eval_generation.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import random\n",
    "\n",
    "import numpy as np\n",
    "import torch\n",
    "from torch.utils.data.dataloader import default_collate\n",
    "\n",
    "from settings import EXPERIMENTS_DIR\n",
    "from experiment import Experiment\n",
    "from utils import to_device, load_weights, load_embeddings, create_embeddings_matrix\n",
    "from vocab import Vocab\n",
    "from train import create_model\n",
    "from preprocess import load_dataset, create_dataset_reader"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "exp_id = 'train.jkmkvrrr'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Load everything"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "exp = Experiment.load(EXPERIMENTS_DIR, exp_id)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "TrainConfig(model_class=<class 'models.Seq2SeqMeaningStyle'>, preprocess_exp_id='preprocess.buppgpnf', embedding_size=300, hidden_size=256, dropout=0.2, scheduled_sampling_ratio=0.5, pretrained_embeddings=True, trainable_embeddings=False, meaning_size=128, style_size=128, lr=0.001, weight_decay=1e-07, grad_clipping=5, D_num_iterations=10, D_loss_multiplier=1, P_loss_multiplier=10, P_bow_loss_multiplier=1, use_discriminator=True, use_predictor=False, use_predictor_bow=True, use_motivator=True, use_gauss=False, num_epochs=500, batch_size=1024, best_loss='loss')"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "exp.config"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dataset: 453655, val: 10000, test: 10000\n",
      "Vocab: 9419, style vocab: 2\n",
      "W_emb: (9419, 300)\n"
     ]
    }
   ],
   "source": [
    "preprocess_exp = Experiment.load(EXPERIMENTS_DIR, exp.config.preprocess_exp_id)\n",
    "dataset_train, dataset_val, dataset_test, vocab, style_vocab, W_emb = load_dataset(preprocess_exp)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset_reader = create_dataset_reader(preprocess_exp.config)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = create_model(exp.config, vocab, style_vocab, dataset_train.max_len, W_emb)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "load_weights(model, exp.experiment_dir.joinpath('best.th'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = model.eval()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Predict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def create_inputs(instances):\n",
    "    if not isinstance(instances, list):\n",
    "        instances = [instances,]\n",
    "        \n",
    "    if not isinstance(instances[0], dict):\n",
    "        sentences = [\n",
    "            dataset_reader.preprocess_sentence(dataset_reader.spacy( dataset_reader.clean_sentence(sent)))\n",
    "            for sent in instances\n",
    "        ]\n",
    "        \n",
    "        style = list(style_vocab.token2id.keys())[0]\n",
    "        instances = [\n",
    "            {\n",
    "                'sentence': sent,\n",
    "                'style': style,\n",
    "            }\n",
    "            for sent in sentences\n",
    "        ]\n",
    "        \n",
    "        for inst in instances:\n",
    "            inst_encoded = dataset_train.encode_instance(inst)\n",
    "            inst.update(inst_encoded)            \n",
    "    \n",
    "    \n",
    "    instances = [\n",
    "        {\n",
    "            'sentence': inst['sentence_enc'],\n",
    "            'style': inst['style_enc'],\n",
    "        } \n",
    "        for inst in instances\n",
    "    ]\n",
    "    \n",
    "    instances = default_collate(instances)\n",
    "    instances = to_device(instances)      \n",
    "    \n",
    "    return instances"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_sentences(outputs):\n",
    "    predicted_indices = outputs[\"predictions\"]\n",
    "    end_idx = vocab[Vocab.END_TOKEN]\n",
    "    \n",
    "    if not isinstance(predicted_indices, np.ndarray):\n",
    "        predicted_indices = predicted_indices.detach().cpu().numpy()\n",
    "\n",
    "    all_predicted_tokens = []\n",
    "    for indices in predicted_indices:\n",
    "        indices = list(indices)\n",
    "\n",
    "        # Collect indices till the first end_symbol\n",
    "        if end_idx in indices:\n",
    "            indices = indices[:indices.index(end_idx)]\n",
    "\n",
    "        predicted_tokens = [vocab.id2token[x] for x in indices]\n",
    "        all_predicted_tokens.append(predicted_tokens)\n",
    "        \n",
    "    return all_predicted_tokens"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "sentence =  ' '.join(dataset_val.instances[1]['sentence'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'they are really good people .'"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sentence"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "inputs = create_inputs(sentence)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "outputs = model(inputs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "sentences = get_sentences(outputs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'they are really good people .'"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "' '.join(sentences[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Swap style"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "possible_styles = list(style_vocab.token2id.keys()) #['negative', 'positive']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['negative', 'positive']"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "possible_styles"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "sentences0 = [s for s in dataset_val.instances if s['style'] == possible_styles[0]]\n",
    "sentences1 = [s for s in dataset_val.instances if s['style'] == possible_styles[1]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3239 if i could give negative stars i certainly would for this place .\n",
      "2083 if it was not broke , why in the world would you fix it ?\n",
      "3874 the rice had hard things in it .\n",
      "1569 quite possibly the worst experience of my life .\n",
      "3584 however our little one ordered buttered noodles and was pleased as punch .\n"
     ]
    }
   ],
   "source": [
    "for i in np.random.choice(np.arange(len(sentences0)), 5):\n",
    "    print(i, ' '.join(sentences0[i]['sentence']))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "4935 which is awesome !\n",
      "1561 it is also a good place to go just for dessert .\n",
      "1208 they are amazing , truly , could not be happier .\n",
      "1347 i had the tamales and they were the best i have ever had !\n",
      "3450 the capistrami is the best thing ever .\n"
     ]
    }
   ],
   "source": [
    "for i in np.random.choice(np.arange(len(sentences1)), 5):\n",
    "    print(i, ' '.join(sentences1[i]['sentence']))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Swap"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "target0 = 3874 # np.random.choice(np.arange(len(sentences0)))\n",
    "target1 = 4935 # np.random.choice(np.arange(len(sentences0)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "the rice had hard things in it .\n"
     ]
    }
   ],
   "source": [
    "print(' '.join(sentences0[target0]['sentence']))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "which is awesome !\n"
     ]
    }
   ],
   "source": [
    "print(' '.join(sentences1[target1]['sentence']))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "inputs = create_inputs([\n",
    "    sentences0[target0],\n",
    "    sentences1[target1],\n",
    "])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [],
   "source": [
    "z_hidden = model(inputs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([2, 128])"
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "z_hidden['style_hidden'].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([2, 128])"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "z_hidden['meaning_hidden'].shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "original_decoded = model.decode(z_hidden)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [],
   "source": [
    "original_sentences = get_sentences(original_decoded)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "the rice had hard things in it .\n",
      "which is awesome !\n"
     ]
    }
   ],
   "source": [
    "print(' '.join(original_sentences[0]))\n",
    "print(' '.join(original_sentences[1]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "z_hidden_swapped = {\n",
    "    'meaning_hidden': torch.stack([\n",
    "        z_hidden['meaning_hidden'][0].clone(),\n",
    "        z_hidden['meaning_hidden'][1].clone(),        \n",
    "    ], dim=0),\n",
    "    'style_hidden': torch.stack([\n",
    "        z_hidden['style_hidden'][1].clone(),\n",
    "        z_hidden['style_hidden'][0].clone(),        \n",
    "    ], dim=0),\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "swaped_decoded = model.decode(z_hidden_swapped)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [],
   "source": [
    "swaped_sentences = get_sentences(swaped_decoded)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "the rice had hard things in it .\n",
      "which is awesome !\n",
      "\n",
      "plus is really hard to it .\n",
      "the rice was awesome .\n"
     ]
    }
   ],
   "source": [
    "print(' '.join(original_sentences[0]))\n",
    "print(' '.join(original_sentences[1]))\n",
    "print()\n",
    "print(' '.join(swaped_sentences[0]))\n",
    "print(' '.join(swaped_sentences[1]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}