
   ],
   "source": [
    "is_per_host = tf.contrib.tpu.InputPipelineConfig.PER_HOST_V2\n",
    "run_config = tf.contrib.tpu.RunConfig(\n",
    "    master=None,\n",
    "    tpu_config=tf.contrib.tpu.TPUConfig(\n",
    "        num_shards=1,\n",
    "        per_host_input_for_training=is_per_host))\n",
    "\n",
    "model_fn = model_fn_builder(\n",
    "    bert_config=bert_config,\n",
    "    init_checkpoint=init_checkpoint,\n",
    "    layer_indexes=layer_indexes,\n",
    "    use_tpu=False,\n",
    "    use_one_hot_embeddings=False)\n",
    "\n",
    "# If TPU is not available, this will fall back to normal Estimator on CPU\n",
    "# or GPU.\n",
    "estimator = tf.contrib.tpu.TPUEstimator(\n",
    "    use_tpu=False,\n",
    "    model_fn=model_fn,\n",
    "    config=run_config,\n",
    "    predict_batch_size=1)\n",
    "\n",
    "input_fn = input_fn_builder(\n",
    "    features=features, seq_length=max_seq_length)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:09:46.413197Z",
     "start_time": "2018-11-03T02:09:40.834621Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Could not find trained model in model_dir: /var/folders/y2/py87pn6115bdsdftbc6394nh0000gn/T/tmpmcfk2tyr, running initialization to predict.\n",
      "INFO:tensorflow:Calling model_fn.\n",
      "INFO:tensorflow:Running infer on CPU\n",
      "INFO:tensorflow:Done calling model_fn.\n",
      "INFO:tensorflow:Graph was finalized.\n",
      "INFO:tensorflow:Running local_init_op.\n",
      "INFO:tensorflow:Done running local_init_op.\n",
      "extracting layer 0\n",
      "extracting layer 1\n",
      "extracting layer 2\n",
      "extracting layer 3\n",
      "extracting layer 4\n",
      "extracting layer 5\n",
      "extracting layer 6\n",
      "extracting layer 7\n",
      "extracting layer 8\n",
      "extracting layer 9\n",
      "extracting layer 10\n",
      "extracting layer 11\n",
      "INFO:tensorflow:prediction_loop marked as finished\n",
      "INFO:tensorflow:prediction_loop marked as finished\n"
     ]
    }
   ],
   "source": [
    "tensorflow_all_out = []\n",
    "for result in estimator.predict(input_fn, yield_single_examples=True):\n",
    "    unique_id = int(result[\"unique_id\"])\n",
    "    feature = unique_id_to_feature[unique_id]\n",
    "    output_json = collections.OrderedDict()\n",
    "    output_json[\"linex_index\"] = unique_id\n",
    "    tensorflow_all_out_features = []\n",
    "    # for (i, token) in enumerate(feature.tokens):\n",
    "    all_layers = []\n",
    "    for (j, layer_index) in enumerate(layer_indexes):\n",
    "        print(\"extracting layer {}\".format(j))\n",
    "        layer_output = result[\"layer_output_%d\" % j]\n",
    "        layers = collections.OrderedDict()\n",
    "        layers[\"index\"] = layer_index\n",
    "        layers[\"values\"] = layer_output\n",
    "        all_layers.append(layers)\n",
    "    tensorflow_out_features = collections.OrderedDict()\n",
    "    tensorflow_out_features[\"layers\"] = all_layers\n",
    "    tensorflow_all_out_features.append(tensorflow_out_features)\n",
    "\n",
    "    output_json[\"features\"] = tensorflow_all_out_features\n",
    "    tensorflow_all_out.append(output_json)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:09:46.460128Z",
     "start_time": "2018-11-03T02:09:46.416138Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1\n",
      "2\n",
      "odict_keys(['linex_index', 'features'])\n",
      "number of tokens 1\n",
      "number of layers 12\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(128, 768)"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "print(len(tensorflow_all_out))\n",
    "print(len(tensorflow_all_out[0]))\n",
    "print(tensorflow_all_out[0].keys())\n",
    "print(\"number of tokens\", len(tensorflow_all_out[0]['features']))\n",
    "print(\"number of layers\", len(tensorflow_all_out[0]['features'][0]['layers']))\n",
    "tensorflow_all_out[0]['features'][0]['layers'][0]['values'].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:09:46.498637Z",
     "start_time": "2018-11-03T02:09:46.463115Z"
    }
   },
   "outputs": [],
   "source": [
    "tensorflow_outputs = list(tensorflow_all_out[0]['features'][0]['layers'][t]['values'] for t in layer_indexes)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2/ PyTorch code"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:09:46.660303Z",
     "start_time": "2018-11-03T02:09:46.501325Z"
    }
   },
   "outputs": [],
   "source": [
    "import extract_features_pytorch\n",
    "from extract_features_pytorch import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "init_checkpoint_pt = \"../pytorch_model/uncased_L-12_H-768_A-12/pytorch_model.bin\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:09:48.292135Z",
     "start_time": "2018-11-03T02:09:46.661921Z"
    },
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "BertModel(\n",
       "  (embeddings): BERTEmbeddings(\n",
       "    (word_embeddings): Embedding(30522, 768)\n",
       "    (position_embeddings): Embedding(512, 768)\n",
       "    (token_type_embeddings): Embedding(2, 768)\n",
       "    (LayerNorm): BERTLayerNorm()\n",
       "    (dropout): Dropout(p=0.1)\n",
       "  )\n",
       "  (encoder): BERTEncoder(\n",
       "    (layer): ModuleList(\n",
       "      (0): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (1): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (2): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (3): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (4): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (5): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (6): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (7): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (8): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (9): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (10): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (11): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "    )\n",
       "  )\n",
       "  (pooler): BERTPooler(\n",
       "    (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "    (activation): Tanh()\n",
       "  )\n",
       ")"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "device = torch.device(\"cpu\")\n",
    "model = extract_features_pytorch.BertModel(bert_config)\n",
    "model.load_state_dict(torch.load(init_checkpoint_pt, map_location='cpu'))\n",
    "model.to(device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:09:48.332982Z",
     "start_time": "2018-11-03T02:09:48.294056Z"
    },
    "code_folding": []
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "BertModel(\n",
       "  (embeddings): BERTEmbeddings(\n",
       "    (word_embeddings): Embedding(30522, 768)\n",
       "    (position_embeddings): Embedding(512, 768)\n",
       "    (token_type_embeddings): Embedding(2, 768)\n",
       "    (LayerNorm): BERTLayerNorm()\n",
       "    (dropout): Dropout(p=0.1)\n",
       "  )\n",
       "  (encoder): BERTEncoder(\n",
       "    (layer): ModuleList(\n",
       "      (0): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (1): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (2): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (3): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (4): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (5): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (6): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (7): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (8): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (9): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (10): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "      (11): BERTLayer(\n",
       "        (attention): BERTAttention(\n",
       "          (self): BERTSelfAttention(\n",
       "            (query): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (key): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (value): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "          (output): BERTSelfOutput(\n",
       "            (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "            (LayerNorm): BERTLayerNorm()\n",
       "            (dropout): Dropout(p=0.1)\n",
       "          )\n",
       "        )\n",
       "        (intermediate): BERTIntermediate(\n",
       "          (dense): Linear(in_features=768, out_features=3072, bias=True)\n",
       "        )\n",
       "        (output): BERTOutput(\n",
       "          (dense): Linear(in_features=3072, out_features=768, bias=True)\n",
       "          (LayerNorm): BERTLayerNorm()\n",
       "          (dropout): Dropout(p=0.1)\n",
       "        )\n",
       "      )\n",
       "    )\n",
       "  )\n",
       "  (pooler): BERTPooler(\n",
       "    (dense): Linear(in_features=768, out_features=768, bias=True)\n",
       "    (activation): Tanh()\n",
       "  )\n",
       ")"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_input_ids = torch.tensor([f.input_ids for f in features], dtype=torch.long)\n",
    "all_input_mask = torch.tensor([f.input_mask for f in features], dtype=torch.long)\n",
    "all_input_type_ids = torch.tensor([f.input_type_ids for f in features], dtype=torch.long)\n",
    "all_example_index = torch.arange(all_input_ids.size(0), dtype=torch.long)\n",
    "\n",
    "eval_data = TensorDataset(all_input_ids, all_input_mask, all_input_type_ids, all_example_index)\n",
    "eval_sampler = SequentialSampler(eval_data)\n",
    "eval_dataloader = DataLoader(eval_data, sampler=eval_sampler, batch_size=1)\n",
    "\n",
    "model.eval()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:09:54.371188Z",
     "start_time": "2018-11-03T02:09:53.976875Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor([[  101,  2040,  2001,  3958, 27227,  1029,   102,  3958, 27227,  2001,\n",
      "          1037, 13997, 11510,   102,     0,     0,     0,     0,     0,     0,\n",
      "             0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
      "             0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
      "             0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
      "             0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
      "             0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
      "             0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
      "             0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
      "             0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
      "             0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
      "             0,     0,     0,     0,     0,     0,     0,     0,     0,     0,\n",
      "             0,     0,     0,     0,     0,     0,     0,     0]])\n",
      "tensor([[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
      "         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
      "         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
      "         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
      "         0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
      "         0, 0, 0, 0, 0, 0, 0, 0]])\n",
      "tensor([0])\n",
      "layer 0 0\n",
      "layer 1 1\n",
      "layer 2 2\n",
      "layer 3 3\n",
      "layer 4 4\n",
      "layer 5 5\n",
      "layer 6 6\n",
      "layer 7 7\n",
      "layer 8 8\n",
      "layer 9 9\n",
      "layer 10 10\n",
      "layer 11 11\n"
     ]
    }
   ],
   "source": [
    "layer_indexes = list(range(12))\n",
    "\n",
    "pytorch_all_out = []\n",
    "for input_ids, input_mask, input_type_ids, example_indices in eval_dataloader:\n",
    "    print(input_ids)\n",
    "    print(input_mask)\n",
    "    print(example_indices)\n",
    "    input_ids = input_ids.to(device)\n",
    "    input_mask = input_mask.float().to(device)\n",
    "\n",
    "    all_encoder_layers, _ = model(input_ids, token_type_ids=input_type_ids, attention_mask=input_mask)\n",
    "\n",
    "    for b, example_index in enumerate(example_indices):\n",
    "        feature = features[example_index.item()]\n",
    "        unique_id = int(feature.unique_id)\n",
    "        # feature = unique_id_to_feature[unique_id]\n",
    "        output_json = collections.OrderedDict()\n",
    "        output_json[\"linex_index\"] = unique_id\n",
    "        all_out_features = []\n",
    "        # for (i, token) in enumerate(feature.tokens):\n",
    "        all_layers = []\n",
    "        for (j, layer_index) in enumerate(layer_indexes):\n",
    "            print(\"layer\", j, layer_index)\n",
    "            layer_output = all_encoder_layers[int(layer_index)].detach().cpu().numpy()\n",
    "            layer_output = layer_output[b]\n",
    "            layers = collections.OrderedDict()\n",
    "            layers[\"index\"] = layer_index\n",
    "            layer_output = layer_output\n",
    "            layers[\"values\"] = layer_output if not isinstance(layer_output, (int, float)) else [layer_output]\n",
    "            all_layers.append(layers)\n",
    "\n",
    "            out_features = collections.OrderedDict()\n",
    "            out_features[\"layers\"] = all_layers\n",
    "            all_out_features.append(out_features)\n",
    "        output_json[\"features\"] = all_out_features\n",
    "        pytorch_all_out.append(output_json)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:09:57.139854Z",
     "start_time": "2018-11-03T02:09:57.104636Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1\n",
      "2\n",
      "odict_keys(['linex_index', 'features'])\n",
      "number of tokens 1\n",
      "number of layers 12\n",
      "hidden_size 128\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(128, 768)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "print(len(pytorch_all_out))\n",
    "print(len(pytorch_all_out[0]))\n",
    "print(pytorch_all_out[0].keys())\n",
    "print(\"number of tokens\", len(pytorch_all_out))\n",
    "print(\"number of layers\", len(pytorch_all_out[0]['features'][0]['layers']))\n",
    "print(\"hidden_size\", len(pytorch_all_out[0]['features'][0]['layers'][0]['values']))\n",
    "pytorch_all_out[0]['features'][0]['layers'][0]['values'].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:09:59.000058Z",
     "start_time": "2018-11-03T02:09:58.967575Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(128, 768)\n",
      "(128, 768)\n"
     ]
    }
   ],
   "source": [
    "pytorch_outputs = list(pytorch_all_out[0]['features'][0]['layers'][t]['values'] for t in layer_indexes)\n",
    "print(pytorch_outputs[0].shape)\n",
    "print(pytorch_outputs[1].shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:09:59.462123Z",
     "start_time": "2018-11-03T02:09:59.430932Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(128, 768)\n",
      "(128, 768)\n"
     ]
    }
   ],
   "source": [
    "print(tensorflow_outputs[0].shape)\n",
    "print(tensorflow_outputs[1].shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3/ Comparing the standard deviation on the last layer of both models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:10:00.014784Z",
     "start_time": "2018-11-03T02:09:59.983978Z"
    }
   },
   "outputs": [],
   "source": [
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-11-03T02:10:09.582557Z",
     "start_time": "2018-11-03T02:10:09.549308Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(128, 768) (128, 768)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "4.1671223e-07"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "i = 11\n",
    "print(np.array(tensorflow_outputs[i]).shape, np.array(pytorch_outputs[i]).shape)\n",
    "np.sqrt(np.mean((np.array(tensorflow_outputs[i]) - np.array(pytorch_outputs[i]))**2.0))"
   ]
  }
 ],
 "metadata": {
  "hide_input": false,
  "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": "3.6.5"
  },
  "toc": {
   "colors": {
    "hover_highlight": "#DAA520",
    "running_highlight": "#FF0000",
    "selected_highlight": "#FFD700"
   },
   "moveMenuLeft": true,
   "nav_menu": {
    "height": "48px",
    "width": "252px"
   },
   "navigate_menu": true,
   "number_sections": true,
   "sideBar": true,
   "threshold": 4,
   "toc_cell": false,
   "toc_section_display": "block",
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}