6.2-循环神经网络进阶.ipynb

{"cells":[{"outputs":[],"execution_count":1,"source":"import numpy as np\nimport torch\nfrom torch import nn, optim\nimport torch.nn.functional as F\nimport sys\nsys.path.append(\"../input/\")\nimport d2l_jay9460 as d2l\ndevice = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n\n(corpus_indices, char_to_idx, idx_to_char, vocab_size) = d2l.load_data_jay_lyrics()","cell_type":"code","metadata":{"trusted":true,"collapsed":false,"jupyter":{},"tags":[],"slideshow":{"slide_type":"slide"},"id":"E26FEA30BD3147EB88BBA80310F0EB08","scrolled":false}},{"metadata":{"id":"C3ED4362643C47D38B69B4C0FFB6B2C5","jupyter":{},"tags":[],"slideshow":{"slide_type":"slide"},"collapsed":false,"scrolled":false},"cell_type":"code","outputs":[],"source":"## GRU简洁实现\n# 各参数定义\nnum_hiddens=256\nnum_epochs, num_steps, batch_size, lr, clipping_theta = 160, 35, 32, 1e2, 1e-2\npred_period, pred_len, prefixes = 40, 50, ['分开', '不分开']\nlr = 1e-2 # 注意调整学习率\n# 直接调用pytor.nn模块定义GRU网络","execution_count":3},{"metadata":{"id":"D8BBC078D5DA43FFB4E44E2470FD2FA8","jupyter":{},"tags":[],"slideshow":{"slide_type":"slide"}},"cell_type":"code","outputs":[],"source":"gru_layer = nn.GRU(input_size=vocab_size, hidden_size=num_hiddens)\nmodel = d2l.RNNModel(gru_layer, vocab_size).to(device)  # 调用之前的语言模型函数搭建完整的神经网络\n\nd2l.train_and_predict_rnn_pytorch(model, num_hiddens, vocab_size, device,\n                                corpus_indices, idx_to_char, char_to_idx,\n                                num_epochs, num_steps, lr, clipping_theta,\n                                batch_size, pred_period, pred_len, prefixes)","execution_count":null},{"metadata":{"id":"F6390DAC65DE4E948D128A62F8243E58","jupyter":{},"tags":[],"slideshow":{"slide_type":"slide"},"collapsed":false,"scrolled":false},"cell_type":"code","outputs":[{"output_type":"stream","text":"epoch 40, perplexity 1.012188, time 0.89 sec\n - 分开的玩笑 想通 却又再考倒我 说散 你想很久了吧? 败给你的黑色幽默 说散 你想很久了吧? 我的认真败\n - 不分开不 我不要再想你 爱情来的太快就像龙卷风 离不开暴风圈来不及逃 我不能再想 我不能再想 我不 我不 \nepoch 80, perplexity 2.042774, time 0.92 sec\n - 分开已在蓝娘子依旧每日折一枝杨柳 在小村外的溪边  情绪激动 一颗两颗三颗四颗 连成线背著背默默许下心愿\n - 不分开始想懂不懂 一直走 离不开暴风圈来不及逃 我跟了这节奏 后知后觉 我该好好生活 我该好好生活 我该好\nepoch 120, perplexity 1.049309, time 0.90 sec\n - 分开的春联还残 有一直在一起个人慢温暖  为什么我爸爸一直到我们都睡着 我想就这样牵着你的手不放开 爱可\n - 不分开一定都会有 双截棍 哼哼哈兮 快使用双截棍 哼哼哈兮 如果我有轻功 飞到底 有样牵着你想的太快就像龙\nepoch 160, perplexity 1.017079, time 0.92 sec\n - 分开的春联还残 有一直在一起个人慢温暖  为什么我爸爸种种的见 它一定实现 娘子 娘子却依旧每日 折一枝\n - 不分开一定都会有  我马儿有些瘦 天涯尽头 满脸风霜 不要再这样打我妈妈 难道你手 你想要再想 我 一口一\n","name":"stdout"}],"source":"## LSTM简洁实现\n# 各参数定义（同上，略）\n# 直接调用pytor.nn模块定义LSTM网络\nlstm_layer = nn.LSTM(input_size=vocab_size, hidden_size=num_hiddens)\nmodel = d2l.RNNModel(gru_layer, vocab_size).to(device)  # 调用之前的语言模型函数搭建完整的神经网络（同上）    \nd2l.train_and_predict_rnn_pytorch(model, num_hiddens, vocab_size, device,  # 同上\n                                corpus_indices, idx_to_char, char_to_idx,\n                                num_epochs, num_steps, lr, clipping_theta,\n                                batch_size, pred_period, pred_len, prefixes)","execution_count":7},{"metadata":{"id":"FED3339AE82A4CF5A7EE32EE03B6A665","jupyter":{},"tags":[],"slideshow":{"slide_type":"slide"},"collapsed":false,"scrolled":false},"cell_type":"code","outputs":[],"source":"# 双向LSTM\nlstm_layer = nn.LSTM(input_size=vocab_size, hidden_size=num_hiddens, bidirectional=True)  # bidirectional选择是否双向\nmodel = d2l.RNNModel(lstm_layer, vocab_size).to(device)","execution_count":4},{"metadata":{"id":"7B6ACC3E2F414562BFC9AA57CFB62EA0","jupyter":{},"tags":[],"slideshow":{"slide_type":"slide"},"collapsed":false,"scrolled":false},"cell_type":"code","outputs":[{"output_type":"stream","text":"epoch 40, perplexity 1.000804, time 1.70 sec\n - 分开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开\n - 不分开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开\nepoch 80, perplexity 1.000270, time 1.71 sec\n - 分开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开\n - 不分开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开\nepoch 120, perplexity 1.000138, time 1.69 sec\n - 分开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开\n - 不分开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开\nepoch 160, perplexity 1.000083, time 1.79 sec\n - 分开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开始开\n - 不分开始开始开始开始开始开开开开开开开开开开开开开开开开开开开开开开开开开开开开开开开开开开开开开开开开开\n","name":"stdout"}],"source":"d2l.train_and_predict_rnn_pytorch(model, num_hiddens, vocab_size, device,\n                                corpus_indices, idx_to_char, char_to_idx,\n                                num_epochs, num_steps, lr, clipping_theta,\n                                batch_size, pred_period, pred_len, prefixes)","execution_count":5},{"metadata":{"id":"313FA7C163DD4D418B1E600234754933","jupyter":{},"tags":[],"slideshow":{"slide_type":"slide"},"collapsed":false,"scrolled":false},"cell_type":"code","outputs":[{"output_type":"execute_result","metadata":{},"data":{"text/plain":"[53,\n 314,\n 648,\n 155,\n 720,\n 694,\n 880,\n 53,\n 314,\n 609,\n 312,\n 68,\n 399,\n 505,\n 61,\n 566,\n 880,\n 53,\n 314,\n 609,\n 312,\n 128,\n 1004,\n 243,\n 953,\n 922,\n 880,\n 128,\n 1004,\n 243,\n 505,\n 61,\n 987,\n 880,\n 148,\n 575,\n 490,\n 575,\n 490,\n 575,\n 490,\n 243,\n 53,\n 53,\n 53,\n 53,\n 914,\n 312,\n 880,\n 277,\n 230,\n 534,\n 207,\n 174,\n 880,\n 36,\n 148,\n 743,\n 527,\n 487,\n 398,\n 385,\n 356,\n 160,\n 880,\n 239,\n 748,\n 119,\n 634,\n 765,\n 433,\n 880,\n 36,\n 148,\n 54,\n 399,\n 312,\n 880,\n 828,\n 397,\n 534,\n 36,\n 148,\n 24,\n 315,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 719,\n 268,\n 534,\n 36,\n 148,\n 634,\n 328,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 193,\n 705,\n 534,\n 36,\n 148,\n 239,\n 299,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 236,\n 236,\n 534,\n 36,\n 148,\n 597,\n 523,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 236,\n 236,\n 534,\n 36,\n 148,\n 597,\n 523,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 828,\n 397,\n 534,\n 36,\n 148,\n 24,\n 315,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 719,\n 268,\n 534,\n 36,\n 148,\n 634,\n 328,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 193,\n 705,\n 534,\n 36,\n 148,\n 239,\n 299,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 236,\n 236,\n 534,\n 36,\n 148,\n 597,\n 523,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 236,\n 236,\n 534,\n 36,\n 148,\n 597,\n 523,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 53,\n 314,\n 648,\n 155,\n 720,\n 694,\n 880,\n 53,\n 314,\n 609,\n 312,\n 68,\n 399,\n 505,\n 61,\n 566,\n 880,\n 53,\n 314,\n 609,\n 312,\n 128,\n 1004,\n 243,\n 953,\n 922,\n 880,\n 128,\n 1004,\n 243,\n 505,\n 61,\n 987,\n 880,\n 148,\n 575,\n 490,\n 575,\n 490,\n 575,\n 490,\n 243,\n 53,\n 53,\n 53,\n 53,\n 914,\n 312,\n 880,\n 277,\n 230,\n 534,\n 207,\n 174,\n 880,\n 36,\n 148,\n 743,\n 527,\n 487,\n 398,\n 385,\n 356,\n 160,\n 880,\n 239,\n 748,\n 119,\n 634,\n 765,\n 433,\n 880,\n 36,\n 148,\n 54,\n 399,\n 312,\n 880,\n 828,\n 397,\n 534,\n 36,\n 148,\n 24,\n 315,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 719,\n 268,\n 534,\n 36,\n 148,\n 634,\n 328,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 193,\n 705,\n 534,\n 36,\n 148,\n 239,\n 299,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 236,\n 236,\n 534,\n 36,\n 148,\n 597,\n 523,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 236,\n 236,\n 534,\n 36,\n 148,\n 597,\n 523,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 828,\n 397,\n 534,\n 36,\n 148,\n 24,\n 315,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 719,\n 268,\n 534,\n 36,\n 148,\n 634,\n 328,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 193,\n 705,\n 534,\n 36,\n 148,\n 239,\n 299,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 236,\n 236,\n 534,\n 36,\n 148,\n 597,\n 523,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 236,\n 236,\n 534,\n 36,\n 148,\n 597,\n 523,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 828,\n 397,\n 534,\n 36,\n 148,\n 24,\n 315,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 719,\n 268,\n 534,\n 36,\n 148,\n 634,\n 328,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 193,\n 705,\n 534,\n 36,\n 148,\n 239,\n 299,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 236,\n 236,\n 534,\n 36,\n 148,\n 597,\n 523,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 236,\n 236,\n 534,\n 36,\n 148,\n 597,\n 523,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 828,\n 397,\n 534,\n 36,\n 148,\n 24,\n 315,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 719,\n 268,\n 534,\n 36,\n 148,\n 634,\n 328,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 193,\n 705,\n 534,\n 36,\n 148,\n 239,\n 299,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 236,\n 236,\n 534,\n 36,\n 148,\n 597,\n 523,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 236,\n 236,\n 534,\n 36,\n 148,\n 597,\n 523,\n 534,\n 774,\n 578,\n 41,\n 541,\n 880,\n 859,\n 602,\n 840,\n 665,\n 184,\n 880,\n 774,\n 746,\n 930,\n 574,\n 859,\n 602,\n 840,\n 604,\n 984,\n 880,\n 1008,\n 617,\n 140,\n 621,\n 880,\n 859,\n 602,\n 277,\n 953,\n 221,\n 880,\n 801,\n 534,\n 774,\n 746,\n 880,\n 148,\n 53,\n 314,\n 639,\n 148,\n 534,\n 1013,\n 542,\n 972,\n 386,\n 880,\n 319,\n 760,\n 243,\n 277,\n 987,\n 880,\n 70,\n 514,\n 264,\n 173,\n 880,\n 319,\n 760,\n 639,\n 799,\n 566,\n 880,\n 237,\n 237,\n 719,\n 153,\n 880,\n 36,\n 148,\n 578,\n 1023,\n 312,\n 880,\n 737,\n 249,\n 834,\n 915,\n 880,\n 30,\n 312,\n 117,\n 735,\n 182,\n 4,\n 534,\n 953,\n 249,\n 71,\n 880,\n 635,\n 107,\n 634,\n 628,\n 493,\n 920,\n 880,\n 284,\n 75,\n 75,\n 390,\n 551,\n 880,\n 312,\n 578,\n 799,\n 148,\n 534,\n 738,\n 839,\n 880,\n 36,\n 588,\n 1022,\n 534,\n 451,\n 884,\n 880,\n 769,\n 615,\n 334,\n 678,\n 879,\n 880,\n 253,\n 541,\n 534,\n 735,\n 910,\n 880,\n 312,\n 534,\n 117,\n 735,\n 664,\n 625,\n 880,\n 868,\n 966,\n 511,\n 880,\n 36,\n 148,\n 975,\n 273,\n 363,\n 455,\n 746,\n 937,\n 371,\n 880,\n 639,\n 801,\n 634,\n 611,\n 984,\n 533,\n 334,\n 264,\n 551,\n 880,\n 585,\n 30,\n 953,\n 249,\n 95,\n 915,\n 880,\n 312,\n 534,\n 117,\n 735,\n 664,\n 625,\n 880,\n 868,\n 966,\n 511,\n 880,\n 604,\n 1011,\n 534,\n 344,\n 585,\n 244,\n 728,\n 950,\n 433,\n 880,\n 148,\n 212,\n 796,\n 1008,\n 245,\n 433,\n 284,\n 372,\n 922,\n 880,\n 736,\n 256,\n 816,\n 353,\n 964,\n 780,\n 880,\n 319,\n 760,\n 243,\n 277,\n 987,\n 880,\n 70,\n 514,\n 264,\n 173,\n 880,\n 319,\n 760,\n 639,\n 799,\n 566,\n 880,\n 237,\n 237,\n 719,\n 153,\n 880,\n 36,\n 148,\n 578,\n 1023,\n 312,\n 880,\n 737,\n 249,\n 834,\n 915,\n 880,\n 30,\n 312,\n 117,\n 735,\n 182,\n 4,\n 534,\n 953,\n 249,\n 71,\n 880,\n 635,\n 107,\n 634,\n 628,\n 493,\n 920,\n 880,\n 284,\n 75,\n 75,\n 390,\n 551,\n 880,\n 312,\n 578,\n 799,\n 148,\n 534,\n 738,\n 839,\n 880,\n 36,\n 588,\n 1022,\n 534,\n 451,\n 884,\n 880,\n 769,\n 615,\n 334,\n 678,\n 879,\n 880,\n 253,\n 541,\n 534,\n 735,\n 910,\n 880,\n 312,\n 534,\n 117,\n 735,\n 664,\n 625,\n 880,\n 868,\n 966,\n 511,\n 880,\n 36,\n 148,\n 975,\n 273,\n 363,\n 455,\n 746,\n 937,\n 371,\n 880,\n 639,\n 801,\n 634,\n 611,\n 984,\n 533,\n 334,\n 264,\n 551,\n 880,\n 585,\n 30,\n 953,\n 249,\n 95,\n 915,\n 880,\n 312,\n 534,\n 117,\n 735,\n 664,\n 625,\n 880,\n 868,\n 966,\n 511,\n 880,\n 604,\n 1011,\n 534,\n 344,\n 585,\n 244,\n 728,\n 950,\n 433,\n 880,\n 148,\n 212,\n 796,\n 1008,\n 245,\n 433,\n 284,\n 372,\n 922,\n 880,\n 736,\n 256,\n 816,\n 353,\n 964,\n 780,\n 880,\n 115,\n 775,\n 197,\n 880,\n 115,\n 775,\n 197,\n 880,\n 953,\n 475,\n 240,\n 475,\n 679,\n 475,\n 409,\n 475,\n 706,\n 914,\n 490,\n 880,\n 692,\n 1010,\n 1010,\n 880,\n 953,\n 699,\n 240,\n 699,\n 679,\n 699,\n 409,\n 699,\n 880,\n 975,\n 334,\n 260,\n 953,\n 475,\n 240,\n 475,\n 679,\n 475,\n 409,\n 475,\n 706,\n 914,\n 490,\n 880,\n 692,\n 1010,\n 1010,\n 880,\n 953,\n 699,\n 240,\n 699,\n 679,\n 699,\n 409,\n 699,\n 880,\n 975,\n 334,\n 260,\n 303,\n 914,\n 188,\n 880,\n 1024,\n 710,\n 243,\n 5,\n 490,\n 845,\n 880,\n 953,\n 589,\n 11,\n 121,\n 831,\n 243,\n 148,\n 24,\n 20,\n 880,\n 441,\n 334,\n 312,\n 534,\n 218,\n 491,\n ...]"},"transient":{},"execution_count":19}],"source":"# len(corpus_indices)\n# vocab_size\ncorpus_indices","execution_count":19}],"metadata":{"kernelspec":{"name":"python3","display_name":"Python 3","language":"python"},"language_info":{"name":"python","version":"3.7.3","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"}},"nbformat":4,"nbformat_minor":0}