动手实战人工智能 AI By Doing

自动化深度学习实践

huhuhang 更新于：2024-08-31 12:05:12

cover
Pro 计划

Pro 会员计划

Pro 计划是作者为了维护和更新本教程而推出的会员计划，你可以获得更多的福利和服务，点击查看详情。

78. 自动化深度学习实践#

78.1. 介绍#

前面，我们学习了深度学习框架 TensorFlow，其高阶接口 Keras 以良好的易用性深受欢迎。本次实验中，我们将了解到 Keras 团队维护的自动化深度学习框架 Auto-Keras，并使用其完成基础的应用。

78.2. 知识点#

  • Auto-Keras 介绍

  • 图像分类任务

  • 文本分类任务

  • 最优模型可视化

  • AutoML 优劣分析

78.3. Auto-Keras 介绍#

本次实验中，我们选择 Auto-Keras 框架来介绍自动化深度学习的应用。Auto-Keras 是由 DATA Lab 开发，由 Keras 官方团队维护的深度学习框架。目前来讲，Auto-Keras 应该是自动化深度学习应用的理想选择之一。

https://cdn.aibydoing.com/aibydoing/images/uid214893-20190702-1562036444816.png

深度神经网络结构是深度学习的核心，合适的结构和功能层决定了深度学习的效率和结果。实际生产环境中，我们通常会沿用一些经典的网络结构，并根据机器学习专家的经验进行调整。而这个过程对于具有有限数据科学或机器学习背景的人士而言是极其困难的。

Auto-Keras 主要是提供了神经结构搜索（NAS）和超参数自动优化（HPO）这两项功能。其中，Auto-Keras 使用了 4 种常用的 NAS 方法：

  • 随机搜索：通过随机调整网络结构来探索搜索空间，因此生成的神经网络结构的实际性能对后续的搜索没有影响。

  • 网格搜索：通过手动指定超参数空间的子集进行搜索，即网络功能层的数量和层的宽度是预定义的。

  • 贪婪搜索：以贪婪的方式探索搜索空间。这里的「贪婪」意味着下一次搜索迭代的网络基础结构是从当前迭代生成的，对训练/验证集具有最佳性能的结构。

  • 贝叶斯优化：它是目前 Auto-Keras 的默认搜索策略。有关更多详细信息，请 参阅论文。

通过 Auto-Keras 官方仓库 可以看出，目前其后端主要依赖于 scikit-learn 和 TensorFlow 2 进行开发。

安装 Auto-Keras，可以使用 pip install autokeras 命令。

78.4. 计算机视觉#

计算机视觉是深度学习中最常见的应用场景，其中主要有：图像分类、图像生成，对象检测、目标跟踪、语义分割、实例分割等。目前，Auto-Keras 可以完成图像分类，而图像分割的相关工具类还在构建当中。

78.4.1. 图像分类#

例如前面实验中完成的 DIGITS 和 MNIST 手写字符识别，实际上就是一个图像分类任务。Auto-Keras 中用于图像分类的类为 autokeras.ImageClassifier。

其中，Auto-Keras 支持 2 种情形下的图像分类任务。分别是：

  • 图片已处理为 NumPy 数组，例如前面实验中使用过的 DIGITS 和 MNIST 数据集。

  • 图片仍未原始的 JPG，PNG 等格式，需要进行预处理步骤。

接下来，我们尝试使用 Auto-Keras 来完成这两种情形下的应用。对于已经被处理成 NumPy 数组的任务，我们同样选择 MNIST 手写字符分类。这里选择 TensorFlow 自带的数据 API 加载数据，并完成导入。 

import tensorflow as tf

# 加载数据集
(X_train, y_train), (X_test, y_test) = tf.keras.datasets.mnist.load_data()
# 查看数组形状
X_train.shape, y_train.shape, X_test.shape, y_test.shape

((60000, 28, 28), (60000,), (10000, 28, 28), (10000,))

可以看到，训练集有 6 万个样本，测试集有 1 万个样本，样本数据形状为 \(28 \times 28\) 的灰度图像。

接下来，我们导入 Auto-Keras 提供的图像分类的类，并完成模型训练过程。模型类主要有 4 个方法，分别是：

  • fit：用于传入 NumPy 数组训练模型。

  • predict：用于模型的推理过程。

  • evaluate：用于模型的评估过程。

  • export_model:选取测试模型中最佳的模型并返回。

让我们尝试使用该函数对 MNIST 手写字符进行识别（下面代码总共需要进行二次训练，所以请耐心等待 5~7 min）： 

import autokeras as ak

# 实例化模型，max_trials=1 ，表示尝试的最大的 神经网络模型数
clf = ak.ImageClassifier(max_trials=1)
# epochs 表示每个模型训练的最大世代数
# batch_size 指定每个 batch 的大小
clf.fit(X_train, y_train, batch_size=1000, epochs=1)
print("训练完成")

Trial 1 Complete [00h 00m 24s]
val_loss: 0.2161960005760193

Best val_loss So Far: 0.2161960005760193
Total elapsed time: 00h 00m 24s

60/60 [==============================] - 27s 441ms/step - loss: 0.5438 - accuracy: 0.8388
INFO:tensorflow:Assets written to: ./image_classifier/best_model/assets

训练完成

由于模型的训练时间很长，因此上面代码只尝试了 1 个模型，即 max_trials=1（在实际实践中，你可以根据自身需要，调节想要尝试的模型数量）。

在训练过程中，Auto-Keras 会分别对 max_trials 个模型进行训练。训练过程中，Autokeras 会把训练数据分成两份，一份用于训练，一份用于评估。然后根据评估结果，选出这些模型中最佳的模型结构，再把所有数据放入该结构中进行训练。因此，如果需要从 max_trials 个模型中输出最佳模型，就必须训练 max_trials+1 次。

现在让我利用选出的最佳模型对测试数据进行测试：

clf.evaluate(X_test, y_test)

313/313 [==============================] - 2s 6ms/step - loss: 0.1587 - accuracy: 0.9566

[0.1587049812078476, 0.95660001039505]

从结果可以看出，利用 Auto-Keras ，我们将搭建神经网络的工作也交给计算机来做，虽然为了找到最佳模型会花费一些时间，但是也能训练出不错的模型。

最后，让我们将模型从 Auto-Keras 中导出，得到我们熟悉的 TensorFlow 模型并展示出来： 

model = clf.export_model()
model.summary()

Model: "model"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_1 (InputLayer)        [(None, 28, 28)]          0         
                                                                 
 cast_to_float32 (CastToFlo  (None, 28, 28)            0         
 at32)                                                           
                                                                 
 expand_last_dim (ExpandLas  (None, 28, 28, 1)         0         
 tDim)                                                           
                                                                 
 normalization (Normalizati  (None, 28, 28, 1)         3         
 on)                                                             
                                                                 
 conv2d (Conv2D)             (None, 26, 26, 32)        320       
                                                                 
 conv2d_1 (Conv2D)           (None, 24, 24, 64)        18496     
                                                                 
 max_pooling2d (MaxPooling2  (None, 12, 12, 64)        0         
 D)                                                              
                                                                 
 dropout (Dropout)           (None, 12, 12, 64)        0         
                                                                 
 flatten (Flatten)           (None, 9216)              0         
                                                                 
 dropout_1 (Dropout)         (None, 9216)              0         
                                                                 
 dense (Dense)               (None, 10)                92170     
                                                                 
 classification_head_1 (Sof  (None, 10)                0         
 tmax)                                                           
                                                                 
=================================================================
Total params: 110989 (433.55 KB)
Trainable params: 110986 (433.54 KB)
Non-trainable params: 3 (16.00 Byte)
_________________________________________________________________

如上图所示，即为 Auto-Keras 选择的最终模型结构。

由此可见，Auto-Keras 的使用的确十分简单，你完全不需要考虑如何构建网络，定义损失函数等，只需要将数据传入 API 即可。

如图所示，传统的机器学习从数据处理开始，到最终完成预测大致需要经过 8 个步骤，而 AutoML 流程可以简化为 2 个步骤，至于数据预处理、模型选择、超参数调优等统统都可以交给框架完成。

78.5. 自然语言处理#

除了计算机视觉，自然语言处理也是深度学习重要应用方向。自然语言处理主要研究人与计算机之间用自然语言进行有效通信的各种理论和方法。目前，自然语言处理有着十分丰富的应用，主要有：信息检索、语音识别、机器翻译、智能问答、对话系统、文本分类、情感分析、文本生成、自动文摘等。

78.5.1. 文本分类#

Auto-Keras中用于完成文本分类的类是 autokeras.TextClassifier。接下来，我们尝试使用 Auto-Keras 完成 IMDB 电影评论文本分类预测。

IMDB 数据来源于著名的互联网电影数据库。该数据集为电影评论数据，并被打上了积极（1）或消极（0）的标签。

我们可以利用 TensorFlow 官方提供的接口加载数据集合：

import autokeras as ak
import numpy as np
import tensorflow as tf

# 加载 IMDB数据并对它进行处理
max_features = 20000
index_offset = 3
# 加载数据
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.imdb.load_data(
    num_words=max_features,
    index_from=index_offset)
x_train = x_train
y_train = y_train.reshape(-1, 1)
x_test = x_test
y_test = y_test.reshape(-1, 1)
x_train

array([list([1, 14, 22, 16, 43, 530, 973, 1622, 1385, 65, 458, 4468, 66, 3941, 4, 173, 36, 256, 5, 25, 100, 43, 838, 112, 50, 670, 2, 9, 35, 480, 284, 5, 150, 4, 172, 112, 167, 2, 336, 385, 39, 4, 172, 4536, 1111, 17, 546, 38, 13, 447, 4, 192, 50, 16, 6, 147, 2025, 19, 14, 22, 4, 1920, 4613, 469, 4, 22, 71, 87, 12, 16, 43, 530, 38, 76, 15, 13, 1247, 4, 22, 17, 515, 17, 12, 16, 626, 18, 19193, 5, 62, 386, 12, 8, 316, 8, 106, 5, 4, 2223, 5244, 16, 480, 66, 3785, 33, 4, 130, 12, 16, 38, 619, 5, 25, 124, 51, 36, 135, 48, 25, 1415, 33, 6, 22, 12, 215, 28, 77, 52, 5, 14, 407, 16, 82, 10311, 8, 4, 107, 117, 5952, 15, 256, 4, 2, 7, 3766, 5, 723, 36, 71, 43, 530, 476, 26, 400, 317, 46, 7, 4, 12118, 1029, 13, 104, 88, 4, 381, 15, 297, 98, 32, 2071, 56, 26, 141, 6, 194, 7486, 18, 4, 226, 22, 21, 134, 476, 26, 480, 5, 144, 30, 5535, 18, 51, 36, 28, 224, 92, 25, 104, 4, 226, 65, 16, 38, 1334, 88, 12, 16, 283, 5, 16, 4472, 113, 103, 32, 15, 16, 5345, 19, 178, 32]),
       list([1, 194, 1153, 194, 8255, 78, 228, 5, 6, 1463, 4369, 5012, 134, 26, 4, 715, 8, 118, 1634, 14, 394, 20, 13, 119, 954, 189, 102, 5, 207, 110, 3103, 21, 14, 69, 188, 8, 30, 23, 7, 4, 249, 126, 93, 4, 114, 9, 2300, 1523, 5, 647, 4, 116, 9, 35, 8163, 4, 229, 9, 340, 1322, 4, 118, 9, 4, 130, 4901, 19, 4, 1002, 5, 89, 29, 952, 46, 37, 4, 455, 9, 45, 43, 38, 1543, 1905, 398, 4, 1649, 26, 6853, 5, 163, 11, 3215, 10156, 4, 1153, 9, 194, 775, 7, 8255, 11596, 349, 2637, 148, 605, 15358, 8003, 15, 123, 125, 68, 2, 6853, 15, 349, 165, 4362, 98, 5, 4, 228, 9, 43, 2, 1157, 15, 299, 120, 5, 120, 174, 11, 220, 175, 136, 50, 9, 4373, 228, 8255, 5, 2, 656, 245, 2350, 5, 4, 9837, 131, 152, 491, 18, 2, 32, 7464, 1212, 14, 9, 6, 371, 78, 22, 625, 64, 1382, 9, 8, 168, 145, 23, 4, 1690, 15, 16, 4, 1355, 5, 28, 6, 52, 154, 462, 33, 89, 78, 285, 16, 145, 95]),
       list([1, 14, 47, 8, 30, 31, 7, 4, 249, 108, 7, 4, 5974, 54, 61, 369, 13, 71, 149, 14, 22, 112, 4, 2401, 311, 12, 16, 3711, 33, 75, 43, 1829, 296, 4, 86, 320, 35, 534, 19, 263, 4821, 1301, 4, 1873, 33, 89, 78, 12, 66, 16, 4, 360, 7, 4, 58, 316, 334, 11, 4, 1716, 43, 645, 662, 8, 257, 85, 1200, 42, 1228, 2578, 83, 68, 3912, 15, 36, 165, 1539, 278, 36, 69, 2, 780, 8, 106, 14, 6905, 1338, 18, 6, 22, 12, 215, 28, 610, 40, 6, 87, 326, 23, 2300, 21, 23, 22, 12, 272, 40, 57, 31, 11, 4, 22, 47, 6, 2307, 51, 9, 170, 23, 595, 116, 595, 1352, 13, 191, 79, 638, 89, 2, 14, 9, 8, 106, 607, 624, 35, 534, 6, 227, 7, 129, 113]),
       ...,
       list([1, 11, 6, 230, 245, 6401, 9, 6, 1225, 446, 2, 45, 2174, 84, 8322, 4007, 21, 4, 912, 84, 14532, 325, 725, 134, 15271, 1715, 84, 5, 36, 28, 57, 1099, 21, 8, 140, 8, 703, 5, 11656, 84, 56, 18, 1644, 14, 9, 31, 7, 4, 9406, 1209, 2295, 2, 1008, 18, 6, 20, 207, 110, 563, 12, 8, 2901, 17793, 8, 97, 6, 20, 53, 4767, 74, 4, 460, 364, 1273, 29, 270, 11, 960, 108, 45, 40, 29, 2961, 395, 11, 6, 4065, 500, 7, 14492, 89, 364, 70, 29, 140, 4, 64, 4780, 11, 4, 2678, 26, 178, 4, 529, 443, 17793, 5, 27, 710, 117, 2, 8123, 165, 47, 84, 37, 131, 818, 14, 595, 10, 10, 61, 1242, 1209, 10, 10, 288, 2260, 1702, 34, 2901, 17793, 4, 65, 496, 4, 231, 7, 790, 5, 6, 320, 234, 2766, 234, 1119, 1574, 7, 496, 4, 139, 929, 2901, 17793, 7750, 5, 4241, 18, 4, 8497, 13164, 250, 11, 1818, 7561, 4, 4217, 5408, 747, 1115, 372, 1890, 1006, 541, 9303, 7, 4, 59, 11027, 4, 3586, 2]),
       list([1, 1446, 7079, 69, 72, 3305, 13, 610, 930, 8, 12, 582, 23, 5, 16, 484, 685, 54, 349, 11, 4120, 2959, 45, 58, 1466, 13, 197, 12, 16, 43, 23, 2, 5, 62, 30, 145, 402, 11, 4131, 51, 575, 32, 61, 369, 71, 66, 770, 12, 1054, 75, 100, 2198, 8, 4, 105, 37, 69, 147, 712, 75, 3543, 44, 257, 390, 5, 69, 263, 514, 105, 50, 286, 1814, 23, 4, 123, 13, 161, 40, 5, 421, 4, 116, 16, 897, 13, 2, 40, 319, 5872, 112, 6700, 11, 4803, 121, 25, 70, 3468, 4, 719, 3798, 13, 18, 31, 62, 40, 8, 7200, 4, 2, 7, 14, 123, 5, 942, 25, 8, 721, 12, 145, 5, 202, 12, 160, 580, 202, 12, 6, 52, 58, 11418, 92, 401, 728, 12, 39, 14, 251, 8, 15, 251, 5, 2, 12, 38, 84, 80, 124, 12, 9, 23]),
       list([1, 17, 6, 194, 337, 7, 4, 204, 22, 45, 254, 8, 106, 14, 123, 4, 12815, 270, 14437, 5, 16923, 12255, 732, 2098, 101, 405, 39, 14, 1034, 4, 1310, 9, 115, 50, 305, 12, 47, 4, 168, 5, 235, 7, 38, 111, 699, 102, 7, 4, 4039, 9245, 9, 24, 6, 78, 1099, 17, 2345, 16553, 21, 27, 9685, 6139, 5, 2, 1603, 92, 1183, 4, 1310, 7, 4, 204, 42, 97, 90, 35, 221, 109, 29, 127, 27, 118, 8, 97, 12, 157, 21, 6789, 2, 9, 6, 66, 78, 1099, 4, 631, 1191, 5, 2642, 272, 191, 1070, 6, 7585, 8, 2197, 2, 10755, 544, 5, 383, 1271, 848, 1468, 12183, 497, 16876, 8, 1597, 8778, 19280, 21, 60, 27, 239, 9, 43, 8368, 209, 405, 10, 10, 12, 764, 40, 4, 248, 20, 12, 16, 5, 174, 1791, 72, 7, 51, 6, 1739, 22, 4, 204, 131, 9])],
      dtype=object)

从结果可以看出此时的数据文本已经被转为了词向量，但是由于 Autokeras 的文本分类器接收的是真实文本，因此我们还需要将数据转为原来的真实文本，因此我们就需要文本单词和数字的对应表，如下： 

# 加载词向量表
word_to_id = tf.keras.datasets.imdb.get_word_index()
word_to_id = {k: (v + index_offset) for k, v in word_to_id.items()}
word_to_id["<PAD>"] = 0
word_to_id["<START>"] = 1
word_to_id["<UNK>"] = 2
id_to_word = {value: key for key, value in word_to_id.items()}
id_to_word

{34704: 'fawn',
 52009: 'tsukino',
 52010: 'nunnery',
 16819: 'sonja',
 63954: 'vani',
 1411: 'woods',
 16118: 'spiders',
 2348: 'hanging',
 2292: 'woody',
 52011: 'trawling',
 52012: "hold's",
 11310: 'comically',
 40833: 'localized',
 30571: 'disobeying',
 52013: "'royale",
 40834: "harpo's",
 52014: 'canet',
 19316: 'aileen',
 52015: 'acurately',
 52016: "diplomat's",
 25245: 'rickman',
 6749: 'arranged',
 52017: 'rumbustious',
 52018: 'familiarness',
 52019: "spider'",
 68807: 'hahahah',
 52020: "wood'",
 40836: 'transvestism',
 34705: "hangin'",
 2341: 'bringing',
 40837: 'seamier',
 34706: 'wooded',
 52021: 'bravora',
 16820: 'grueling',
 1639: 'wooden',
 16821: 'wednesday',
 52022: "'prix",
 34707: 'altagracia',
 52023: 'circuitry',
 11588: 'crotch',
 57769: 'busybody',
 52024: "tart'n'tangy",
 14132: 'burgade',
 52026: 'thrace',
 11041: "tom's",
 52028: 'snuggles',
 29117: 'francesco',
 52030: 'complainers',
 52128: 'templarios',
 40838: '272',
 52031: '273',
 52133: 'zaniacs',
 34709: '275',
 27634: 'consenting',
 40839: 'snuggled',
 15495: 'inanimate',
 52033: 'uality',
 11929: 'bronte',
 4013: 'errors',
 3233: 'dialogs',
 52034: "yomada's",
 34710: "madman's",
 30588: 'dialoge',
 52036: 'usenet',
 40840: 'videodrome',
 26341: "kid'",
 52037: 'pawed',
 30572: "'girlfriend'",
 52038: "'pleasure",
 52039: "'reloaded'",
 40842: "kazakos'",
 52040: 'rocque',
 52041: 'mailings',
 11930: 'brainwashed',
 16822: 'mcanally',
 52042: "tom''",
 25246: 'kurupt',
 21908: 'affiliated',
 52043: 'babaganoosh',
 40843: "noe's",
 40844: 'quart',
 362: 'kids',
 5037: 'uplifting',
 7096: 'controversy',
 21909: 'kida',
 23382: 'kidd',
 52044: "error'",
 52045: 'neurologist',
 18513: 'spotty',
 30573: 'cobblers',
 9881: 'projection',
 40845: 'fastforwarding',
 52046: 'sters',
 52047: "eggar's",
 52048: 'etherything',
 40846: 'gateshead',
 34711: 'airball',
 25247: 'unsinkable',
 7183: 'stern',
 52049: "cervi's",
 40847: 'dnd',
 11589: 'dna',
 20601: 'insecurity',
 52050: "'reboot'",
 11040: 'trelkovsky',
 52051: 'jaekel',
 52052: 'sidebars',
 52053: "sforza's",
 17636: 'distortions',
 52054: 'mutinies',
 30605: 'sermons',
 40849: '7ft',
 52055: 'boobage',
 52056: "o'bannon's",
 23383: 'populations',
 52057: 'chulak',
 27636: 'mesmerize',
 52058: 'quinnell',
 10310: 'yahoo',
 52060: 'meteorologist',
 42580: 'beswick',
 15496: 'boorman',
 40850: 'voicework',
 52061: "ster'",
 22925: 'blustering',
 52062: 'hj',
 27637: 'intake',
 5624: 'morally',
 40852: 'jumbling',
 52063: 'bowersock',
 52064: "'porky's'",
 16824: 'gershon',
 40853: 'ludicrosity',
 52065: 'coprophilia',
 40854: 'expressively',
 19503: "india's",
 34713: "post's",
 52066: 'wana',
 5286: 'wang',
 30574: 'wand',
 25248: 'wane',
 52324: 'edgeways',
 34714: 'titanium',
 40855: 'pinta',
 181: 'want',
 30575: 'pinto',
 52068: 'whoopdedoodles',
 21911: 'tchaikovsky',
 2106: 'travel',
 52069: "'victory'",
 11931: 'copious',
 22436: 'gouge',
 52070: "chapters'",
 6705: 'barbra',
 30576: 'uselessness',
 52071: "wan'",
 27638: 'assimilated',
 16119: 'petiot',
 52072: 'most\x85and',
 3933: 'dinosaurs',
 355: 'wrong',
 52073: 'seda',
 52074: 'stollen',
 34715: 'sentencing',
 40856: 'ouroboros',
 40857: 'assimilates',
 40858: 'colorfully',
 27639: 'glenne',
 52075: 'dongen',
 4763: 'subplots',
 52076: 'kiloton',
 23384: 'chandon',
 34716: "effect'",
 27640: 'snugly',
 40859: 'kuei',
 9095: 'welcomed',
 30074: 'dishonor',
 52078: 'concurrence',
 23385: 'stoicism',
 14899: "guys'",
 52080: "beroemd'",
 6706: 'butcher',
 40860: "melfi's",
 30626: 'aargh',
 20602: 'playhouse',
 11311: 'wickedly',
 1183: 'fit',
 52081: 'labratory',
 40862: 'lifeline',
 1930: 'screaming',
 4290: 'fix',
 52082: 'cineliterate',
 52083: 'fic',
 52084: 'fia',
 34717: 'fig',
 52085: 'fmvs',
 52086: 'fie',
 52087: 'reentered',
 30577: 'fin',
 52088: 'doctresses',
 52089: 'fil',
 12609: 'zucker',
 31934: 'ached',
 52091: 'counsil',
 52092: 'paterfamilias',
 13888: 'songwriter',
 34718: 'shivam',
 9657: 'hurting',
 302: 'effects',
 52093: 'slauther',
 52094: "'flame'",
 52095: 'sommerset',
 52096: 'interwhined',
 27641: 'whacking',
 52097: 'bartok',
 8778: 'barton',
 21912: 'frewer',
 52098: "fi'",
 6195: 'ingrid',
 30578: 'stribor',
 52099: 'approporiately',
 52100: 'wobblyhand',
 52101: 'tantalisingly',
 52102: 'ankylosaurus',
 17637: 'parasites',
 52103: 'childen',
 52104: "jenkins'",
 52105: 'metafiction',
 17638: 'golem',
 40863: 'indiscretion',
 23386: "reeves'",
 57784: "inamorata's",
 52107: 'brittannica',
 7919: 'adapt',
 30579: "russo's",
 48249: 'guitarists',
 10556: 'abbott',
 40864: 'abbots',
 17652: 'lanisha',
 40866: 'magickal',
 52108: 'mattter',
 52109: "'willy",
 34719: 'pumpkins',
 52110: 'stuntpeople',
 30580: 'estimate',
 40867: 'ugghhh',
 11312: 'gameplay',
 52111: "wern't",
 40868: "n'sync",
 16120: 'sickeningly',
 40869: 'chiara',
 4014: 'disturbed',
 40870: 'portmanteau',
 52112: 'ineffectively',
 82146: "duchonvey's",
 37522: "nasty'",
 1288: 'purpose',
 52115: 'lazers',
 28108: 'lightened',
 52116: 'kaliganj',
 52117: 'popularism',
 18514: "damme's",
 30581: 'stylistics',
 52118: 'mindgaming',
 46452: 'spoilerish',
 52120: "'corny'",
 34721: 'boerner',
 6795: 'olds',
 52121: 'bakelite',
 27642: 'renovated',
 27643: 'forrester',
 52122: "lumiere's",
 52027: 'gaskets',
 887: 'needed',
 34722: 'smight',
 1300: 'master',
 25908: "edie's",
 40871: 'seeber',
 52123: 'hiya',
 52124: 'fuzziness',
 14900: 'genesis',
 12610: 'rewards',
 30582: 'enthrall',
 40872: "'about",
 52125: "recollection's",
 11042: 'mutilated',
 52126: 'fatherlands',
 52127: "fischer's",
 5402: 'positively',
 34708: '270',
 34723: 'ahmed',
 9839: 'zatoichi',
 13889: 'bannister',
 52130: 'anniversaries',
 30583: "helm's",
 52131: "'work'",
 34724: 'exclaimed',
 52132: "'unfunny'",
 52032: '274',
 547: 'feeling',
 52134: "wanda's",
 33269: 'dolan',
 52136: '278',
 52137: 'peacoat',
 40873: 'brawny',
 40874: 'mishra',
 40875: 'worlders',
 52138: 'protags',
 52139: 'skullcap',
 57599: 'dastagir',
 5625: 'affairs',
 7802: 'wholesome',
 52140: 'hymen',
 25249: 'paramedics',
 52141: 'unpersons',
 52142: 'heavyarms',
 52143: 'affaire',
 52144: 'coulisses',
 40876: 'hymer',
 52145: 'kremlin',
 30584: 'shipments',
 52146: 'pixilated',
 30585: "'00s",
 18515: 'diminishing',
 1360: 'cinematic',
 14901: 'resonates',
 40877: 'simplify',
 40878: "nature'",
 40879: 'temptresses',
 16825: 'reverence',
 19505: 'resonated',
 34725: 'dailey',
 52147: '2\x85',
 27644: 'treize',
 52148: 'majo',
 21913: 'kiya',
 52149: 'woolnough',
 39800: 'thanatos',
 35734: 'sandoval',
 40882: 'dorama',
 52150: "o'shaughnessy",
 4991: 'tech',
 32021: 'fugitives',
 30586: 'teck',
 76128: "'e'",
 40884: 'doesn’t',
 52152: 'purged',
 660: 'saying',
 41098: "martians'",
 23421: 'norliss',
 27645: 'dickey',
 52155: 'dicker',
 52156: "'sependipity",
 8425: 'padded',
 57795: 'ordell',
 40885: "sturges'",
 52157: 'independentcritics',
 5748: 'tempted',
 34727: "atkinson's",
 25250: 'hounded',
 52158: 'apace',
 15497: 'clicked',
 30587: "'humor'",
 17180: "martino's",
 52159: "'supporting",
 52035: 'warmongering',
 34728: "zemeckis's",
 21914: 'lube',
 52160: 'shocky',
 7479: 'plate',
 40886: 'plata',
 40887: 'sturgess',
 40888: "nerds'",
 20603: 'plato',
 34729: 'plath',
 40889: 'platt',
 52162: 'mcnab',
 27646: 'clumsiness',
 3902: 'altogether',
 42587: 'massacring',
 52163: 'bicenntinial',
 40890: 'skaal',
 14363: 'droning',
 8779: 'lds',
 21915: 'jaguar',
 34730: "cale's",
 1780: 'nicely',
 4591: 'mummy',
 18516: "lot's",
 10089: 'patch',
 50205: 'kerkhof',
 52164: "leader's",
 27647: "'movie",
 52165: 'uncomfirmed',
 40891: 'heirloom',
 47363: 'wrangle',
 52166: 'emotion\x85',
 52167: "'stargate'",
 40892: 'pinoy',
 40893: 'conchatta',
 41131: 'broeke',
 40894: 'advisedly',
 17639: "barker's",
 52169: 'descours',
 775: 'lots',
 9262: 'lotr',
 9882: 'irs',
 52170: 'lott',
 40895: 'xvi',
 34731: 'irk',
 52171: 'irl',
 6890: 'ira',
 21916: 'belzer',
 52172: 'irc',
 27648: 'ire',
 40896: 'requisites',
 7696: 'discipline',
 52964: 'lyoko',
 11313: 'extend',
 876: 'nature',
 52173: "'dickie'",
 40897: 'optimist',
 30589: 'lapping',
 3903: 'superficial',
 52174: 'vestment',
 2826: 'extent',
 52175: 'tendons',
 52176: "heller's",
 52177: 'quagmires',
 52178: 'miyako',
 20604: 'moocow',
 52179: "coles'",
 40898: 'lookit',
 52180: 'ravenously',
 40899: 'levitating',
 52181: 'perfunctorily',
 30590: 'lookin',
 40901: "lot'",
 52182: 'lookie',
 34873: 'fearlessly',
 52184: 'libyan',
 40902: 'fondles',
 35717: 'gopher',
 40904: 'wearying',
 52185: "nz's",
 27649: 'minuses',
 52186: 'puposelessly',
 52187: 'shandling',
 31271: 'decapitates',
 11932: 'humming',
 40905: "'nother",
 21917: 'smackdown',
 30591: 'underdone',
 40906: 'frf',
 52188: 'triviality',
 25251: 'fro',
 8780: 'bothers',
 52189: "'kensington",
 76: 'much',
 34733: 'muco',
 22618: 'wiseguy',
 27651: "richie's",
 40907: 'tonino',
 52190: 'unleavened',
 11590: 'fry',
 40908: "'tv'",
 40909: 'toning',
 14364: 'obese',
 30592: 'sensationalized',
 40910: 'spiv',
 6262: 'spit',
 7367: 'arkin',
 21918: 'charleton',
 16826: 'jeon',
 21919: 'boardroom',
 4992: 'doubts',
 3087: 'spin',
 53086: 'hepo',
 27652: 'wildcat',
 10587: 'venoms',
 52194: 'misconstrues',
 18517: 'mesmerising',
 40911: 'misconstrued',
 52195: 'rescinds',
 52196: 'prostrate',
 40912: 'majid',
 16482: 'climbed',
 34734: 'canoeing',
 52198: 'majin',
 57807: 'animie',
 40913: 'sylke',
 14902: 'conditioned',
 40914: 'waddell',
 52199: '3\x85',
 41191: 'hyperdrive',
 34735: 'conditioner',
 53156: 'bricklayer',
 2579: 'hong',
 52201: 'memoriam',
 30595: 'inventively',
 25252: "levant's",
 20641: 'portobello',
 52203: 'remand',
 19507: 'mummified',
 27653: 'honk',
 19508: 'spews',
 40915: 'visitations',
 52204: 'mummifies',
 25253: 'cavanaugh',
 23388: 'zeon',
 40916: "jungle's",
 34736: 'viertel',
 27654: 'frenchmen',
 52205: 'torpedoes',
 52206: 'schlessinger',
 34737: 'torpedoed',
 69879: 'blister',
 52207: 'cinefest',
 34738: 'furlough',
 52208: 'mainsequence',
 40917: 'mentors',
 9097: 'academic',
 20605: 'stillness',
 40918: 'academia',
 52209: 'lonelier',
 52210: 'nibby',
 52211: "losers'",
 40919: 'cineastes',
 4452: 'corporate',
 40920: 'massaging',
 30596: 'bellow',
 19509: 'absurdities',
 53244: 'expetations',
 40921: 'nyfiken',
 75641: 'mehras',
 52212: 'lasse',
 52213: 'visability',
 33949: 'militarily',
 52214: "elder'",
 19026: 'gainsbourg',
 20606: 'hah',
 13423: 'hai',
 34739: 'haj',
 25254: 'hak',
 4314: 'hal',
 4895: 'ham',
 53262: 'duffer',
 52216: 'haa',
 69: 'had',
 11933: 'advancement',
 16828: 'hag',
 25255: "hand'",
 13424: 'hay',
 20607: 'mcnamara',
 52217: "mozart's",
 30734: 'duffel',
 30597: 'haq',
 13890: 'har',
 47: 'has',
 2404: 'hat',
 40922: 'hav',
 30598: 'haw',
 52218: 'figtings',
 15498: 'elders',
 52219: 'underpanted',
 52220: 'pninson',
 27655: 'unequivocally',
 23676: "barbara's",
 52222: "bello'",
 13000: 'indicative',
 40923: 'yawnfest',
 52223: 'hexploitation',
 52224: "loder's",
 27656: 'sleuthing',
 32625: "justin's",
 52225: "'ball",
 52226: "'summer",
 34938: "'demons'",
 52228: "mormon's",
 34740: "laughton's",
 52229: 'debell',
 39727: 'shipyard',
 30600: 'unabashedly',
 40404: 'disks',
 2293: 'crowd',
 10090: 'crowe',
 56437: "vancouver's",
 34741: 'mosques',
 6630: 'crown',
 52230: 'culpas',
 27657: 'crows',
 53347: 'surrell',
 52232: 'flowless',
 52233: 'sheirk',
 40926: "'three",
 52234: "peterson'",
 52235: 'ooverall',
 40927: 'perchance',
 1324: 'bottom',
 53366: 'chabert',
 52236: 'sneha',
 13891: 'inhuman',
 52237: 'ichii',
 52238: 'ursla',
 30601: 'completly',
 40928: 'moviedom',
 52239: 'raddick',
 51998: 'brundage',
 40929: 'brigades',
 1184: 'starring',
 52240: "'goal'",
 52241: 'caskets',
 52242: 'willcock',
 52243: "threesome's",
 52244: "mosque'",
 52245: "cover's",
 17640: 'spaceships',
 40930: 'anomalous',
 27658: 'ptsd',
 52246: 'shirdan',
 21965: 'obscenity',
 30602: 'lemmings',
 30603: 'duccio',
 52247: "levene's",
 52248: "'gorby'",
 25258: "teenager's",
 5343: 'marshall',
 9098: 'honeymoon',
 3234: 'shoots',
 12261: 'despised',
 52249: 'okabasho',
 8292: 'fabric',
 18518: 'cannavale',
 3540: 'raped',
 52250: "tutt's",
 17641: 'grasping',
 18519: 'despises',
 40931: "thief's",
 8929: 'rapes',
 52251: 'raper',
 27659: "eyre'",
 52252: 'walchek',
 23389: "elmo's",
 40932: 'perfumes',
 21921: 'spurting',
 52253: "exposition'\x85",
 52254: 'denoting',
 34743: 'thesaurus',
 40933: "shoot'",
 49762: 'bonejack',
 52256: 'simpsonian',
 30604: 'hebetude',
 34744: "hallow's",
 52257: 'desperation\x85',
 34745: 'incinerator',
 10311: 'congratulations',
 52258: 'humbled',
 5927: "else's",
 40848: 'trelkovski',
 52259: "rape'",
 59389: "'chapters'",
 52260: '1600s',
 7256: 'martian',
 25259: 'nicest',
 52262: 'eyred',
 9460: 'passenger',
 6044: 'disgrace',
 52263: 'moderne',
 5123: 'barrymore',
 52264: 'yankovich',
 40934: 'moderns',
 52265: 'studliest',
 52266: 'bedsheet',
 14903: 'decapitation',
 52267: 'slurring',
 52268: "'nunsploitation'",
 34746: "'character'",
 9883: 'cambodia',
 52269: 'rebelious',
 27660: 'pasadena',
 40935: 'crowne',
 52270: "'bedchamber",
 52271: 'conjectural',
 52272: 'appologize',
 52273: 'halfassing',
 57819: 'paycheque',
 20609: 'palms',
 52274: "'islands",
 40936: 'hawked',
 21922: 'palme',
 40937: 'conservatively',
 64010: 'larp',
 5561: 'palma',
 21923: 'smelling',
 13001: 'aragorn',
 52275: 'hawker',
 52276: 'hawkes',
 3978: 'explosions',
 8062: 'loren',
 52277: "pyle's",
 6707: 'shootout',
 18520: "mike's",
 52278: "driscoll's",
 40938: 'cogsworth',
 52279: "britian's",
 34747: 'childs',
 52280: "portrait's",
 3629: 'chain',
 2500: 'whoever',
 52281: 'puttered',
 52282: 'childe',
 52283: 'maywether',
 3039: 'chair',
 52284: "rance's",
 34748: 'machu',
 4520: 'ballet',
 34749: 'grapples',
 76155: 'summerize',
 30606: 'freelance',
 52286: "andrea's",
 52287: '\x91very',
 45882: 'coolidge',
 18521: 'mache',
 52288: 'balled',
 40940: 'grappled',
 18522: 'macha',
 21924: 'underlining',
 5626: 'macho',
 19510: 'oversight',
 25260: 'machi',
 11314: 'verbally',
 21925: 'tenacious',
 40941: 'windshields',
 18560: 'paychecks',
 3399: 'jerk',
 11934: "good'",
 34751: 'prancer',
 21926: 'prances',
 52289: 'olympus',
 21927: 'lark',
 10788: 'embark',
 7368: 'gloomy',
 52290: 'jehaan',
 52291: 'turaqui',
 20610: "child'",
 2897: 'locked',
 52292: 'pranced',
 2591: 'exact',
 52293: 'unattuned',
 786: 'minute',
 16121: 'skewed',
 40943: 'hodgins',
 34752: 'skewer',
 52294: 'think\x85',
 38768: 'rosenstein',
 52295: 'helmit',
 34753: 'wrestlemanias',
 16829: 'hindered',
 30607: "martha's",
 52296: 'cheree',
 52297: "pluckin'",
 40944: 'ogles',
 11935: 'heavyweight',
 82193: 'aada',
 11315: 'chopping',
 61537: 'strongboy',
 41345: 'hegemonic',
 40945: 'adorns',
 41349: 'xxth',
 34754: 'nobuhiro',
 52301: 'capitães',
 52302: 'kavogianni',
 13425: 'antwerp',
 6541: 'celebrated',
 52303: 'roarke',
 40946: 'baggins',
 31273: 'cheeseburgers',
 52304: 'matras',
 52305: "nineties'",
 52306: "'craig'",
 13002: 'celebrates',
 3386: 'unintentionally',
 14365: 'drafted',
 52307: 'climby',
 52308: '303',
 18523: 'oldies',
 9099: 'climbs',
 9658: 'honour',
 34755: 'plucking',
 30077: '305',
 5517: 'address',
 40947: 'menjou',
 42595: "'freak'",
 19511: 'dwindling',
 9461: 'benson',
 52310: 'white’s',
 40948: 'shamelessness',
 21928: 'impacted',
 52311: 'upatz',
 3843: 'cusack',
 37570: "flavia's",
 52312: 'effette',
 34756: 'influx',
 52313: 'boooooooo',
 52314: 'dimitrova',
 13426: 'houseman',
 25262: 'bigas',
 52315: 'boylen',
 52316: 'phillipenes',
 40949: 'fakery',
 27661: "grandpa's",
 27662: 'darnell',
 19512: 'undergone',
 52318: 'handbags',
 21929: 'perished',
 37781: 'pooped',
 27663: 'vigour',
 3630: 'opposed',
 52319: 'etude',
 11802: "caine's",
 52320: 'doozers',
 34757: 'photojournals',
 52321: 'perishes',
 34758: 'constrains',
 40951: 'migenes',
 30608: 'consoled',
 16830: 'alastair',
 52322: 'wvs',
 52323: 'ooooooh',
 34759: 'approving',
 40952: 'consoles',
 52067: 'disparagement',
 52325: 'futureistic',
 52326: 'rebounding',
 52327: "'date",
 52328: 'gregoire',
 21930: 'rutherford',
 34760: 'americanised',
 82199: 'novikov',
 1045: 'following',
 34761: 'munroe',
 52329: "morita'",
 52330: 'christenssen',
 23109: 'oatmeal',
 25263: 'fossey',
 40953: 'livered',
 13003: 'listens',
 76167: "'marci",
 52333: "otis's",
 23390: 'thanking',
 16022: 'maude',
 34762: 'extensions',
 52335: 'ameteurish',
 52336: "commender's",
 27664: 'agricultural',
 4521: 'convincingly',
 17642: 'fueled',
 54017: 'mahattan',
 40955: "paris's",
 52339: 'vulkan',
 52340: 'stapes',
 52341: 'odysessy',
 12262: 'harmon',
 4255: 'surfing',
 23497: 'halloran',
 49583: 'unbelieveably',
 52342: "'offed'",
 30610: 'quadrant',
 19513: 'inhabiting',
 34763: 'nebbish',
 40956: 'forebears',
 34764: 'skirmish',
 52343: 'ocassionally',
 52344: "'resist",
 21931: 'impactful',
 52345: 'spicier',
 40957: 'touristy',
 52346: "'football'",
 40958: 'webpage',
 52348: 'exurbia',
 52349: 'jucier',
 14904: 'professors',
 34765: 'structuring',
 30611: 'jig',
 40959: 'overlord',
 25264: 'disconnect',
 82204: 'sniffle',
 40960: 'slimeball',
 40961: 'jia',
 16831: 'milked',
 40962: 'banjoes',
 1240: 'jim',
 52351: 'workforces',
 52352: 'jip',
 52353: 'rotweiller',
 34766: 'mundaneness',
 52354: "'ninja'",
 11043: "dead'",
 40963: "cipriani's",
 20611: 'modestly',
 52355: "professor'",
 40964: 'shacked',
 34767: 'bashful',
 23391: 'sorter',
 16123: 'overpowering',
 18524: 'workmanlike',
 27665: 'henpecked',
 18525: 'sorted',
 52357: "jōb's",
 52358: "'always",
 34768: "'baptists",
 52359: 'dreamcatchers',
 52360: "'silence'",
 21932: 'hickory',
 52361: 'fun\x97yet',
 52362: 'breakumentary',
 15499: 'didn',
 52363: 'didi',
 52364: 'pealing',
 40965: 'dispite',
 25265: "italy's",
 21933: 'instability',
 6542: 'quarter',
 12611: 'quartet',
 52365: 'padmé',
 52366: "'bleedmedry",
 52367: 'pahalniuk',
 52368: 'honduras',
 10789: 'bursting',
 41468: "pablo's",
 52370: 'irremediably',
 40966: 'presages',
 57835: 'bowlegged',
 65186: 'dalip',
 6263: 'entering',
 76175: 'newsradio',
 54153: 'presaged',
 27666: "giallo's",
 40967: 'bouyant',
 52371: 'amerterish',
 18526: 'rajni',
 30613: 'leeves',
 34770: 'macauley',
 615: 'seriously',
 52372: 'sugercoma',
 52373: 'grimstead',
 52374: "'fairy'",
 30614: 'zenda',
 52375: "'twins'",
 17643: 'realisation',
 27667: 'highsmith',
 7820: 'raunchy',
 40968: 'incentives',
 52377: 'flatson',
 35100: 'snooker',
 16832: 'crazies',
 14905: 'crazier',
 7097: 'grandma',
 52378: 'napunsaktha',
 30615: 'workmanship',
 52379: 'reisner',
 61309: "sanford's",
 52380: '\x91doña',
 6111: 'modest',
 19156: "everything's",
 40969: 'hamer',
 52382: "couldn't'",
 13004: 'quibble',
 52383: 'socking',
 21934: 'tingler',
 52384: 'gutman',
 40970: 'lachlan',
 52385: 'tableaus',
 52386: 'headbanger',
 2850: 'spoken',
 34771: 'cerebrally',
 23493: "'road",
 21935: 'tableaux',
 40971: "proust's",
 40972: 'periodical',
 52388: "shoveller's",
 25266: 'tamara',
 17644: 'affords',
 3252: 'concert',
 87958: "yara's",
 52389: 'someome',
 8427: 'lingering',
 41514: "abraham's",
 34772: 'beesley',
 34773: 'cherbourg',
 28627: 'kagan',
 9100: 'snatch',
 9263: "miyazaki's",
 25267: 'absorbs',
 40973: "koltai's",
 64030: 'tingled',
 19514: 'crossroads',
 16124: 'rehab',
 52392: 'falworth',
 52393: 'sequals',
 ...}

如上所示，这个对应表就像字典一样，将每个单词和一个 id 相对应，进而将所有的字符串转为了数字。

接下来，我们就需要利用对照该表将训练数据和测试数据转换为原文本： 

x_train = list(map(lambda sentence: ' '.join(id_to_word[i] for i in sentence), x_train))
x_test = list(map(lambda sentence: ' '.join(id_to_word[i] for i in sentence), x_test))
x_train = np.array(x_train, dtype=np.str_)
x_test = np.array(x_test, dtype=np.str_)
print(x_train)
print(x_train.shape)  # (25000,)
print(y_train.shape)  # (25000, 1)
print(x_train[0][:50])  # <START> this film was just brilliant casting <UNK>

["<START> this film was just brilliant casting location scenery story direction everyone's really suited the part they played and you could just imagine being there robert <UNK> is an amazing actor and now the same being director <UNK> father came from the same scottish island as myself so i loved the fact there was a real connection with this film the witty remarks throughout the film were great it was just brilliant so much that i bought the film as soon as it was released for retail and would recommend it to everyone to watch and the fly fishing was amazing really cried at the end it was so sad and you know what they say if you cry at a film it must have been good and this definitely was also congratulations to the two little boy's that played the <UNK> of norman and paul they were just brilliant children are often left out of the praising list i think because the stars that play them all grown up are such a big profile for the whole film but these children are amazing and should be praised for what they have done don't you think the whole story was so lovely because it was true and was someone's life after all that was shared with us all"
 "<START> big hair big boobs bad music and a giant safety pin these are the words to best describe this terrible movie i love cheesy horror movies and i've seen hundreds but this had got to be on of the worst ever made the plot is paper thin and ridiculous the acting is an abomination the script is completely laughable the best is the end showdown with the cop and how he worked out who the killer is it's just so damn terribly written the clothes are sickening and funny in equal measures the hair is big lots of boobs bounce men wear those cut tee shirts that show off their <UNK> sickening that men actually wore them and the music is just <UNK> trash that plays over and over again in almost every scene there is trashy music boobs and <UNK> taking away bodies and the gym still doesn't close for <UNK> all joking aside this is a truly bad film whose only charm is to look back on the disaster that was the 80's and have a good old laugh at how bad everything was back then"
 "<START> this has to be one of the worst films of the 1990s when my friends i were watching this film being the target audience it was aimed at we just sat watched the first half an hour with our jaws touching the floor at how bad it really was the rest of the time everyone else in the theatre just started talking to each other leaving or generally crying into their popcorn that they actually paid money they had <UNK> working to watch this feeble excuse for a film it must have looked like a great idea on paper but on film it looks like no one in the film has a clue what is going on crap acting crap costumes i can't get across how <UNK> this is to watch save yourself an hour a bit of your life"
 ...
 "<START> in a far away galaxy is a planet called <UNK> it's native people worship cats but the dog people wage war upon these feline loving people and they have no choice but to go to earth and grind people up for food this is one of the stupidest f k <UNK> ideas for a movie i've seen leave it to ted mikels to make a movie more incompetent than the already low standard he set in previous films it's like he enjoying playing in a celluloid game of limbo how low can he go the only losers in the scenario are us the viewer mr mikels and his silly little <UNK> mustache actually has people who still buy this crap br br my grade f br br dvd extras commentary by ted mikels the story behind the making of 9 and a half minutes 17 minutes 15 seconds of behind the scenes footage ted mikels filmography and trailers for the worm eaters girl in gold boots the doll squad ten violent women featuring nudity blood orgy of the she devils the corpse <UNK>"
 "<START> six degrees had me hooked i looked forward to it coming on and was totally disappointed when men in trees replaced it's time spot i thought it was just on <UNK> and would be back early in 2007 what happened all my friends were really surprised it ended we could relate to the characters who had real problems we talked about each episode and had our favorite characters there wasn't anybody on the show i didn't like and felt the acting was superb i <UNK> like seeing programs being taped in cities where you can identify the local areas i for one would like to protest the <UNK> of this show and ask you to bring it back and give it another chance give it a good time slot don't keep moving it from this day to that day and <UNK> it so people will know it is on"
 "<START> as a big fan of the original film it's hard to watch this show the garish set decor and harshly lighted sets rob any style from this remake the mood is never there instead it has the look and feel of so many television movies of the seventies crenna is not a bad choice as walter neff but his snappy wardrobe and <UNK> apartment don't fit the mood of the original or make him an interesting character he does his best to make it work but samantha <UNK> is a really bad choice the english accent and california looks can't hold a candle to barbara <UNK> velvet voice and sex appeal lee j cobb tries mightily to fashion barton keyes but even his performance is just gruff without style br br it feels like the tv movie it was and again reminds me of what a remarkable film the original still is"]
(25000,)
(25000, 1)
<START> this film was just brilliant casting locat

如上图所示，我们需要对上面的真实文本进行评估和预测。和图像分类的使用的方法 ImageClassifier 类似，TextClassifier 的使用方法如下（下面代码将运行 8~10 min,请耐心等待）： 

# 这里还是只尝试一个模型，1 个 epoch。
# 为了节约资源，这里我们将 max_trials 设置为 1
# 在本地运行时，你可以将 max_trials 设置为 10 左右，进而找到较优的模型
clf = ak.TextClassifier(max_trials=1)
clf.fit(x_train, y_train, epochs=1)
print("训练完成")

Trial 1 Complete [00h 00m 26s]
val_loss: 0.29259392619132996

Best val_loss So Far: 0.29259392619132996
Total elapsed time: 00h 00m 26s

782/782 [==============================] - 32s 40ms/step - loss: 0.4363 - accuracy: 0.7761
INFO:tensorflow:Assets written to: ./text_classifier/best_model/assets

训练完成

Note

上面的结果展示，每迭代超过 一半的 batch ，就会重新换行。因此，如果你发现进度条静止不动时，可以下拉展示框，结果可能在下面。

同理，让我们放入测试数据，观察寻找出来的模型的准确率： 

clf.evaluate(x_test, y_test)

782/782 [==============================] - 10s 13ms/step - loss: 0.2807 - accuracy: 0.8835

[0.28066298365592957, 0.8834800124168396]

78.6. 结构化的数据分类#

除了上述的计算机视觉和自然语言处理之外， Auto-Keras 还可以对结构化数据进行分类。用于对结构化数据进行分类的类为 StructuredDataClassifier，该类和上面的类的使用方法大致相同。但是该类的 fit(x,y) 函数的主要参数 x,y 发生了一些改变。

  • x:表示模型的输入数据。该数据的类型除了 numpy.ndarray、pandas.DataFrame 和 tensorflow.Dataset 等之外，还可以是一个字符串。该字符串表示训练数据所在的 csv 文件的路径。换句话说，我们只需要将这个文件放入 fit 函数中就行了，连读数据的步骤都可以省掉。

  • y：表示模型的输出数据，即 target 。该数据的类型除了 numpy.ndarray、pandas.DataFrame 和 tensorflow.Dataset 等之外，也可以为一个字符串。该字符串表示通过 x 传入的 csv 文件中的某列的列名。即把该列作为 target，其他列作为 input。

接下来，让我们利用该工具类完成泰坦尼克号的分类。

首先，让我们先来加载一下这个数据集合：

wget -nc "https://cdn.aibydoing.com/aibydoing/files/titanic_eval.csv"
wget -nc "https://cdn.aibydoing.com/aibydoing/files/titanic_train.csv"

--2023-11-14 11:01:33--  https://cdn.aibydoing.com/aibydoing/files/titanic_eval.csv
正在解析主机 cdn.aibydoing.com (cdn.aibydoing.com)... 198.18.7.59
正在连接 cdn.aibydoing.com (cdn.aibydoing.com)|198.18.7.59|:443... 已连接。
已发出 HTTP 请求，正在等待回应... 200 OK
长度：13049 (13K) [text/csv]
正在保存至: “titanic_eval.csv”

titanic_eval.csv    100%[===================>]  12.74K  --.-KB/s  用时 0.08s     

2023-11-14 11:01:34 (152 KB/s) - 已保存 “titanic_eval.csv” [13049/13049])

--2023-11-14 11:01:34--  https://cdn.aibydoing.com/aibydoing/files/titanic_train.csv
正在解析主机 cdn.aibydoing.com (cdn.aibydoing.com)... 198.18.7.59
正在连接 cdn.aibydoing.com (cdn.aibydoing.com)|198.18.7.59|:443... 已连接。
已发出 HTTP 请求，正在等待回应... 200 OK
长度：30874 (30K) [text/csv]
正在保存至: “titanic_train.csv”

titanic_train.csv   100%[===================>]  30.15K  --.-KB/s  用时 0.1s      

2023-11-14 11:01:36 (258 KB/s) - 已保存 “titanic_train.csv” [30874/30874])

import pandas as pd
df = pd.read_csv("titanic_train.csv")
df.head()
survived sex age n_siblings_spouses parch fare class deck embark_town alone
0 0 male 22.0 1 0 7.2500 Third unknown Southampton n
1 1 female 38.0 1 0 71.2833 First C Cherbourg n
2 1 female 26.0 0 0 7.9250 Third unknown Southampton y
3 1 female 35.0 1 0 53.1000 First C Southampton n
4 0 male 28.0 0 0 8.4583 Third unknown Queenstown y

从上面的结果我们可以看出，表中存储了泰坦尼克号的乘客的一些基本信息，且其中的 survived 表示该乘客是否在事故中幸存了下来。本实验的目的就是通过 Auto-Keras 进行训练，进而得到一个能够预测乘客是否获救的分类模型（其中 1 代表获救，0 代表没有获救）。

我们无需对上面数据进行任何处理，直接将数据放入分类器中进行分类即可，如下(下面代码可能会运行 8~10 min，请耐心等待)： 

import autokeras as ak
# 定义分类器，这里我们尝试 3 个模型，并最终返回三个中的最佳模型
clf = ak.StructuredDataClassifier(max_trials=3)
# 只需传入文件路径和文件列名即可
# verbose=2 表示每一个 epoch 显示一次日志信息
clf.fit(x='titanic_train.csv', y='survived', verbose=2, epochs=500)
print("训练完成")

Trial 3 Complete [00h 00m 04s]
val_accuracy: 0.8695651888847351

Best val_accuracy So Far: 0.8782608509063721
Total elapsed time: 00h 00m 24s

Epoch 1/500
20/20 - 0s - loss: 0.6435 - accuracy: 0.6970 - 236ms/epoch - 12ms/step
Epoch 2/500
20/20 - 0s - loss: 0.5628 - accuracy: 0.7974 - 13ms/epoch - 632us/step
Epoch 3/500
20/20 - 0s - loss: 0.5063 - accuracy: 0.8086 - 13ms/epoch - 634us/step
Epoch 4/500
20/20 - 0s - loss: 0.4680 - accuracy: 0.8166 - 15ms/epoch - 744us/step
Epoch 5/500
20/20 - 0s - loss: 0.4453 - accuracy: 0.8198 - 12ms/epoch - 605us/step
Epoch 6/500
20/20 - 0s - loss: 0.4316 - accuracy: 0.8214 - 12ms/epoch - 593us/step
Epoch 7/500
20/20 - 0s - loss: 0.4226 - accuracy: 0.8246 - 16ms/epoch - 777us/step
Epoch 8/500
20/20 - 0s - loss: 0.4156 - accuracy: 0.8230 - 18ms/epoch - 904us/step
Epoch 9/500
20/20 - 0s - loss: 0.4097 - accuracy: 0.8278 - 12ms/epoch - 613us/step
Epoch 10/500
20/20 - 0s - loss: 0.4051 - accuracy: 0.8309 - 12ms/epoch - 624us/step
Epoch 11/500
20/20 - 0s - loss: 0.4010 - accuracy: 0.8309 - 13ms/epoch - 640us/step
Epoch 12/500
20/20 - 0s - loss: 0.3973 - accuracy: 0.8325 - 18ms/epoch - 891us/step
Epoch 13/500
20/20 - 0s - loss: 0.3941 - accuracy: 0.8341 - 13ms/epoch - 638us/step
Epoch 14/500
20/20 - 0s - loss: 0.3912 - accuracy: 0.8341 - 13ms/epoch - 643us/step
Epoch 15/500
20/20 - 0s - loss: 0.3885 - accuracy: 0.8389 - 12ms/epoch - 582us/step
Epoch 16/500
20/20 - 0s - loss: 0.3862 - accuracy: 0.8373 - 12ms/epoch - 581us/step
Epoch 17/500
20/20 - 0s - loss: 0.3838 - accuracy: 0.8373 - 12ms/epoch - 616us/step
Epoch 18/500
20/20 - 0s - loss: 0.3818 - accuracy: 0.8421 - 14ms/epoch - 689us/step
Epoch 19/500
20/20 - 0s - loss: 0.3797 - accuracy: 0.8453 - 12ms/epoch - 608us/step
Epoch 20/500
20/20 - 0s - loss: 0.3779 - accuracy: 0.8437 - 11ms/epoch - 572us/step
Epoch 21/500
20/20 - 0s - loss: 0.3762 - accuracy: 0.8437 - 11ms/epoch - 573us/step
Epoch 22/500
20/20 - 0s - loss: 0.3746 - accuracy: 0.8453 - 13ms/epoch - 657us/step
Epoch 23/500
20/20 - 0s - loss: 0.3730 - accuracy: 0.8469 - 13ms/epoch - 626us/step
Epoch 24/500
20/20 - 0s - loss: 0.3716 - accuracy: 0.8453 - 12ms/epoch - 595us/step
Epoch 25/500
20/20 - 0s - loss: 0.3702 - accuracy: 0.8437 - 13ms/epoch - 642us/step
Epoch 26/500
20/20 - 0s - loss: 0.3688 - accuracy: 0.8437 - 13ms/epoch - 649us/step
Epoch 27/500
20/20 - 0s - loss: 0.3675 - accuracy: 0.8437 - 13ms/epoch - 665us/step
Epoch 28/500
20/20 - 0s - loss: 0.3661 - accuracy: 0.8453 - 12ms/epoch - 620us/step
Epoch 29/500
20/20 - 0s - loss: 0.3649 - accuracy: 0.8453 - 13ms/epoch - 644us/step
Epoch 30/500
20/20 - 0s - loss: 0.3637 - accuracy: 0.8437 - 12ms/epoch - 594us/step
Epoch 31/500
20/20 - 0s - loss: 0.3625 - accuracy: 0.8421 - 11ms/epoch - 571us/step
Epoch 32/500
20/20 - 0s - loss: 0.3613 - accuracy: 0.8421 - 11ms/epoch - 563us/step
Epoch 33/500
20/20 - 0s - loss: 0.3602 - accuracy: 0.8437 - 12ms/epoch - 584us/step
Epoch 34/500
20/20 - 0s - loss: 0.3590 - accuracy: 0.8453 - 11ms/epoch - 553us/step
Epoch 35/500
20/20 - 0s - loss: 0.3580 - accuracy: 0.8453 - 12ms/epoch - 592us/step
Epoch 36/500
20/20 - 0s - loss: 0.3570 - accuracy: 0.8453 - 12ms/epoch - 582us/step
Epoch 37/500
20/20 - 0s - loss: 0.3559 - accuracy: 0.8453 - 11ms/epoch - 572us/step
Epoch 38/500
20/20 - 0s - loss: 0.3550 - accuracy: 0.8437 - 12ms/epoch - 592us/step
Epoch 39/500
20/20 - 0s - loss: 0.3539 - accuracy: 0.8453 - 12ms/epoch - 598us/step
Epoch 40/500
20/20 - 0s - loss: 0.3531 - accuracy: 0.8453 - 13ms/epoch - 638us/step
Epoch 41/500
20/20 - 0s - loss: 0.3521 - accuracy: 0.8469 - 12ms/epoch - 612us/step
Epoch 42/500
20/20 - 0s - loss: 0.3513 - accuracy: 0.8469 - 12ms/epoch - 587us/step
Epoch 43/500
20/20 - 0s - loss: 0.3505 - accuracy: 0.8453 - 13ms/epoch - 636us/step
Epoch 44/500
20/20 - 0s - loss: 0.3495 - accuracy: 0.8453 - 12ms/epoch - 607us/step
Epoch 45/500
20/20 - 0s - loss: 0.3488 - accuracy: 0.8453 - 13ms/epoch - 625us/step
Epoch 46/500
20/20 - 0s - loss: 0.3480 - accuracy: 0.8453 - 12ms/epoch - 617us/step
Epoch 47/500
20/20 - 0s - loss: 0.3472 - accuracy: 0.8453 - 12ms/epoch - 623us/step
Epoch 48/500
20/20 - 0s - loss: 0.3464 - accuracy: 0.8453 - 13ms/epoch - 660us/step
Epoch 49/500
20/20 - 0s - loss: 0.3456 - accuracy: 0.8453 - 12ms/epoch - 601us/step
Epoch 50/500
20/20 - 0s - loss: 0.3449 - accuracy: 0.8453 - 12ms/epoch - 605us/step
Epoch 51/500
20/20 - 0s - loss: 0.3441 - accuracy: 0.8453 - 12ms/epoch - 620us/step
Epoch 52/500
20/20 - 0s - loss: 0.3435 - accuracy: 0.8453 - 12ms/epoch - 606us/step
Epoch 53/500
20/20 - 0s - loss: 0.3425 - accuracy: 0.8437 - 12ms/epoch - 606us/step
Epoch 54/500
20/20 - 0s - loss: 0.3419 - accuracy: 0.8453 - 12ms/epoch - 589us/step
Epoch 55/500
20/20 - 0s - loss: 0.3411 - accuracy: 0.8453 - 11ms/epoch - 561us/step
Epoch 56/500
20/20 - 0s - loss: 0.3403 - accuracy: 0.8501 - 13ms/epoch - 638us/step
Epoch 57/500
20/20 - 0s - loss: 0.3396 - accuracy: 0.8485 - 11ms/epoch - 559us/step
Epoch 58/500
20/20 - 0s - loss: 0.3389 - accuracy: 0.8517 - 11ms/epoch - 561us/step
Epoch 59/500
20/20 - 0s - loss: 0.3382 - accuracy: 0.8501 - 12ms/epoch - 620us/step
Epoch 60/500
20/20 - 0s - loss: 0.3374 - accuracy: 0.8501 - 11ms/epoch - 558us/step
Epoch 61/500
20/20 - 0s - loss: 0.3369 - accuracy: 0.8517 - 11ms/epoch - 558us/step
Epoch 62/500
20/20 - 0s - loss: 0.3359 - accuracy: 0.8517 - 12ms/epoch - 589us/step
Epoch 63/500
20/20 - 0s - loss: 0.3354 - accuracy: 0.8517 - 11ms/epoch - 569us/step
Epoch 64/500
20/20 - 0s - loss: 0.3347 - accuracy: 0.8517 - 12ms/epoch - 586us/step
Epoch 65/500
20/20 - 0s - loss: 0.3339 - accuracy: 0.8533 - 11ms/epoch - 548us/step
Epoch 66/500
20/20 - 0s - loss: 0.3335 - accuracy: 0.8533 - 13ms/epoch - 639us/step
Epoch 67/500
20/20 - 0s - loss: 0.3326 - accuracy: 0.8565 - 13ms/epoch - 646us/step
Epoch 68/500
20/20 - 0s - loss: 0.3321 - accuracy: 0.8533 - 12ms/epoch - 613us/step
Epoch 69/500
20/20 - 0s - loss: 0.3315 - accuracy: 0.8549 - 12ms/epoch - 597us/step
Epoch 70/500
20/20 - 0s - loss: 0.3308 - accuracy: 0.8549 - 12ms/epoch - 623us/step
Epoch 71/500
20/20 - 0s - loss: 0.3302 - accuracy: 0.8533 - 13ms/epoch - 630us/step
Epoch 72/500
20/20 - 0s - loss: 0.3296 - accuracy: 0.8533 - 13ms/epoch - 636us/step
Epoch 73/500
20/20 - 0s - loss: 0.3290 - accuracy: 0.8533 - 14ms/epoch - 683us/step
Epoch 74/500
20/20 - 0s - loss: 0.3285 - accuracy: 0.8549 - 13ms/epoch - 642us/step
Epoch 75/500
20/20 - 0s - loss: 0.3278 - accuracy: 0.8565 - 12ms/epoch - 576us/step
Epoch 76/500
20/20 - 0s - loss: 0.3272 - accuracy: 0.8565 - 12ms/epoch - 597us/step
Epoch 77/500
20/20 - 0s - loss: 0.3268 - accuracy: 0.8581 - 11ms/epoch - 543us/step
Epoch 78/500
20/20 - 0s - loss: 0.3261 - accuracy: 0.8549 - 11ms/epoch - 536us/step
Epoch 79/500
20/20 - 0s - loss: 0.3255 - accuracy: 0.8581 - 12ms/epoch - 614us/step
Epoch 80/500
20/20 - 0s - loss: 0.3249 - accuracy: 0.8565 - 15ms/epoch - 747us/step
Epoch 81/500
20/20 - 0s - loss: 0.3243 - accuracy: 0.8596 - 14ms/epoch - 724us/step
Epoch 82/500
20/20 - 0s - loss: 0.3237 - accuracy: 0.8596 - 12ms/epoch - 593us/step
Epoch 83/500
20/20 - 0s - loss: 0.3231 - accuracy: 0.8612 - 15ms/epoch - 737us/step
Epoch 84/500
20/20 - 0s - loss: 0.3225 - accuracy: 0.8612 - 11ms/epoch - 564us/step
Epoch 85/500
20/20 - 0s - loss: 0.3220 - accuracy: 0.8612 - 12ms/epoch - 621us/step
Epoch 86/500
20/20 - 0s - loss: 0.3215 - accuracy: 0.8612 - 11ms/epoch - 541us/step
Epoch 87/500
20/20 - 0s - loss: 0.3208 - accuracy: 0.8612 - 12ms/epoch - 591us/step
Epoch 88/500
20/20 - 0s - loss: 0.3203 - accuracy: 0.8596 - 11ms/epoch - 553us/step
Epoch 89/500
20/20 - 0s - loss: 0.3198 - accuracy: 0.8596 - 11ms/epoch - 552us/step
Epoch 90/500
20/20 - 0s - loss: 0.3192 - accuracy: 0.8612 - 12ms/epoch - 623us/step
Epoch 91/500
20/20 - 0s - loss: 0.3186 - accuracy: 0.8596 - 12ms/epoch - 606us/step
Epoch 92/500
20/20 - 0s - loss: 0.3180 - accuracy: 0.8612 - 13ms/epoch - 633us/step
Epoch 93/500
20/20 - 0s - loss: 0.3174 - accuracy: 0.8612 - 17ms/epoch - 864us/step
Epoch 94/500
20/20 - 0s - loss: 0.3169 - accuracy: 0.8612 - 44ms/epoch - 2ms/step
Epoch 95/500
20/20 - 0s - loss: 0.3163 - accuracy: 0.8612 - 21ms/epoch - 1ms/step
Epoch 96/500
20/20 - 0s - loss: 0.3158 - accuracy: 0.8628 - 12ms/epoch - 600us/step
Epoch 97/500
20/20 - 0s - loss: 0.3152 - accuracy: 0.8644 - 12ms/epoch - 603us/step
Epoch 98/500
20/20 - 0s - loss: 0.3147 - accuracy: 0.8644 - 30ms/epoch - 1ms/step

Epoch 99/500
20/20 - 0s - loss: 0.3141 - accuracy: 0.8644 - 12ms/epoch - 622us/step
Epoch 100/500
20/20 - 0s - loss: 0.3136 - accuracy: 0.8628 - 12ms/epoch - 590us/step
Epoch 101/500
20/20 - 0s - loss: 0.3131 - accuracy: 0.8644 - 12ms/epoch - 606us/step
Epoch 102/500
20/20 - 0s - loss: 0.3125 - accuracy: 0.8660 - 11ms/epoch - 547us/step
Epoch 103/500
20/20 - 0s - loss: 0.3119 - accuracy: 0.8644 - 11ms/epoch - 559us/step
Epoch 104/500
20/20 - 0s - loss: 0.3114 - accuracy: 0.8644 - 11ms/epoch - 560us/step
Epoch 105/500
20/20 - 0s - loss: 0.3109 - accuracy: 0.8628 - 11ms/epoch - 555us/step
Epoch 106/500
20/20 - 0s - loss: 0.3104 - accuracy: 0.8644 - 11ms/epoch - 551us/step
Epoch 107/500
20/20 - 0s - loss: 0.3098 - accuracy: 0.8644 - 11ms/epoch - 554us/step
Epoch 108/500
20/20 - 0s - loss: 0.3092 - accuracy: 0.8628 - 11ms/epoch - 563us/step
Epoch 109/500
20/20 - 0s - loss: 0.3088 - accuracy: 0.8660 - 11ms/epoch - 554us/step
Epoch 110/500
20/20 - 0s - loss: 0.3083 - accuracy: 0.8660 - 11ms/epoch - 567us/step
Epoch 111/500
20/20 - 0s - loss: 0.3076 - accuracy: 0.8644 - 11ms/epoch - 553us/step
Epoch 112/500
20/20 - 0s - loss: 0.3073 - accuracy: 0.8676 - 11ms/epoch - 553us/step
Epoch 113/500
20/20 - 0s - loss: 0.3067 - accuracy: 0.8676 - 11ms/epoch - 547us/step
Epoch 114/500
20/20 - 0s - loss: 0.3062 - accuracy: 0.8676 - 11ms/epoch - 556us/step
Epoch 115/500
20/20 - 0s - loss: 0.3056 - accuracy: 0.8676 - 12ms/epoch - 594us/step
Epoch 116/500
20/20 - 0s - loss: 0.3051 - accuracy: 0.8676 - 13ms/epoch - 645us/step
Epoch 117/500
20/20 - 0s - loss: 0.3046 - accuracy: 0.8660 - 12ms/epoch - 586us/step
Epoch 118/500
20/20 - 0s - loss: 0.3042 - accuracy: 0.8692 - 12ms/epoch - 594us/step
Epoch 119/500
20/20 - 0s - loss: 0.3036 - accuracy: 0.8676 - 12ms/epoch - 578us/step
Epoch 120/500
20/20 - 0s - loss: 0.3031 - accuracy: 0.8724 - 12ms/epoch - 579us/step
Epoch 121/500
20/20 - 0s - loss: 0.3025 - accuracy: 0.8724 - 12ms/epoch - 590us/step
Epoch 122/500
20/20 - 0s - loss: 0.3022 - accuracy: 0.8724 - 12ms/epoch - 588us/step
Epoch 123/500
20/20 - 0s - loss: 0.3016 - accuracy: 0.8724 - 12ms/epoch - 586us/step
Epoch 124/500
20/20 - 0s - loss: 0.3012 - accuracy: 0.8708 - 12ms/epoch - 586us/step
Epoch 125/500
20/20 - 0s - loss: 0.3005 - accuracy: 0.8740 - 12ms/epoch - 586us/step
Epoch 126/500
20/20 - 0s - loss: 0.3002 - accuracy: 0.8740 - 12ms/epoch - 585us/step
Epoch 127/500
20/20 - 0s - loss: 0.2996 - accuracy: 0.8740 - 12ms/epoch - 583us/step
Epoch 128/500
20/20 - 0s - loss: 0.2991 - accuracy: 0.8724 - 11ms/epoch - 573us/step
Epoch 129/500
20/20 - 0s - loss: 0.2987 - accuracy: 0.8724 - 12ms/epoch - 584us/step
Epoch 130/500
20/20 - 0s - loss: 0.2983 - accuracy: 0.8724 - 12ms/epoch - 585us/step
Epoch 131/500
20/20 - 0s - loss: 0.2977 - accuracy: 0.8740 - 11ms/epoch - 574us/step
Epoch 132/500
20/20 - 0s - loss: 0.2973 - accuracy: 0.8740 - 12ms/epoch - 592us/step
Epoch 133/500
20/20 - 0s - loss: 0.2967 - accuracy: 0.8756 - 12ms/epoch - 587us/step
Epoch 134/500
20/20 - 0s - loss: 0.2964 - accuracy: 0.8740 - 11ms/epoch - 564us/step
Epoch 135/500
20/20 - 0s - loss: 0.2959 - accuracy: 0.8756 - 11ms/epoch - 572us/step
Epoch 136/500
20/20 - 0s - loss: 0.2954 - accuracy: 0.8740 - 12ms/epoch - 578us/step
Epoch 137/500
20/20 - 0s - loss: 0.2950 - accuracy: 0.8740 - 12ms/epoch - 601us/step
Epoch 138/500
20/20 - 0s - loss: 0.2944 - accuracy: 0.8756 - 11ms/epoch - 562us/step
Epoch 139/500
20/20 - 0s - loss: 0.2941 - accuracy: 0.8740 - 12ms/epoch - 611us/step
Epoch 140/500
20/20 - 0s - loss: 0.2935 - accuracy: 0.8740 - 11ms/epoch - 557us/step
Epoch 141/500
20/20 - 0s - loss: 0.2931 - accuracy: 0.8740 - 12ms/epoch - 596us/step
Epoch 142/500
20/20 - 0s - loss: 0.2926 - accuracy: 0.8740 - 12ms/epoch - 580us/step
Epoch 143/500
20/20 - 0s - loss: 0.2922 - accuracy: 0.8756 - 11ms/epoch - 573us/step
Epoch 144/500
20/20 - 0s - loss: 0.2917 - accuracy: 0.8756 - 12ms/epoch - 586us/step
Epoch 145/500
20/20 - 0s - loss: 0.2912 - accuracy: 0.8740 - 11ms/epoch - 569us/step
Epoch 146/500
20/20 - 0s - loss: 0.2910 - accuracy: 0.8740 - 11ms/epoch - 574us/step
Epoch 147/500
20/20 - 0s - loss: 0.2903 - accuracy: 0.8740 - 11ms/epoch - 569us/step
Epoch 148/500
20/20 - 0s - loss: 0.2900 - accuracy: 0.8756 - 12ms/epoch - 605us/step
Epoch 149/500
20/20 - 0s - loss: 0.2896 - accuracy: 0.8740 - 11ms/epoch - 560us/step
Epoch 150/500
20/20 - 0s - loss: 0.2891 - accuracy: 0.8756 - 12ms/epoch - 577us/step
Epoch 151/500
20/20 - 0s - loss: 0.2886 - accuracy: 0.8772 - 12ms/epoch - 598us/step
Epoch 152/500
20/20 - 0s - loss: 0.2882 - accuracy: 0.8788 - 12ms/epoch - 583us/step
Epoch 153/500
20/20 - 0s - loss: 0.2878 - accuracy: 0.8772 - 12ms/epoch - 594us/step
Epoch 154/500
20/20 - 0s - loss: 0.2874 - accuracy: 0.8788 - 12ms/epoch - 579us/step
Epoch 155/500
20/20 - 0s - loss: 0.2869 - accuracy: 0.8756 - 12ms/epoch - 579us/step
Epoch 156/500
20/20 - 0s - loss: 0.2866 - accuracy: 0.8788 - 11ms/epoch - 564us/step
Epoch 157/500
20/20 - 0s - loss: 0.2861 - accuracy: 0.8756 - 11ms/epoch - 572us/step
Epoch 158/500
20/20 - 0s - loss: 0.2854 - accuracy: 0.8772 - 11ms/epoch - 565us/step
Epoch 159/500
20/20 - 0s - loss: 0.2853 - accuracy: 0.8772 - 11ms/epoch - 564us/step
Epoch 160/500
20/20 - 0s - loss: 0.2847 - accuracy: 0.8772 - 11ms/epoch - 570us/step
Epoch 161/500
20/20 - 0s - loss: 0.2843 - accuracy: 0.8772 - 11ms/epoch - 566us/step
Epoch 162/500
20/20 - 0s - loss: 0.2840 - accuracy: 0.8788 - 12ms/epoch - 587us/step
Epoch 163/500
20/20 - 0s - loss: 0.2834 - accuracy: 0.8788 - 11ms/epoch - 574us/step
Epoch 164/500
20/20 - 0s - loss: 0.2831 - accuracy: 0.8788 - 12ms/epoch - 606us/step
Epoch 165/500
20/20 - 0s - loss: 0.2827 - accuracy: 0.8788 - 11ms/epoch - 561us/step
Epoch 166/500
20/20 - 0s - loss: 0.2821 - accuracy: 0.8788 - 12ms/epoch - 589us/step
Epoch 167/500
20/20 - 0s - loss: 0.2819 - accuracy: 0.8788 - 11ms/epoch - 570us/step
Epoch 168/500
20/20 - 0s - loss: 0.2812 - accuracy: 0.8804 - 11ms/epoch - 568us/step
Epoch 169/500
20/20 - 0s - loss: 0.2811 - accuracy: 0.8804 - 12ms/epoch - 580us/step
Epoch 170/500
20/20 - 0s - loss: 0.2805 - accuracy: 0.8804 - 12ms/epoch - 587us/step
Epoch 171/500
20/20 - 0s - loss: 0.2802 - accuracy: 0.8804 - 11ms/epoch - 570us/step
Epoch 172/500
20/20 - 0s - loss: 0.2798 - accuracy: 0.8804 - 13ms/epoch - 657us/step
Epoch 173/500
20/20 - 0s - loss: 0.2793 - accuracy: 0.8804 - 12ms/epoch - 588us/step
Epoch 174/500
20/20 - 0s - loss: 0.2789 - accuracy: 0.8804 - 12ms/epoch - 578us/step
Epoch 175/500
20/20 - 0s - loss: 0.2786 - accuracy: 0.8804 - 12ms/epoch - 593us/step
Epoch 176/500
20/20 - 0s - loss: 0.2779 - accuracy: 0.8804 - 12ms/epoch - 595us/step
Epoch 177/500
20/20 - 0s - loss: 0.2779 - accuracy: 0.8788 - 12ms/epoch - 580us/step
Epoch 178/500
20/20 - 0s - loss: 0.2773 - accuracy: 0.8804 - 12ms/epoch - 587us/step
Epoch 179/500
20/20 - 0s - loss: 0.2769 - accuracy: 0.8804 - 12ms/epoch - 598us/step
Epoch 180/500
20/20 - 0s - loss: 0.2765 - accuracy: 0.8804 - 12ms/epoch - 583us/step
Epoch 181/500
20/20 - 0s - loss: 0.2760 - accuracy: 0.8804 - 12ms/epoch - 586us/step
Epoch 182/500
20/20 - 0s - loss: 0.2757 - accuracy: 0.8804 - 12ms/epoch - 582us/step
Epoch 183/500
20/20 - 0s - loss: 0.2753 - accuracy: 0.8804 - 12ms/epoch - 596us/step
Epoch 184/500
20/20 - 0s - loss: 0.2748 - accuracy: 0.8804 - 12ms/epoch - 590us/step
Epoch 185/500
20/20 - 0s - loss: 0.2745 - accuracy: 0.8804 - 12ms/epoch - 596us/step
Epoch 186/500
20/20 - 0s - loss: 0.2741 - accuracy: 0.8804 - 12ms/epoch - 583us/step
Epoch 187/500
20/20 - 0s - loss: 0.2737 - accuracy: 0.8804 - 12ms/epoch - 583us/step
Epoch 188/500
20/20 - 0s - loss: 0.2734 - accuracy: 0.8804 - 12ms/epoch - 585us/step
Epoch 189/500
20/20 - 0s - loss: 0.2729 - accuracy: 0.8804 - 12ms/epoch - 589us/step
Epoch 190/500
20/20 - 0s - loss: 0.2727 - accuracy: 0.8852 - 12ms/epoch - 588us/step
Epoch 191/500
20/20 - 0s - loss: 0.2719 - accuracy: 0.8820 - 12ms/epoch - 584us/step
Epoch 192/500
20/20 - 0s - loss: 0.2717 - accuracy: 0.8804 - 12ms/epoch - 598us/step
Epoch 193/500
20/20 - 0s - loss: 0.2714 - accuracy: 0.8852 - 12ms/epoch - 588us/step
Epoch 194/500
20/20 - 0s - loss: 0.2709 - accuracy: 0.8836 - 11ms/epoch - 551us/step
Epoch 195/500
20/20 - 0s - loss: 0.2705 - accuracy: 0.8852 - 12ms/epoch - 603us/step

Epoch 196/500
20/20 - 0s - loss: 0.2702 - accuracy: 0.8836 - 12ms/epoch - 593us/step
Epoch 197/500
20/20 - 0s - loss: 0.2698 - accuracy: 0.8868 - 12ms/epoch - 578us/step
Epoch 198/500
20/20 - 0s - loss: 0.2694 - accuracy: 0.8900 - 12ms/epoch - 599us/step
Epoch 199/500
20/20 - 0s - loss: 0.2689 - accuracy: 0.8900 - 12ms/epoch - 587us/step
Epoch 200/500
20/20 - 0s - loss: 0.2687 - accuracy: 0.8900 - 12ms/epoch - 584us/step
Epoch 201/500
20/20 - 0s - loss: 0.2682 - accuracy: 0.8931 - 12ms/epoch - 599us/step
Epoch 202/500
20/20 - 0s - loss: 0.2678 - accuracy: 0.8931 - 11ms/epoch - 552us/step
Epoch 203/500
20/20 - 0s - loss: 0.2676 - accuracy: 0.8931 - 12ms/epoch - 586us/step
Epoch 204/500
20/20 - 0s - loss: 0.2672 - accuracy: 0.8931 - 12ms/epoch - 599us/step
Epoch 205/500
20/20 - 0s - loss: 0.2668 - accuracy: 0.8931 - 12ms/epoch - 582us/step
Epoch 206/500
20/20 - 0s - loss: 0.2664 - accuracy: 0.8931 - 12ms/epoch - 585us/step
Epoch 207/500
20/20 - 0s - loss: 0.2661 - accuracy: 0.8931 - 12ms/epoch - 583us/step
Epoch 208/500
20/20 - 0s - loss: 0.2658 - accuracy: 0.8947 - 12ms/epoch - 592us/step
Epoch 209/500
20/20 - 0s - loss: 0.2653 - accuracy: 0.8931 - 12ms/epoch - 579us/step
Epoch 210/500
20/20 - 0s - loss: 0.2651 - accuracy: 0.8963 - 11ms/epoch - 574us/step
Epoch 211/500
20/20 - 0s - loss: 0.2647 - accuracy: 0.8931 - 22ms/epoch - 1ms/step
Epoch 212/500
20/20 - 0s - loss: 0.2644 - accuracy: 0.8931 - 12ms/epoch - 579us/step
Epoch 213/500
20/20 - 0s - loss: 0.2639 - accuracy: 0.8947 - 12ms/epoch - 593us/step
Epoch 214/500
20/20 - 0s - loss: 0.2636 - accuracy: 0.8963 - 12ms/epoch - 593us/step
Epoch 215/500
20/20 - 0s - loss: 0.2634 - accuracy: 0.8947 - 12ms/epoch - 622us/step
Epoch 216/500
20/20 - 0s - loss: 0.2629 - accuracy: 0.8963 - 12ms/epoch - 588us/step
Epoch 217/500
20/20 - 0s - loss: 0.2626 - accuracy: 0.8947 - 12ms/epoch - 584us/step
Epoch 218/500
20/20 - 0s - loss: 0.2622 - accuracy: 0.8963 - 12ms/epoch - 585us/step
Epoch 219/500
20/20 - 0s - loss: 0.2619 - accuracy: 0.8963 - 12ms/epoch - 606us/step
Epoch 220/500
20/20 - 0s - loss: 0.2614 - accuracy: 0.8963 - 12ms/epoch - 587us/step
Epoch 221/500
20/20 - 0s - loss: 0.2614 - accuracy: 0.8979 - 12ms/epoch - 598us/step
Epoch 222/500
20/20 - 0s - loss: 0.2609 - accuracy: 0.8979 - 12ms/epoch - 592us/step
Epoch 223/500
20/20 - 0s - loss: 0.2604 - accuracy: 0.8963 - 12ms/epoch - 603us/step
Epoch 224/500
20/20 - 0s - loss: 0.2603 - accuracy: 0.8979 - 11ms/epoch - 572us/step
Epoch 225/500
20/20 - 0s - loss: 0.2597 - accuracy: 0.8979 - 12ms/epoch - 585us/step
Epoch 226/500
20/20 - 0s - loss: 0.2597 - accuracy: 0.8979 - 12ms/epoch - 597us/step
Epoch 227/500
20/20 - 0s - loss: 0.2591 - accuracy: 0.8979 - 11ms/epoch - 548us/step
Epoch 228/500
20/20 - 0s - loss: 0.2589 - accuracy: 0.8979 - 12ms/epoch - 605us/step
Epoch 229/500
20/20 - 0s - loss: 0.2587 - accuracy: 0.8979 - 11ms/epoch - 537us/step
Epoch 230/500
20/20 - 0s - loss: 0.2582 - accuracy: 0.8979 - 12ms/epoch - 596us/step
Epoch 231/500
20/20 - 0s - loss: 0.2580 - accuracy: 0.8979 - 12ms/epoch - 576us/step
Epoch 232/500
20/20 - 0s - loss: 0.2575 - accuracy: 0.8979 - 12ms/epoch - 604us/step
Epoch 233/500
20/20 - 0s - loss: 0.2572 - accuracy: 0.8979 - 12ms/epoch - 575us/step
Epoch 234/500
20/20 - 0s - loss: 0.2571 - accuracy: 0.8979 - 13ms/epoch - 651us/step
Epoch 235/500
20/20 - 0s - loss: 0.2566 - accuracy: 0.8979 - 12ms/epoch - 607us/step
Epoch 236/500
20/20 - 0s - loss: 0.2564 - accuracy: 0.8979 - 12ms/epoch - 591us/step
Epoch 237/500
20/20 - 0s - loss: 0.2561 - accuracy: 0.8995 - 12ms/epoch - 596us/step
Epoch 238/500
20/20 - 0s - loss: 0.2556 - accuracy: 0.8995 - 12ms/epoch - 601us/step
Epoch 239/500
20/20 - 0s - loss: 0.2556 - accuracy: 0.8995 - 12ms/epoch - 598us/step
Epoch 240/500
20/20 - 0s - loss: 0.2551 - accuracy: 0.8995 - 12ms/epoch - 583us/step
Epoch 241/500
20/20 - 0s - loss: 0.2549 - accuracy: 0.8995 - 12ms/epoch - 582us/step
Epoch 242/500
20/20 - 0s - loss: 0.2544 - accuracy: 0.8995 - 12ms/epoch - 594us/step
Epoch 243/500
20/20 - 0s - loss: 0.2542 - accuracy: 0.8995 - 12ms/epoch - 585us/step
Epoch 244/500
20/20 - 0s - loss: 0.2538 - accuracy: 0.9027 - 12ms/epoch - 593us/step
Epoch 245/500
20/20 - 0s - loss: 0.2538 - accuracy: 0.9011 - 12ms/epoch - 595us/step
Epoch 246/500
20/20 - 0s - loss: 0.2533 - accuracy: 0.9027 - 12ms/epoch - 583us/step
Epoch 247/500
20/20 - 0s - loss: 0.2531 - accuracy: 0.9027 - 12ms/epoch - 608us/step
Epoch 248/500
20/20 - 0s - loss: 0.2526 - accuracy: 0.9027 - 12ms/epoch - 575us/step
Epoch 249/500
20/20 - 0s - loss: 0.2525 - accuracy: 0.9027 - 12ms/epoch - 579us/step
Epoch 250/500
20/20 - 0s - loss: 0.2521 - accuracy: 0.9027 - 12ms/epoch - 587us/step
Epoch 251/500
20/20 - 0s - loss: 0.2516 - accuracy: 0.9027 - 12ms/epoch - 596us/step
Epoch 252/500
20/20 - 0s - loss: 0.2516 - accuracy: 0.9011 - 12ms/epoch - 586us/step
Epoch 253/500
20/20 - 0s - loss: 0.2515 - accuracy: 0.9027 - 12ms/epoch - 598us/step
Epoch 254/500
20/20 - 0s - loss: 0.2507 - accuracy: 0.9059 - 12ms/epoch - 599us/step
Epoch 255/500
20/20 - 0s - loss: 0.2509 - accuracy: 0.9043 - 12ms/epoch - 580us/step
Epoch 256/500
20/20 - 0s - loss: 0.2502 - accuracy: 0.9043 - 12ms/epoch - 587us/step
Epoch 257/500
20/20 - 0s - loss: 0.2501 - accuracy: 0.9043 - 12ms/epoch - 585us/step
Epoch 258/500
20/20 - 0s - loss: 0.2498 - accuracy: 0.9027 - 12ms/epoch - 611us/step
Epoch 259/500
20/20 - 0s - loss: 0.2494 - accuracy: 0.9027 - 12ms/epoch - 605us/step
Epoch 260/500
20/20 - 0s - loss: 0.2491 - accuracy: 0.9075 - 12ms/epoch - 587us/step
Epoch 261/500
20/20 - 0s - loss: 0.2488 - accuracy: 0.9075 - 12ms/epoch - 589us/step
Epoch 262/500
20/20 - 0s - loss: 0.2486 - accuracy: 0.9059 - 12ms/epoch - 586us/step
Epoch 263/500
20/20 - 0s - loss: 0.2482 - accuracy: 0.9075 - 12ms/epoch - 591us/step
Epoch 264/500
20/20 - 0s - loss: 0.2477 - accuracy: 0.9059 - 12ms/epoch - 603us/step
Epoch 265/500
20/20 - 0s - loss: 0.2481 - accuracy: 0.9043 - 12ms/epoch - 609us/step
Epoch 266/500
20/20 - 0s - loss: 0.2472 - accuracy: 0.9091 - 12ms/epoch - 601us/step
Epoch 267/500
20/20 - 0s - loss: 0.2473 - accuracy: 0.9091 - 12ms/epoch - 578us/step
Epoch 268/500
20/20 - 0s - loss: 0.2470 - accuracy: 0.9075 - 12ms/epoch - 601us/step
Epoch 269/500
20/20 - 0s - loss: 0.2465 - accuracy: 0.9091 - 12ms/epoch - 617us/step
Epoch 270/500
20/20 - 0s - loss: 0.2465 - accuracy: 0.9091 - 12ms/epoch - 608us/step
Epoch 271/500
20/20 - 0s - loss: 0.2460 - accuracy: 0.9091 - 12ms/epoch - 610us/step
Epoch 272/500
20/20 - 0s - loss: 0.2458 - accuracy: 0.9075 - 13ms/epoch - 641us/step
Epoch 273/500
20/20 - 0s - loss: 0.2455 - accuracy: 0.9075 - 12ms/epoch - 604us/step
Epoch 274/500
20/20 - 0s - loss: 0.2453 - accuracy: 0.9091 - 12ms/epoch - 612us/step
Epoch 275/500
20/20 - 0s - loss: 0.2450 - accuracy: 0.9107 - 12ms/epoch - 597us/step
Epoch 276/500
20/20 - 0s - loss: 0.2447 - accuracy: 0.9107 - 13ms/epoch - 639us/step
Epoch 277/500
20/20 - 0s - loss: 0.2444 - accuracy: 0.9123 - 12ms/epoch - 608us/step
Epoch 278/500
20/20 - 0s - loss: 0.2444 - accuracy: 0.9107 - 13ms/epoch - 639us/step
Epoch 279/500
20/20 - 0s - loss: 0.2439 - accuracy: 0.9091 - 12ms/epoch - 580us/step
Epoch 280/500
20/20 - 0s - loss: 0.2439 - accuracy: 0.9091 - 11ms/epoch - 559us/step
Epoch 281/500
20/20 - 0s - loss: 0.2434 - accuracy: 0.9107 - 11ms/epoch - 549us/step
Epoch 282/500
20/20 - 0s - loss: 0.2434 - accuracy: 0.9107 - 11ms/epoch - 556us/step
Epoch 283/500
20/20 - 0s - loss: 0.2429 - accuracy: 0.9123 - 11ms/epoch - 556us/step
Epoch 284/500
20/20 - 0s - loss: 0.2427 - accuracy: 0.9091 - 11ms/epoch - 555us/step
Epoch 285/500
20/20 - 0s - loss: 0.2424 - accuracy: 0.9107 - 11ms/epoch - 557us/step
Epoch 286/500
20/20 - 0s - loss: 0.2423 - accuracy: 0.9107 - 11ms/epoch - 567us/step
Epoch 287/500
20/20 - 0s - loss: 0.2419 - accuracy: 0.9107 - 11ms/epoch - 560us/step
Epoch 288/500
20/20 - 0s - loss: 0.2416 - accuracy: 0.9107 - 11ms/epoch - 547us/step
Epoch 289/500
20/20 - 0s - loss: 0.2414 - accuracy: 0.9107 - 11ms/epoch - 546us/step
Epoch 290/500
20/20 - 0s - loss: 0.2414 - accuracy: 0.9123 - 11ms/epoch - 541us/step
Epoch 291/500
20/20 - 0s - loss: 0.2408 - accuracy: 0.9107 - 11ms/epoch - 557us/step
Epoch 292/500
20/20 - 0s - loss: 0.2408 - accuracy: 0.9123 - 11ms/epoch - 552us/step

Epoch 293/500
20/20 - 0s - loss: 0.2402 - accuracy: 0.9107 - 11ms/epoch - 538us/step
Epoch 294/500
20/20 - 0s - loss: 0.2402 - accuracy: 0.9107 - 11ms/epoch - 553us/step
Epoch 295/500
20/20 - 0s - loss: 0.2401 - accuracy: 0.9107 - 11ms/epoch - 554us/step
Epoch 296/500
20/20 - 0s - loss: 0.2395 - accuracy: 0.9107 - 11ms/epoch - 545us/step
Epoch 297/500
20/20 - 0s - loss: 0.2394 - accuracy: 0.9123 - 11ms/epoch - 549us/step
Epoch 298/500
20/20 - 0s - loss: 0.2394 - accuracy: 0.9107 - 11ms/epoch - 557us/step
Epoch 299/500
20/20 - 0s - loss: 0.2388 - accuracy: 0.9123 - 11ms/epoch - 552us/step
Epoch 300/500
20/20 - 0s - loss: 0.2387 - accuracy: 0.9107 - 11ms/epoch - 548us/step
Epoch 301/500
20/20 - 0s - loss: 0.2385 - accuracy: 0.9123 - 11ms/epoch - 547us/step
Epoch 302/500
20/20 - 0s - loss: 0.2383 - accuracy: 0.9107 - 11ms/epoch - 548us/step
Epoch 303/500
20/20 - 0s - loss: 0.2380 - accuracy: 0.9123 - 11ms/epoch - 549us/step
Epoch 304/500
20/20 - 0s - loss: 0.2376 - accuracy: 0.9123 - 11ms/epoch - 547us/step
Epoch 305/500
20/20 - 0s - loss: 0.2377 - accuracy: 0.9107 - 11ms/epoch - 546us/step
Epoch 306/500
20/20 - 0s - loss: 0.2373 - accuracy: 0.9107 - 11ms/epoch - 556us/step
Epoch 307/500
20/20 - 0s - loss: 0.2371 - accuracy: 0.9123 - 11ms/epoch - 562us/step
Epoch 308/500
20/20 - 0s - loss: 0.2367 - accuracy: 0.9107 - 11ms/epoch - 568us/step
Epoch 309/500
20/20 - 0s - loss: 0.2367 - accuracy: 0.9107 - 12ms/epoch - 588us/step
Epoch 310/500
20/20 - 0s - loss: 0.2366 - accuracy: 0.9107 - 13ms/epoch - 633us/step
Epoch 311/500
20/20 - 0s - loss: 0.2361 - accuracy: 0.9107 - 12ms/epoch - 591us/step
Epoch 312/500
20/20 - 0s - loss: 0.2360 - accuracy: 0.9107 - 12ms/epoch - 589us/step
Epoch 313/500
20/20 - 0s - loss: 0.2357 - accuracy: 0.9123 - 12ms/epoch - 599us/step
Epoch 314/500
20/20 - 0s - loss: 0.2355 - accuracy: 0.9107 - 11ms/epoch - 572us/step
Epoch 315/500
20/20 - 0s - loss: 0.2351 - accuracy: 0.9107 - 12ms/epoch - 582us/step
Epoch 316/500
20/20 - 0s - loss: 0.2352 - accuracy: 0.9123 - 11ms/epoch - 562us/step
Epoch 317/500
20/20 - 0s - loss: 0.2348 - accuracy: 0.9107 - 12ms/epoch - 587us/step
Epoch 318/500
20/20 - 0s - loss: 0.2346 - accuracy: 0.9107 - 12ms/epoch - 587us/step
Epoch 319/500
20/20 - 0s - loss: 0.2344 - accuracy: 0.9107 - 12ms/epoch - 579us/step
Epoch 320/500
20/20 - 0s - loss: 0.2341 - accuracy: 0.9107 - 12ms/epoch - 599us/step
Epoch 321/500
20/20 - 0s - loss: 0.2341 - accuracy: 0.9107 - 12ms/epoch - 592us/step
Epoch 322/500
20/20 - 0s - loss: 0.2336 - accuracy: 0.9107 - 12ms/epoch - 588us/step
Epoch 323/500
20/20 - 0s - loss: 0.2336 - accuracy: 0.9107 - 12ms/epoch - 599us/step
Epoch 324/500
20/20 - 0s - loss: 0.2331 - accuracy: 0.9123 - 12ms/epoch - 595us/step
Epoch 325/500
20/20 - 0s - loss: 0.2329 - accuracy: 0.9123 - 12ms/epoch - 587us/step
Epoch 326/500
20/20 - 0s - loss: 0.2331 - accuracy: 0.9107 - 19ms/epoch - 974us/step
Epoch 327/500
20/20 - 0s - loss: 0.2326 - accuracy: 0.9107 - 11ms/epoch - 568us/step
Epoch 328/500
20/20 - 0s - loss: 0.2323 - accuracy: 0.9107 - 11ms/epoch - 557us/step
Epoch 329/500
20/20 - 0s - loss: 0.2323 - accuracy: 0.9139 - 11ms/epoch - 552us/step
Epoch 330/500
20/20 - 0s - loss: 0.2320 - accuracy: 0.9091 - 11ms/epoch - 554us/step
Epoch 331/500
20/20 - 0s - loss: 0.2319 - accuracy: 0.9091 - 11ms/epoch - 559us/step
Epoch 332/500
20/20 - 0s - loss: 0.2315 - accuracy: 0.9107 - 11ms/epoch - 561us/step
Epoch 333/500
20/20 - 0s - loss: 0.2312 - accuracy: 0.9107 - 11ms/epoch - 546us/step
Epoch 334/500
20/20 - 0s - loss: 0.2316 - accuracy: 0.9107 - 11ms/epoch - 550us/step
Epoch 335/500
20/20 - 0s - loss: 0.2307 - accuracy: 0.9107 - 12ms/epoch - 611us/step
Epoch 336/500
20/20 - 0s - loss: 0.2310 - accuracy: 0.9091 - 11ms/epoch - 562us/step
Epoch 337/500
20/20 - 0s - loss: 0.2307 - accuracy: 0.9107 - 11ms/epoch - 550us/step
Epoch 338/500
20/20 - 0s - loss: 0.2298 - accuracy: 0.9123 - 12ms/epoch - 576us/step
Epoch 339/500
20/20 - 0s - loss: 0.2308 - accuracy: 0.9107 - 12ms/epoch - 592us/step
Epoch 340/500
20/20 - 0s - loss: 0.2298 - accuracy: 0.9123 - 11ms/epoch - 566us/step
Epoch 341/500
20/20 - 0s - loss: 0.2300 - accuracy: 0.9107 - 12ms/epoch - 592us/step
Epoch 342/500
20/20 - 0s - loss: 0.2297 - accuracy: 0.9107 - 11ms/epoch - 566us/step
Epoch 343/500
20/20 - 0s - loss: 0.2293 - accuracy: 0.9107 - 12ms/epoch - 599us/step
Epoch 344/500
20/20 - 0s - loss: 0.2293 - accuracy: 0.9107 - 11ms/epoch - 569us/step
Epoch 345/500
20/20 - 0s - loss: 0.2290 - accuracy: 0.9107 - 12ms/epoch - 578us/step
Epoch 346/500
20/20 - 0s - loss: 0.2288 - accuracy: 0.9107 - 12ms/epoch - 606us/step
Epoch 347/500
20/20 - 0s - loss: 0.2287 - accuracy: 0.9107 - 12ms/epoch - 577us/step
Epoch 348/500
20/20 - 0s - loss: 0.2286 - accuracy: 0.9107 - 12ms/epoch - 583us/step
Epoch 349/500
20/20 - 0s - loss: 0.2280 - accuracy: 0.9123 - 12ms/epoch - 604us/step
Epoch 350/500
20/20 - 0s - loss: 0.2285 - accuracy: 0.9107 - 12ms/epoch - 594us/step
Epoch 351/500
20/20 - 0s - loss: 0.2276 - accuracy: 0.9123 - 11ms/epoch - 575us/step
Epoch 352/500
20/20 - 0s - loss: 0.2277 - accuracy: 0.9123 - 12ms/epoch - 581us/step
Epoch 353/500
20/20 - 0s - loss: 0.2275 - accuracy: 0.9139 - 11ms/epoch - 560us/step
Epoch 354/500
20/20 - 0s - loss: 0.2273 - accuracy: 0.9139 - 12ms/epoch - 577us/step
Epoch 355/500
20/20 - 0s - loss: 0.2273 - accuracy: 0.9139 - 11ms/epoch - 573us/step
Epoch 356/500
20/20 - 0s - loss: 0.2268 - accuracy: 0.9155 - 11ms/epoch - 574us/step
Epoch 357/500
20/20 - 0s - loss: 0.2267 - accuracy: 0.9139 - 12ms/epoch - 595us/step
Epoch 358/500
20/20 - 0s - loss: 0.2266 - accuracy: 0.9123 - 12ms/epoch - 579us/step
Epoch 359/500
20/20 - 0s - loss: 0.2267 - accuracy: 0.9123 - 12ms/epoch - 586us/step
Epoch 360/500
20/20 - 0s - loss: 0.2260 - accuracy: 0.9139 - 11ms/epoch - 574us/step
Epoch 361/500
20/20 - 0s - loss: 0.2262 - accuracy: 0.9123 - 11ms/epoch - 572us/step
Epoch 362/500
20/20 - 0s - loss: 0.2257 - accuracy: 0.9139 - 11ms/epoch - 562us/step
Epoch 363/500
20/20 - 0s - loss: 0.2258 - accuracy: 0.9139 - 12ms/epoch - 609us/step
Epoch 364/500
20/20 - 0s - loss: 0.2252 - accuracy: 0.9139 - 12ms/epoch - 583us/step
Epoch 365/500
20/20 - 0s - loss: 0.2255 - accuracy: 0.9123 - 12ms/epoch - 587us/step
Epoch 366/500
20/20 - 0s - loss: 0.2249 - accuracy: 0.9123 - 12ms/epoch - 582us/step
Epoch 367/500
20/20 - 0s - loss: 0.2250 - accuracy: 0.9139 - 12ms/epoch - 588us/step
Epoch 368/500
20/20 - 0s - loss: 0.2247 - accuracy: 0.9139 - 12ms/epoch - 586us/step
Epoch 369/500
20/20 - 0s - loss: 0.2245 - accuracy: 0.9139 - 12ms/epoch - 587us/step
Epoch 370/500
20/20 - 0s - loss: 0.2245 - accuracy: 0.9155 - 12ms/epoch - 586us/step
Epoch 371/500
20/20 - 0s - loss: 0.2241 - accuracy: 0.9155 - 12ms/epoch - 594us/step
Epoch 372/500
20/20 - 0s - loss: 0.2240 - accuracy: 0.9139 - 11ms/epoch - 550us/step
Epoch 373/500
20/20 - 0s - loss: 0.2238 - accuracy: 0.9123 - 12ms/epoch - 600us/step
Epoch 374/500
20/20 - 0s - loss: 0.2237 - accuracy: 0.9139 - 12ms/epoch - 578us/step
Epoch 375/500
20/20 - 0s - loss: 0.2235 - accuracy: 0.9139 - 12ms/epoch - 601us/step
Epoch 376/500
20/20 - 0s - loss: 0.2229 - accuracy: 0.9139 - 11ms/epoch - 571us/step
Epoch 377/500
20/20 - 0s - loss: 0.2233 - accuracy: 0.9139 - 12ms/epoch - 585us/step
Epoch 378/500
20/20 - 0s - loss: 0.2227 - accuracy: 0.9155 - 11ms/epoch - 539us/step
Epoch 379/500
20/20 - 0s - loss: 0.2226 - accuracy: 0.9139 - 12ms/epoch - 614us/step
Epoch 380/500
20/20 - 0s - loss: 0.2227 - accuracy: 0.9155 - 12ms/epoch - 612us/step
Epoch 381/500
20/20 - 0s - loss: 0.2222 - accuracy: 0.9139 - 12ms/epoch - 607us/step
Epoch 382/500
20/20 - 0s - loss: 0.2221 - accuracy: 0.9139 - 12ms/epoch - 591us/step
Epoch 383/500
20/20 - 0s - loss: 0.2221 - accuracy: 0.9155 - 12ms/epoch - 582us/step
Epoch 384/500
20/20 - 0s - loss: 0.2219 - accuracy: 0.9155 - 12ms/epoch - 590us/step
Epoch 385/500
20/20 - 0s - loss: 0.2217 - accuracy: 0.9171 - 12ms/epoch - 598us/step
Epoch 386/500
20/20 - 0s - loss: 0.2215 - accuracy: 0.9155 - 12ms/epoch - 599us/step
Epoch 387/500
20/20 - 0s - loss: 0.2212 - accuracy: 0.9139 - 12ms/epoch - 586us/step
Epoch 388/500
20/20 - 0s - loss: 0.2210 - accuracy: 0.9155 - 12ms/epoch - 600us/step
Epoch 389/500
20/20 - 0s - loss: 0.2211 - accuracy: 0.9171 - 12ms/epoch - 618us/step

Epoch 390/500
20/20 - 0s - loss: 0.2204 - accuracy: 0.9139 - 13ms/epoch - 658us/step
Epoch 391/500
20/20 - 0s - loss: 0.2210 - accuracy: 0.9155 - 25ms/epoch - 1ms/step
Epoch 392/500
20/20 - 0s - loss: 0.2203 - accuracy: 0.9155 - 35ms/epoch - 2ms/step
Epoch 393/500
20/20 - 0s - loss: 0.2204 - accuracy: 0.9155 - 17ms/epoch - 848us/step
Epoch 394/500
20/20 - 0s - loss: 0.2199 - accuracy: 0.9171 - 20ms/epoch - 1ms/step
Epoch 395/500
20/20 - 0s - loss: 0.2202 - accuracy: 0.9171 - 18ms/epoch - 879us/step
Epoch 396/500
20/20 - 0s - loss: 0.2198 - accuracy: 0.9155 - 14ms/epoch - 716us/step
Epoch 397/500
20/20 - 0s - loss: 0.2194 - accuracy: 0.9155 - 11ms/epoch - 561us/step
Epoch 398/500
20/20 - 0s - loss: 0.2198 - accuracy: 0.9155 - 11ms/epoch - 544us/step
Epoch 399/500
20/20 - 0s - loss: 0.2194 - accuracy: 0.9155 - 11ms/epoch - 550us/step
Epoch 400/500
20/20 - 0s - loss: 0.2190 - accuracy: 0.9171 - 11ms/epoch - 561us/step
Epoch 401/500
20/20 - 0s - loss: 0.2191 - accuracy: 0.9155 - 11ms/epoch - 549us/step
Epoch 402/500
20/20 - 0s - loss: 0.2187 - accuracy: 0.9139 - 11ms/epoch - 541us/step
Epoch 403/500
20/20 - 0s - loss: 0.2187 - accuracy: 0.9155 - 11ms/epoch - 562us/step
Epoch 404/500
20/20 - 0s - loss: 0.2185 - accuracy: 0.9155 - 11ms/epoch - 544us/step
Epoch 405/500
20/20 - 0s - loss: 0.2186 - accuracy: 0.9155 - 11ms/epoch - 548us/step
Epoch 406/500
20/20 - 0s - loss: 0.2181 - accuracy: 0.9155 - 11ms/epoch - 544us/step
Epoch 407/500
20/20 - 0s - loss: 0.2177 - accuracy: 0.9155 - 12ms/epoch - 594us/step
Epoch 408/500
20/20 - 0s - loss: 0.2183 - accuracy: 0.9171 - 11ms/epoch - 570us/step
Epoch 409/500
20/20 - 0s - loss: 0.2174 - accuracy: 0.9155 - 11ms/epoch - 564us/step
Epoch 410/500
20/20 - 0s - loss: 0.2179 - accuracy: 0.9171 - 11ms/epoch - 549us/step
Epoch 411/500
20/20 - 0s - loss: 0.2173 - accuracy: 0.9171 - 11ms/epoch - 562us/step
Epoch 412/500
20/20 - 0s - loss: 0.2170 - accuracy: 0.9155 - 11ms/epoch - 564us/step
Epoch 413/500
20/20 - 0s - loss: 0.2172 - accuracy: 0.9171 - 11ms/epoch - 553us/step
Epoch 414/500
20/20 - 0s - loss: 0.2168 - accuracy: 0.9171 - 11ms/epoch - 549us/step
Epoch 415/500
20/20 - 0s - loss: 0.2167 - accuracy: 0.9171 - 12ms/epoch - 582us/step
Epoch 416/500
20/20 - 0s - loss: 0.2168 - accuracy: 0.9171 - 12ms/epoch - 576us/step
Epoch 417/500
20/20 - 0s - loss: 0.2163 - accuracy: 0.9155 - 11ms/epoch - 564us/step
Epoch 418/500
20/20 - 0s - loss: 0.2165 - accuracy: 0.9171 - 11ms/epoch - 565us/step
Epoch 419/500
20/20 - 0s - loss: 0.2160 - accuracy: 0.9171 - 12ms/epoch - 583us/step
Epoch 420/500
20/20 - 0s - loss: 0.2162 - accuracy: 0.9171 - 12ms/epoch - 590us/step
Epoch 421/500
20/20 - 0s - loss: 0.2155 - accuracy: 0.9171 - 11ms/epoch - 568us/step
Epoch 422/500
20/20 - 0s - loss: 0.2160 - accuracy: 0.9171 - 11ms/epoch - 565us/step
Epoch 423/500
20/20 - 0s - loss: 0.2150 - accuracy: 0.9171 - 12ms/epoch - 600us/step
Epoch 424/500
20/20 - 0s - loss: 0.2154 - accuracy: 0.9171 - 12ms/epoch - 592us/step
Epoch 425/500
20/20 - 0s - loss: 0.2153 - accuracy: 0.9171 - 11ms/epoch - 547us/step
Epoch 426/500
20/20 - 0s - loss: 0.2149 - accuracy: 0.9171 - 12ms/epoch - 612us/step
Epoch 427/500
20/20 - 0s - loss: 0.2147 - accuracy: 0.9171 - 11ms/epoch - 564us/step
Epoch 428/500
20/20 - 0s - loss: 0.2150 - accuracy: 0.9171 - 12ms/epoch - 575us/step
Epoch 429/500
20/20 - 0s - loss: 0.2140 - accuracy: 0.9187 - 11ms/epoch - 543us/step
Epoch 430/500
20/20 - 0s - loss: 0.2147 - accuracy: 0.9171 - 12ms/epoch - 575us/step
Epoch 431/500
20/20 - 0s - loss: 0.2137 - accuracy: 0.9171 - 12ms/epoch - 580us/step
Epoch 432/500
20/20 - 0s - loss: 0.2144 - accuracy: 0.9187 - 11ms/epoch - 571us/step
Epoch 433/500
20/20 - 0s - loss: 0.2137 - accuracy: 0.9187 - 12ms/epoch - 586us/step
Epoch 434/500
20/20 - 0s - loss: 0.2138 - accuracy: 0.9187 - 16ms/epoch - 788us/step
Epoch 435/500
20/20 - 0s - loss: 0.2132 - accuracy: 0.9187 - 14ms/epoch - 683us/step
Epoch 436/500
20/20 - 0s - loss: 0.2137 - accuracy: 0.9187 - 13ms/epoch - 638us/step
Epoch 437/500
20/20 - 0s - loss: 0.2132 - accuracy: 0.9187 - 26ms/epoch - 1ms/step
Epoch 438/500
20/20 - 0s - loss: 0.2130 - accuracy: 0.9187 - 15ms/epoch - 736us/step
Epoch 439/500
20/20 - 0s - loss: 0.2129 - accuracy: 0.9171 - 12ms/epoch - 616us/step
Epoch 440/500
20/20 - 0s - loss: 0.2131 - accuracy: 0.9187 - 12ms/epoch - 589us/step
Epoch 441/500
20/20 - 0s - loss: 0.2126 - accuracy: 0.9187 - 11ms/epoch - 555us/step
Epoch 442/500
20/20 - 0s - loss: 0.2126 - accuracy: 0.9187 - 11ms/epoch - 569us/step
Epoch 443/500
20/20 - 0s - loss: 0.2124 - accuracy: 0.9187 - 14ms/epoch - 716us/step
Epoch 444/500
20/20 - 0s - loss: 0.2124 - accuracy: 0.9187 - 12ms/epoch - 586us/step
Epoch 445/500
20/20 - 0s - loss: 0.2121 - accuracy: 0.9203 - 12ms/epoch - 592us/step
Epoch 446/500
20/20 - 0s - loss: 0.2119 - accuracy: 0.9187 - 12ms/epoch - 586us/step
Epoch 447/500
20/20 - 0s - loss: 0.2117 - accuracy: 0.9187 - 12ms/epoch - 584us/step
Epoch 448/500
20/20 - 0s - loss: 0.2117 - accuracy: 0.9187 - 12ms/epoch - 585us/step
Epoch 449/500
20/20 - 0s - loss: 0.2117 - accuracy: 0.9203 - 12ms/epoch - 594us/step
Epoch 450/500
20/20 - 0s - loss: 0.2112 - accuracy: 0.9187 - 12ms/epoch - 575us/step
Epoch 451/500
20/20 - 0s - loss: 0.2113 - accuracy: 0.9187 - 12ms/epoch - 600us/step
Epoch 452/500
20/20 - 0s - loss: 0.2111 - accuracy: 0.9187 - 12ms/epoch - 594us/step
Epoch 453/500
20/20 - 0s - loss: 0.2114 - accuracy: 0.9203 - 12ms/epoch - 598us/step
Epoch 454/500
20/20 - 0s - loss: 0.2106 - accuracy: 0.9203 - 12ms/epoch - 595us/step
Epoch 455/500
20/20 - 0s - loss: 0.2107 - accuracy: 0.9203 - 12ms/epoch - 595us/step
Epoch 456/500
20/20 - 0s - loss: 0.2111 - accuracy: 0.9187 - 12ms/epoch - 609us/step
Epoch 457/500
20/20 - 0s - loss: 0.2105 - accuracy: 0.9187 - 12ms/epoch - 580us/step
Epoch 458/500
20/20 - 0s - loss: 0.2104 - accuracy: 0.9187 - 12ms/epoch - 598us/step
Epoch 459/500
20/20 - 0s - loss: 0.2103 - accuracy: 0.9187 - 12ms/epoch - 612us/step
Epoch 460/500
20/20 - 0s - loss: 0.2100 - accuracy: 0.9187 - 11ms/epoch - 575us/step
Epoch 461/500
20/20 - 0s - loss: 0.2103 - accuracy: 0.9203 - 12ms/epoch - 607us/step
Epoch 462/500
20/20 - 0s - loss: 0.2096 - accuracy: 0.9203 - 12ms/epoch - 596us/step
Epoch 463/500
20/20 - 0s - loss: 0.2101 - accuracy: 0.9187 - 12ms/epoch - 603us/step
Epoch 464/500
20/20 - 0s - loss: 0.2096 - accuracy: 0.9203 - 12ms/epoch - 591us/step
Epoch 465/500
20/20 - 0s - loss: 0.2096 - accuracy: 0.9203 - 12ms/epoch - 595us/step
Epoch 466/500
20/20 - 0s - loss: 0.2093 - accuracy: 0.9187 - 12ms/epoch - 623us/step
Epoch 467/500
20/20 - 0s - loss: 0.2090 - accuracy: 0.9203 - 13ms/epoch - 648us/step
Epoch 468/500
20/20 - 0s - loss: 0.2093 - accuracy: 0.9203 - 13ms/epoch - 675us/step
Epoch 469/500
20/20 - 0s - loss: 0.2091 - accuracy: 0.9203 - 12ms/epoch - 609us/step
Epoch 470/500
20/20 - 0s - loss: 0.2088 - accuracy: 0.9187 - 12ms/epoch - 603us/step
Epoch 471/500
20/20 - 0s - loss: 0.2087 - accuracy: 0.9203 - 12ms/epoch - 613us/step
Epoch 472/500
20/20 - 0s - loss: 0.2087 - accuracy: 0.9219 - 20ms/epoch - 984us/step
Epoch 473/500
20/20 - 0s - loss: 0.2083 - accuracy: 0.9203 - 13ms/epoch - 634us/step
Epoch 474/500
20/20 - 0s - loss: 0.2086 - accuracy: 0.9203 - 12ms/epoch - 622us/step
Epoch 475/500
20/20 - 0s - loss: 0.2084 - accuracy: 0.9187 - 12ms/epoch - 614us/step
Epoch 476/500
20/20 - 0s - loss: 0.2078 - accuracy: 0.9203 - 12ms/epoch - 603us/step
Epoch 477/500
20/20 - 0s - loss: 0.2085 - accuracy: 0.9203 - 12ms/epoch - 613us/step
Epoch 478/500
20/20 - 0s - loss: 0.2074 - accuracy: 0.9203 - 12ms/epoch - 622us/step
Epoch 479/500
20/20 - 0s - loss: 0.2079 - accuracy: 0.9203 - 12ms/epoch - 620us/step
Epoch 480/500
20/20 - 0s - loss: 0.2078 - accuracy: 0.9203 - 13ms/epoch - 626us/step
Epoch 481/500
20/20 - 0s - loss: 0.2073 - accuracy: 0.9203 - 13ms/epoch - 625us/step
Epoch 482/500
20/20 - 0s - loss: 0.2072 - accuracy: 0.9203 - 13ms/epoch - 630us/step
Epoch 483/500
20/20 - 0s - loss: 0.2074 - accuracy: 0.9203 - 11ms/epoch - 563us/step
Epoch 484/500
20/20 - 0s - loss: 0.2070 - accuracy: 0.9203 - 12ms/epoch - 623us/step
Epoch 485/500
20/20 - 0s - loss: 0.2069 - accuracy: 0.9187 - 11ms/epoch - 556us/step
Epoch 486/500
20/20 - 0s - loss: 0.2069 - accuracy: 0.9203 - 11ms/epoch - 556us/step

Epoch 487/500
20/20 - 0s - loss: 0.2069 - accuracy: 0.9203 - 11ms/epoch - 559us/step
Epoch 488/500
20/20 - 0s - loss: 0.2065 - accuracy: 0.9203 - 11ms/epoch - 552us/step
Epoch 489/500
20/20 - 0s - loss: 0.2065 - accuracy: 0.9219 - 11ms/epoch - 558us/step
Epoch 490/500
20/20 - 0s - loss: 0.2064 - accuracy: 0.9203 - 11ms/epoch - 548us/step
Epoch 491/500
20/20 - 0s - loss: 0.2067 - accuracy: 0.9203 - 11ms/epoch - 555us/step
Epoch 492/500
20/20 - 0s - loss: 0.2057 - accuracy: 0.9203 - 11ms/epoch - 540us/step
Epoch 493/500
20/20 - 0s - loss: 0.2066 - accuracy: 0.9219 - 11ms/epoch - 575us/step
Epoch 494/500
20/20 - 0s - loss: 0.2057 - accuracy: 0.9219 - 11ms/epoch - 561us/step
Epoch 495/500
20/20 - 0s - loss: 0.2061 - accuracy: 0.9187 - 11ms/epoch - 547us/step
Epoch 496/500
20/20 - 0s - loss: 0.2057 - accuracy: 0.9219 - 11ms/epoch - 560us/step
Epoch 497/500
20/20 - 0s - loss: 0.2058 - accuracy: 0.9203 - 11ms/epoch - 556us/step
Epoch 498/500
20/20 - 0s - loss: 0.2052 - accuracy: 0.9203 - 11ms/epoch - 546us/step
Epoch 499/500
20/20 - 0s - loss: 0.2057 - accuracy: 0.9219 - 11ms/epoch - 549us/step
Epoch 500/500
20/20 - 0s - loss: 0.2054 - accuracy: 0.9203 - 11ms/epoch - 541us/step
INFO:tensorflow:Assets written to: ./structured_data_classifier/best_model/assets

训练完成

由于结构化数据的运算速度较快，因此，这里我们尝试了三个模型（上面的图像和文本分类只尝试了一个模型）。如果你在本地运行时，可以将 max_trials 和 epochs 的值设置的更大，进而找到更加合适的模型。

接下来，让我利用找到的最佳模型，对 titanic_eval.csv 文件中的测试数据进行测试： 

print('Accuracy: {accuracy}'.format(
    accuracy=clf.evaluate(x='titanic_eval.csv', y='survived')))

9/9 [==============================] - 0s 708us/step - loss: 0.6394 - accuracy: 0.8182
Accuracy: [0.6393933296203613, 0.8181818127632141]

从结果可以看出，在只尝试了三个模型的情况下，最佳模型的准确率也还不错。当然，我们可以尝试更多的模型，进而得到更好的准确率。而这一过程，我们无需复杂的数据处理，模型搭建等步骤，我们只需要将数据传入 StructuredDataClassifier 工具类中就能实现。这也就是 Auto-Keras 的巨大优势所在。

78.7. 预训练模型#

目前，Auto-Keras 还支持多个预训练模型任务，分别是：目标检测、情绪分析、文本分类、语音生成和语音识别。你可以直接利用这些预训练模型完成快速推理预测。同时，你需要安装 Auto-Keras 支持预训练模型的依赖库 autokeras-pretrained。

由于这些预训练模型都挂在在 Google Cloud 中，国内下载速度极慢。使用这些预训练模型的方法倒是非常简单，如下所示可以完成对 example.jpg 图片中的目标检测任务。 

from autokeras_pretrained import ObjectDetector

detector = ObjectDetector()
detector.predict("./example.jpg", output_file_path="./")

图片描述

更多内容，可以参考 官方示例 内容。

78.8. 优劣分析#

Auto-Keras 使用方法是非常简单的，尤其是当你已经学习过 scikit-learn 和其他深度学习框架之后，学习 Auto-Keras 的使用基本没有任何障碍。简单来讲，先前的建模任务中，你需要从不同模块下导入不同算法，然后针对算法设置超参数，甚至需要使用网格搜索等高阶调参方法。但是在 Auto-Keras 中，你只需要导入 ImageClassifier 或者 TextClassifier 即可。如果你有足够的时间，对于这两个 API 甚至都不需要设置参数。

但核心的问题也出现了，那就是时间。在前面的示例中，为了更快速地得到结果，实验只是尝试了少量的模型。如果你在本地尝试不设置限制的话，Auto-Keras 执行一次的时间极其漫长。重点是，上面的数据集还较为简单，真实环境中使用的数据往往会复杂很多。

所以，通过 Auto-Keras 的实验，你应该能够进一步感受到自动化深度学习带来的优势和劣势。优势在于使用简单，甚至完全不需要了解算法原理。由于其集成了超参数自动优化，数据自动处理，甚至使用起来要方便快捷很多。而劣势在于，需要耗费漫长的搜索时间，几十分钟、几个小时甚至几天。除此之外，自动化深度学习并不是万能的，它也可能出现花费大量时间后找不到理想模型的情况。而在这种情况下，如果由机器学习专家处理，可能会更快地发现数据或者算法的问题。

最后，由于 Auto-Keras 还在发展的初期，框架问题很多，目前只能算作勉强可用。不过，好消息是该团队正在完全重构 Auto-Keras 代码，我们也一同期待后续能拥有更好的体验。

78.9. 总结#

本次实验中，我们学习了 Auto-Keras 自动化深度学习框架的使用。其中，重点介绍了 Auto-Keras 应用于图像和文本分类问题的方法及步骤，熟悉了 API 参数并进行了示例练习。此外，实验的重点是体会自动化深度学习带来的优势和劣势，这一点非常重要。

相关链接

○ 欢迎分享本文链接到你的社交账号、博客、论坛等。更多的外链会增加搜索引擎对本站收录的权重，从而让更多人看到这些内容。

上一章

77. 自动化深度学习综述

下一章

79. 仙人掌航拍照片分类识别