File size: 36,014 Bytes
eec604d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 | {
"cells": [
{
"cell_type": "markdown",
"metadata": {
"collapsed": true
},
"source": [
"# Machine Translation Project (English to French)"
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [],
"source": [
"import collections\n",
"import numpy as np\n",
"import json\n",
"\n",
"from keras.preprocessing.text import Tokenizer\n",
"from keras.utils import pad_sequences\n",
"from keras.models import Model, Sequential\n",
"from keras.layers import Input, Dense, Embedding, GRU, LSTM, Bidirectional, Dropout, Activation, TimeDistributed, RepeatVector\n",
"from keras.optimizers import Adam\n",
"from keras.losses import sparse_categorical_crossentropy"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Verify access to the GPU"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[name: \"/device:CPU:0\"\n",
"device_type: \"CPU\"\n",
"memory_limit: 268435456\n",
"locality {\n",
"}\n",
"incarnation: 8951901467623568752\n",
"xla_global_id: -1\n",
", name: \"/device:GPU:0\"\n",
"device_type: \"GPU\"\n",
"memory_limit: 1733715559\n",
"locality {\n",
" bus_id: 1\n",
" links {\n",
" }\n",
"}\n",
"incarnation: 7542354691675806642\n",
"physical_device_desc: \"device: 0, name: NVIDIA GeForce RTX 2050, pci bus id: 0000:01:00.0, compute capability: 8.6\"\n",
"xla_global_id: 416903419\n",
"]\n"
]
}
],
"source": [
"from tensorflow.python.client import device_lib\n",
"print(device_lib.list_local_devices())"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Dataset\n",
"For our machine translation project, we opt for a dataset featuring a limited vocabulary, specifically designed to facilitate a more manageable and efficient training process. Unlike the extensive [WMT](http://www.statmt.org/) datasets, our chosen dataset ensures a quicker training time and demands fewer computational resources. This strategic decision aims to balance the learning experience while still achieving meaningful results within practical time constraints.\n",
"### Load Data"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [],
"source": [
"def load_data(path):\n",
" input_file = path\n",
" with open(input_file, \"r\") as f:\n",
" data = f.read()\n",
" return data.split('\\n')\n",
"\n",
"english_sentences = load_data('data/english')\n",
"french_sentences = load_data('data/french')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Sample Data"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"['new jersey is sometimes quiet during autumn , and it is snowy in april .',\n",
" 'the united states is usually chilly during july , and it is usually freezing in november .',\n",
" 'california is usually quiet during march , and it is usually hot in june .',\n",
" 'the united states is sometimes mild during june , and it is cold in september .',\n",
" 'your least liked fruit is the grape , but my least liked is the apple .']"
]
},
"execution_count": 5,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"english_sentences[:5]"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"By examining the sentences, it's apparent that they have undergone preprocessing: punctuation has been delimited with spaces, and all the text has been converted to lowercase. This preprocessing serves a crucial purpose in text preparation. Firstly, delimiting punctuation with spaces ensures that each punctuation mark is treated as a separate token, aiding the model in understanding sentence structure. Secondly, converting the entire text to lowercase standardizes the input, preventing the model from distinguishing between words solely based on their casing. This uniformity facilitates more effective training and generalization, enhancing the model's ability to grasp patterns and generate accurate translations."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Structure of the Dataset"
]
},
{
"cell_type": "code",
"execution_count": 6,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"1823250 English words.\n",
"227 unique English words.\n",
"10 Most common words in the English dataset:\n",
"\"is\" \",\" \".\" \"in\" \"it\" \"during\" \"the\" \"but\" \"and\" \"sometimes\"\n",
"\n",
"1961295 French words.\n",
"355 unique French words.\n",
"10 Most common words in the French dataset:\n",
"\"est\" \".\" \",\" \"en\" \"il\" \"les\" \"mais\" \"et\" \"la\" \"parfois\"\n"
]
}
],
"source": [
"english_words_counter = collections.Counter([word for sentence in english_sentences for word in sentence.split()])\n",
"french_words_counter = collections.Counter([word for sentence in french_sentences for word in sentence.split()])\n",
"\n",
"print('{} English words.'.format(len([word for sentence in english_sentences for word in sentence.split()])))\n",
"print('{} unique English words.'.format(len(english_words_counter)))\n",
"print('10 Most common words in the English dataset:')\n",
"print('\"' + '\" \"'.join(list(zip(*english_words_counter.most_common(10)))[0]) + '\"')\n",
"\n",
"print()\n",
"print('{} French words.'.format(len([word for sentence in french_sentences for word in sentence.split()])))\n",
"print('{} unique French words.'.format(len(french_words_counter)))\n",
"print('10 Most common words in the French dataset:')\n",
"print('\"' + '\" \"'.join(list(zip(*french_words_counter.most_common(10)))[0]) + '\"')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Preprocess\n",
"1. Tokenize the words into ids\n",
"2. Add padding to make all the sequences the same length."
]
},
{
"cell_type": "code",
"execution_count": 7,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{'the': 1, 'quick': 2, 'a': 3, 'brown': 4, 'fox': 5, 'jumps': 6, 'over': 7, 'lazy': 8, 'dog': 9, 'by': 10, 'jove': 11, 'my': 12, 'study': 13, 'of': 14, 'lexicography': 15, 'won': 16, 'prize': 17, 'this': 18, 'is': 19, 'short': 20, 'sentence': 21}\n",
"\n",
"Sequence 1 in x\n",
" Input: The quick brown fox jumps over the lazy dog .\n",
" Output: [1, 2, 4, 5, 6, 7, 1, 8, 9]\n",
"Sequence 2 in x\n",
" Input: By Jove , my quick study of lexicography won a prize .\n",
" Output: [10, 11, 12, 2, 13, 14, 15, 16, 3, 17]\n",
"Sequence 3 in x\n",
" Input: This is a short sentence .\n",
" Output: [18, 19, 3, 20, 21]\n"
]
}
],
"source": [
"def tokenize(x):\n",
" tokenizer = Tokenizer()\n",
" tokenizer.fit_on_texts(x)\n",
" return tokenizer.texts_to_sequences(x), tokenizer\n",
"\n",
"text_sentences = [\n",
" 'The quick brown fox jumps over the lazy dog .',\n",
" 'By Jove , my quick study of lexicography won a prize .',\n",
" 'This is a short sentence .']\n",
"\n",
"text_tokenized, text_tokenizer = tokenize(text_sentences)\n",
"print(text_tokenizer.word_index)\n",
"print()\n",
"for sample_i, (sent, token_sent) in enumerate(zip(text_sentences, text_tokenized)):\n",
" print('Sequence {} in x'.format(sample_i + 1))\n",
" print(' Input: {}'.format(sent))\n",
" print(' Output: {}'.format(token_sent))"
]
},
{
"cell_type": "code",
"execution_count": 8,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Sequence 1 in x\n",
" Input: [1 2 4 5 6 7 1 8 9]\n",
" Output: [1 2 4 5 6 7 1 8 9 0]\n",
"Sequence 2 in x\n",
" Input: [10 11 12 2 13 14 15 16 3 17]\n",
" Output: [10 11 12 2 13 14 15 16 3 17]\n",
"Sequence 3 in x\n",
" Input: [18 19 3 20 21]\n",
" Output: [18 19 3 20 21 0 0 0 0 0]\n"
]
}
],
"source": [
"def pad(x, length=None):\n",
" if length is None:\n",
" length = max([len(sentence) for sentence in x])\n",
" return pad_sequences(x, maxlen=length, padding='post')\n",
"\n",
"test_pad = pad(text_tokenized)\n",
"for sample_i, (token_sent, pad_sent) in enumerate(zip(text_tokenized, test_pad)):\n",
" print('Sequence {} in x'.format(sample_i + 1))\n",
" print(' Input: {}'.format(np.array(token_sent)))\n",
" print(' Output: {}'.format(pad_sent))"
]
},
{
"cell_type": "code",
"execution_count": 9,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Data Preprocessed\n",
"Max English sentence length: 15\n",
"Max French sentence length: 21\n",
"English vocabulary size: 199\n",
"French vocabulary size: 344\n"
]
}
],
"source": [
"def preprocess(x,y):\n",
" preprocess_x, x_tk = tokenize(x)\n",
" preprocess_y, y_tk = tokenize(y)\n",
" \n",
" preprocess_x = pad(preprocess_x)\n",
" preprocess_y = pad(preprocess_y)\n",
" \n",
" preprocess_y = preprocess_y.reshape(*preprocess_y.shape, 1)\n",
" \n",
" return preprocess_x, preprocess_y, x_tk, y_tk\n",
"\n",
"preproc_english_sentences, preproc_french_sentences, english_tokenizer, french_tokenizer = preprocess(english_sentences, french_sentences)\n",
"\n",
"max_english_sequence_length = preproc_english_sentences.shape[1]\n",
"max_french_sequence_length = preproc_french_sentences.shape[1]\n",
"english_vocab_size = len(english_tokenizer.word_index)\n",
"french_vocab_size = len(french_tokenizer.word_index)\n",
"\n",
"print('Data Preprocessed')\n",
"print(\"Max English sentence length:\", max_english_sequence_length)\n",
"print(\"Max French sentence length:\", max_french_sequence_length)\n",
"print(\"English vocabulary size:\", english_vocab_size)\n",
"print(\"French vocabulary size:\", french_vocab_size)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Models\n",
"- Model 1 is a simple RNN\n",
"- Model 2 is a Bidirectional RNN\n",
"- Model 3 is an Embedding RNN\n",
"\n",
"### Ids Back to Text\n",
"The neural network will be translating the input to words ids, which isn't the final form we want. We want the French translation. The function `logits_to_text` will bridge the gab between the logits from the neural network to the French translation. You'll be using this function to better understand the output of the neural network."
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {},
"outputs": [],
"source": [
"def logits_to_text(logits, tokenizer):\n",
" index_to_words = {id: word for word, id in tokenizer.word_index.items()}\n",
" index_to_words[0] = '<PAD>'\n",
" \n",
" return ' '.join([index_to_words[prediction] for prediction in np.argmax(logits, 1)])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Model 1: RNN\n",
"\n",
"A basic RNN model is a good baseline for sequence data. In this model, you'll build a RNN that translates English to French."
]
},
{
"cell_type": "code",
"execution_count": 11,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Epoch 1/20\n",
"108/108 [==============================] - 17s 90ms/step - loss: 1.9094 - accuracy: 0.5446 - val_loss: nan - val_accuracy: 0.6307\n",
"Epoch 2/20\n",
"108/108 [==============================] - 9s 84ms/step - loss: 1.2243 - accuracy: 0.6429 - val_loss: nan - val_accuracy: 0.6716\n",
"Epoch 3/20\n",
"108/108 [==============================] - 9s 83ms/step - loss: 1.0848 - accuracy: 0.6683 - val_loss: nan - val_accuracy: 0.6864\n",
"Epoch 4/20\n",
"108/108 [==============================] - 9s 83ms/step - loss: 1.0057 - accuracy: 0.6832 - val_loss: nan - val_accuracy: 0.6927\n",
"Epoch 5/20\n",
"108/108 [==============================] - 9s 83ms/step - loss: 0.9484 - accuracy: 0.6931 - val_loss: nan - val_accuracy: 0.7108\n",
"Epoch 6/20\n",
"108/108 [==============================] - 9s 83ms/step - loss: 0.9019 - accuracy: 0.7036 - val_loss: nan - val_accuracy: 0.7007\n",
"Epoch 7/20\n",
"108/108 [==============================] - 9s 84ms/step - loss: 0.8916 - accuracy: 0.6999 - val_loss: nan - val_accuracy: 0.7244\n",
"Epoch 8/20\n",
"108/108 [==============================] - 9s 83ms/step - loss: 0.8407 - accuracy: 0.7178 - val_loss: nan - val_accuracy: 0.7567\n",
"Epoch 9/20\n",
"108/108 [==============================] - 9s 83ms/step - loss: 0.7807 - accuracy: 0.7405 - val_loss: nan - val_accuracy: 0.7405\n",
"Epoch 10/20\n",
"108/108 [==============================] - 9s 84ms/step - loss: 0.7474 - accuracy: 0.7496 - val_loss: nan - val_accuracy: 0.7721\n",
"Epoch 11/20\n",
"108/108 [==============================] - 9s 83ms/step - loss: 0.7739 - accuracy: 0.7392 - val_loss: nan - val_accuracy: 0.7392\n",
"Epoch 12/20\n",
"108/108 [==============================] - 9s 84ms/step - loss: 0.7552 - accuracy: 0.7420 - val_loss: nan - val_accuracy: 0.7851\n",
"Epoch 13/20\n",
"108/108 [==============================] - 9s 83ms/step - loss: 0.7238 - accuracy: 0.7550 - val_loss: nan - val_accuracy: 0.7937\n",
"Epoch 14/20\n",
"108/108 [==============================] - 9s 83ms/step - loss: 0.7126 - accuracy: 0.7568 - val_loss: nan - val_accuracy: 0.7830\n",
"Epoch 15/20\n",
"108/108 [==============================] - 9s 84ms/step - loss: 0.6838 - accuracy: 0.7650 - val_loss: nan - val_accuracy: 0.7976\n",
"Epoch 16/20\n",
"108/108 [==============================] - 9s 83ms/step - loss: 0.6577 - accuracy: 0.7776 - val_loss: nan - val_accuracy: 0.7995\n",
"Epoch 17/20\n",
"108/108 [==============================] - 9s 83ms/step - loss: 0.6447 - accuracy: 0.7821 - val_loss: nan - val_accuracy: 0.8072\n",
"Epoch 18/20\n",
"108/108 [==============================] - 9s 83ms/step - loss: 0.6309 - accuracy: 0.7858 - val_loss: nan - val_accuracy: 0.8100\n",
"Epoch 19/20\n",
"108/108 [==============================] - 9s 84ms/step - loss: 0.6073 - accuracy: 0.7930 - val_loss: nan - val_accuracy: 0.8111\n",
"Epoch 20/20\n",
"108/108 [==============================] - 9s 83ms/step - loss: 0.6100 - accuracy: 0.7912 - val_loss: nan - val_accuracy: 0.8150\n"
]
},
{
"data": {
"text/plain": [
"<keras.callbacks.History at 0x2573cec49a0>"
]
},
"execution_count": 11,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"def simple_model(input_shape, output_sequence_length, english_vocab_size, french_vocab_size):\n",
" \n",
" #Hyperparameters\n",
" learning_rate = 0.005\n",
" \n",
" # Build the layers\n",
" model = Sequential()\n",
" model.add(GRU(256, input_shape=input_shape[1:], return_sequences=True))\n",
" model.add(TimeDistributed(Dense(1024, activation='relu')))\n",
" model.add(Dropout(0.5))\n",
" model.add(TimeDistributed(Dense(french_vocab_size, activation='softmax')))\n",
" \n",
" # Compile model\n",
" model.compile(loss = sparse_categorical_crossentropy,\n",
" optimizer = Adam(learning_rate),\n",
" metrics = ['accuracy'])\n",
" \n",
" return model\n",
"\n",
"tmp_x = pad(preproc_english_sentences, max_french_sequence_length)\n",
"tmp_x = tmp_x.reshape((-1, preproc_french_sentences.shape[-2], 1))\n",
"\n",
"#Train the neural network\n",
"simple_rnn_model = simple_model(\n",
" tmp_x.shape,\n",
" max_french_sequence_length,\n",
" english_vocab_size,\n",
" french_vocab_size)\n",
"\n",
"simple_rnn_model.fit(tmp_x, preproc_french_sentences, batch_size=1024, epochs=20, validation_split=0.2)"
]
},
{
"cell_type": "code",
"execution_count": 12,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Prediciton:\n",
"1/1 [==============================] - 0s 259ms/step\n",
"new jersey est parfois calme en mois et il est il est en <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD>\n",
"\n",
"Correct Translation:\n",
"[\"new jersey est parfois calme pendant l' automne , et il est neigeux en avril .\"]\n",
"\n",
"Original text:\n",
"['new jersey is sometimes quiet during autumn , and it is snowy in april .']\n"
]
}
],
"source": [
"# Print prediction(s)\n",
"print(\"Prediciton:\")\n",
"print(logits_to_text(simple_rnn_model.predict(tmp_x[:1])[0], french_tokenizer))\n",
"\n",
"print(\"\\nCorrect Translation:\")\n",
"print(french_sentences[:1])\n",
"\n",
"print('\\nOriginal text:')\n",
"print(english_sentences[:1])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Model 2: Bidirectional RNNs\n",
"\n",
"One restriction of a RNN is that it can't see the future input, only the past. This is where bidirectional recurrent neural networks come in. They are able to see the future data."
]
},
{
"cell_type": "code",
"execution_count": 13,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Model: \"sequential_1\"\n",
"_________________________________________________________________\n",
" Layer (type) Output Shape Param # \n",
"=================================================================\n",
" bidirectional (Bidirectiona (None, 21, 256) 100608 \n",
" l) \n",
" \n",
" time_distributed_2 (TimeDis (None, 21, 1024) 263168 \n",
" tributed) \n",
" \n",
" dropout_1 (Dropout) (None, 21, 1024) 0 \n",
" \n",
" time_distributed_3 (TimeDis (None, 21, 344) 352600 \n",
" tributed) \n",
" \n",
"=================================================================\n",
"Total params: 716,376\n",
"Trainable params: 716,376\n",
"Non-trainable params: 0\n",
"_________________________________________________________________\n",
"None\n",
"Epoch 1/20\n",
"108/108 [==============================] - 12s 90ms/step - loss: 1.7553 - accuracy: 0.5756 - val_loss: nan - val_accuracy: 0.6505\n",
"Epoch 2/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 1.1655 - accuracy: 0.6550 - val_loss: nan - val_accuracy: 0.6802\n",
"Epoch 3/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 1.0423 - accuracy: 0.6759 - val_loss: nan - val_accuracy: 0.6903\n",
"Epoch 4/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.9663 - accuracy: 0.6880 - val_loss: nan - val_accuracy: 0.7003\n",
"Epoch 5/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.9119 - accuracy: 0.6974 - val_loss: nan - val_accuracy: 0.7207\n",
"Epoch 6/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.8700 - accuracy: 0.7059 - val_loss: nan - val_accuracy: 0.7287\n",
"Epoch 7/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.8359 - accuracy: 0.7129 - val_loss: nan - val_accuracy: 0.7301\n",
"Epoch 8/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.8495 - accuracy: 0.7090 - val_loss: nan - val_accuracy: 0.7300\n",
"Epoch 9/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.8025 - accuracy: 0.7197 - val_loss: nan - val_accuracy: 0.7386\n",
"Epoch 10/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.7839 - accuracy: 0.7228 - val_loss: nan - val_accuracy: 0.7429\n",
"Epoch 11/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.7671 - accuracy: 0.7248 - val_loss: nan - val_accuracy: 0.7461\n",
"Epoch 12/20\n",
"108/108 [==============================] - 9s 86ms/step - loss: 0.7490 - accuracy: 0.7278 - val_loss: nan - val_accuracy: 0.7487\n",
"Epoch 13/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.7341 - accuracy: 0.7307 - val_loss: nan - val_accuracy: 0.7473\n",
"Epoch 14/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.7183 - accuracy: 0.7363 - val_loss: nan - val_accuracy: 0.7614\n",
"Epoch 15/20\n",
"108/108 [==============================] - 9s 86ms/step - loss: 0.6998 - accuracy: 0.7427 - val_loss: nan - val_accuracy: 0.7594\n",
"Epoch 16/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.7086 - accuracy: 0.7361 - val_loss: nan - val_accuracy: 0.7596\n",
"Epoch 17/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.6889 - accuracy: 0.7424 - val_loss: nan - val_accuracy: 0.7679\n",
"Epoch 18/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.6780 - accuracy: 0.7493 - val_loss: nan - val_accuracy: 0.7763\n",
"Epoch 19/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.6625 - accuracy: 0.7535 - val_loss: nan - val_accuracy: 0.7771\n",
"Epoch 20/20\n",
"108/108 [==============================] - 9s 85ms/step - loss: 0.6572 - accuracy: 0.7553 - val_loss: nan - val_accuracy: 0.7560\n"
]
},
{
"data": {
"text/plain": [
"<keras.callbacks.History at 0x2573f78f0a0>"
]
},
"execution_count": 13,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"def bd_model(input_shape, output_sequence_length, english_vocab_size, french_vocab_size):\n",
" \n",
" #Hyperparameters\n",
" learning_rate = 0.005\n",
" \n",
" # Build the layers\n",
" model = Sequential()\n",
" model.add(Bidirectional(GRU(128, return_sequences=True), input_shape=input_shape[1:]))\n",
" model.add(TimeDistributed(Dense(1024, activation='relu')))\n",
" model.add(Dropout(0.5))\n",
" model.add(TimeDistributed(Dense(french_vocab_size, activation='softmax')))\n",
" \n",
" # Compile model\n",
" model.compile(loss = sparse_categorical_crossentropy,\n",
" optimizer = Adam(learning_rate),\n",
" metrics = ['accuracy'])\n",
" \n",
" return model\n",
"\n",
"tmp_x = pad(preproc_english_sentences, max_french_sequence_length)\n",
"tmp_x = tmp_x.reshape((-1, preproc_french_sentences.shape[-2], 1))\n",
"\n",
"# Train the neural network\n",
"bd_rnn_model = bd_model(\n",
" tmp_x.shape,\n",
" max_french_sequence_length,\n",
" english_vocab_size,\n",
" french_vocab_size)\n",
"\n",
"print(bd_rnn_model.summary())\n",
"\n",
"bd_rnn_model.fit(tmp_x, preproc_french_sentences, batch_size=1024, epochs=20, validation_split=0.2)"
]
},
{
"cell_type": "code",
"execution_count": 14,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Prediciton:\n",
"1/1 [==============================] - 1s 544ms/step\n",
"new jersey est parfois chaud en mois et il et il est en en <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD>\n",
"\n",
"Correct Translation:\n",
"[\"new jersey est parfois calme pendant l' automne , et il est neigeux en avril .\"]\n",
"\n",
"Original text:\n",
"['new jersey is sometimes quiet during autumn , and it is snowy in april .']\n"
]
}
],
"source": [
"# Print prediction(s)\n",
"print(\"Prediciton:\")\n",
"print(logits_to_text(bd_rnn_model.predict(tmp_x[:1])[0], french_tokenizer))\n",
"\n",
"print(\"\\nCorrect Translation:\")\n",
"print(french_sentences[:1])\n",
"\n",
"print('\\nOriginal text:')\n",
"print(english_sentences[:1])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Model 3: Embedding\n",
"\n",
"You've turned the words into ids, but there's a better representation of a word. This is called word embeddings. An embedding is a vector representation of the word that is close to similar words in n-dimensional space, where the n represents the size of the embedding vectors."
]
},
{
"cell_type": "code",
"execution_count": 15,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Model: \"sequential_2\"\n",
"_________________________________________________________________\n",
" Layer (type) Output Shape Param # \n",
"=================================================================\n",
" embedding (Embedding) (None, 21, 256) 50944 \n",
" \n",
" bidirectional_1 (Bidirectio (None, 21, 512) 789504 \n",
" nal) \n",
" \n",
" time_distributed_4 (TimeDis (None, 21, 1024) 525312 \n",
" tributed) \n",
" \n",
" dropout_2 (Dropout) (None, 21, 1024) 0 \n",
" \n",
" time_distributed_5 (TimeDis (None, 21, 344) 352600 \n",
" tributed) \n",
" \n",
"=================================================================\n",
"Total params: 1,718,360\n",
"Trainable params: 1,718,360\n",
"Non-trainable params: 0\n",
"_________________________________________________________________\n",
"None\n",
"Epoch 1/20\n",
"108/108 [==============================] - 17s 130ms/step - loss: 1.3473 - accuracy: 0.6924 - val_loss: nan - val_accuracy: 0.8697\n",
"Epoch 2/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.3152 - accuracy: 0.9003 - val_loss: nan - val_accuracy: 0.9346\n",
"Epoch 3/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.1808 - accuracy: 0.9434 - val_loss: nan - val_accuracy: 0.9578\n",
"Epoch 4/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.1291 - accuracy: 0.9601 - val_loss: nan - val_accuracy: 0.9702\n",
"Epoch 5/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.1022 - accuracy: 0.9688 - val_loss: nan - val_accuracy: 0.9737\n",
"Epoch 6/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.0854 - accuracy: 0.9739 - val_loss: nan - val_accuracy: 0.9772\n",
"Epoch 7/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.0763 - accuracy: 0.9767 - val_loss: nan - val_accuracy: 0.9780\n",
"Epoch 8/20\n",
"108/108 [==============================] - 14s 127ms/step - loss: 0.0658 - accuracy: 0.9798 - val_loss: nan - val_accuracy: 0.9798\n",
"Epoch 9/20\n",
"108/108 [==============================] - 14s 127ms/step - loss: 0.0604 - accuracy: 0.9815 - val_loss: nan - val_accuracy: 0.9816\n",
"Epoch 10/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.0572 - accuracy: 0.9825 - val_loss: nan - val_accuracy: 0.9823\n",
"Epoch 11/20\n",
"108/108 [==============================] - 14s 127ms/step - loss: 0.0511 - accuracy: 0.9842 - val_loss: nan - val_accuracy: 0.9836\n",
"Epoch 12/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.0519 - accuracy: 0.9840 - val_loss: nan - val_accuracy: 0.9839\n",
"Epoch 13/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.0494 - accuracy: 0.9850 - val_loss: nan - val_accuracy: 0.9813\n",
"Epoch 14/20\n",
"108/108 [==============================] - 14s 127ms/step - loss: 0.0499 - accuracy: 0.9847 - val_loss: nan - val_accuracy: 0.9837\n",
"Epoch 15/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.0471 - accuracy: 0.9856 - val_loss: nan - val_accuracy: 0.9838\n",
"Epoch 16/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.0411 - accuracy: 0.9874 - val_loss: nan - val_accuracy: 0.9843\n",
"Epoch 17/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.0378 - accuracy: 0.9883 - val_loss: nan - val_accuracy: 0.9851\n",
"Epoch 18/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.0368 - accuracy: 0.9886 - val_loss: nan - val_accuracy: 0.9853\n",
"Epoch 19/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.0386 - accuracy: 0.9883 - val_loss: nan - val_accuracy: 0.9845\n",
"Epoch 20/20\n",
"108/108 [==============================] - 14s 126ms/step - loss: 0.0459 - accuracy: 0.9863 - val_loss: nan - val_accuracy: 0.9843\n"
]
},
{
"data": {
"text/plain": [
"<keras.callbacks.History at 0x2573f340d90>"
]
},
"execution_count": 15,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"def bidirectional_embed_model(input_shape, output_sequence_length, english_vocab_size, french_vocab_size):\n",
" \n",
" # Hyperparameters\n",
" learning_rate = 0.005\n",
" \n",
" # Build the layers\n",
" model = Sequential()\n",
" model.add(Embedding(english_vocab_size, 256, input_length=input_shape[1], input_shape=input_shape[1:]))\n",
" model.add(Bidirectional(GRU(256, return_sequences=True)))\n",
" model.add(TimeDistributed(Dense(1024, activation='relu')))\n",
" model.add(Dropout(0.5))\n",
" model.add(TimeDistributed(Dense(french_vocab_size, activation='softmax')))\n",
" \n",
" # Compile model\n",
" model.compile(loss = sparse_categorical_crossentropy,\n",
" optimizer = Adam(learning_rate),\n",
" metrics = ['accuracy'])\n",
" \n",
" return model\n",
"\n",
"tmp_x = pad(preproc_english_sentences, max_french_sequence_length)\n",
"tmp_x = tmp_x.reshape((-1, preproc_french_sentences.shape[-2]))\n",
"\n",
"# Build the model\n",
"embed_rnn_model = bidirectional_embed_model(\n",
" tmp_x.shape,\n",
" max_french_sequence_length,\n",
" english_vocab_size,\n",
" french_vocab_size)\n",
"\n",
"print(embed_rnn_model.summary())\n",
"\n",
"embed_rnn_model.fit(tmp_x, preproc_french_sentences, batch_size=1024, epochs=20, validation_split=0.2)\n",
" "
]
},
{
"cell_type": "code",
"execution_count": 16,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Prediciton:\n",
"1/1 [==============================] - 0s 410ms/step\n",
"new jersey est parfois calme pendant l' automne et il est neigeux en avril <PAD> <PAD> <PAD> <PAD> <PAD> <PAD> <PAD>\n",
"\n",
"Correct Translation:\n",
"[\"new jersey est parfois calme pendant l' automne , et il est neigeux en avril .\"]\n",
"\n",
"Original text:\n",
"['new jersey is sometimes quiet during autumn , and it is snowy in april .']\n"
]
}
],
"source": [
"# Print prediction(s)\n",
"print(\"Prediciton:\")\n",
"print(logits_to_text(embed_rnn_model.predict(tmp_x[:1])[0], french_tokenizer))\n",
"\n",
"print(\"\\nCorrect Translation:\")\n",
"print(french_sentences[:1])\n",
"\n",
"print('\\nOriginal text:')\n",
"print(english_sentences[:1])"
]
},
{
"cell_type": "code",
"execution_count": 17,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"WARNING:absl:Found untraced functions such as gru_cell_5_layer_call_fn, gru_cell_5_layer_call_and_return_conditional_losses, gru_cell_6_layer_call_fn, gru_cell_6_layer_call_and_return_conditional_losses while saving (showing 4 of 4). These functions will not be directly callable after loading.\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"INFO:tensorflow:Assets written to: english_to_french_model\\assets\n"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"INFO:tensorflow:Assets written to: english_to_french_model\\assets\n"
]
}
],
"source": [
"embed_rnn_model.save('english_to_french_model')\n",
"# Serialize English Tokenizer to JSON\n",
"with open('english_tokenizer.json', 'w', encoding='utf8') as f:\n",
" f.write(json.dumps(english_tokenizer.to_json(), ensure_ascii=False))\n",
" \n",
"# Serialize French Tokenizer to JSON\n",
"with open('french_tokenizer.json', 'w', encoding='utf8') as f:\n",
" f.write(json.dumps(french_tokenizer.to_json(), ensure_ascii=False))\n",
" \n",
"# Save max lengths\n",
"max_french_sequence_length_json = max_french_sequence_length\n",
"with open('sequence_length.json', 'w', encoding='utf8') as f:\n",
" f.write(json.dumps(max_french_sequence_length_json, ensure_ascii=False))"
]
},
{
"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": "3.9.19"
}
},
"nbformat": 4,
"nbformat_minor": 1
}
|