tflearn中使用cnn做分类并做预测

数据情况在上一篇博客中已经提到，直接看代码，定义网络结构比较简单:

1. from __future__ import division, print_function, absolute_import
3. import tensorflow as tf
4. import tflearn
5. from tflearn.layers.core import input_data, dropout, fully_connected
6. from tflearn.layers.conv import conv_1d, global_max_pool
7. from tflearn.layers.merge_ops import merge
8. from tflearn.layers.estimator import regression
9. from tflearn.data_utils import to_categorical, pad_sequences
10. from tflearn.datasets import imdb
11. import pickle
12. import numpy as np
14. """
15. 还是加载imdb.pkl数据
16. """
17. train, test, _ = imdb.load_data(path='imdb.pkl', n_words=10000,
18. valid_portion=0.1)
19. trainX, trainY = train
20. testX, testY = test
22. """
23. 转化为固定长度的向量，这里固定长度为100
24. """
25. trainX = pad_sequences(trainX, maxlen=100, value=0.)
26. testX = pad_sequences(testX, maxlen=100, value=0.)
27. """
28. 二值化向量
29. """
30. trainY = to_categorical(trainY, nb_classes=2)
31. testY = to_categorical(testY, nb_classes=2)
33. """
34. 构建卷积神经网络，这里卷积神经网网络为1d卷积
35. """
36. network = input_data(shape=[None, 100], name='input')
37. network = tflearn.embedding(network, input_dim=10000, output_dim=128)
38. branch1 = conv_1d(network, 128, 3, padding='valid', activation='relu', regularizer="L2")
39. branch2 = conv_1d(network, 128, 4, padding='valid', activation='relu', regularizer="L2")
40. branch3 = conv_1d(network, 128, 5, padding='valid', activation='relu', regularizer="L2")
41. network = merge([branch1, branch2, branch3], mode='concat', axis=1)
42. network = tf.expand_dims(network, 2)
43. network = global_max_pool(network)
44. network = dropout(network, 0.5)
45. network = fully_connected(network, 2, activation='softmax')
46. network = regression(network, optimizer='adam', learning_rate=0.001,
47. loss='categorical_crossentropy', name='target')
48. """
49. 训练开始
50. """
51. model = tflearn.DNN(network, tensorboard_verbose=0)
52. model.fit(trainX, trainY, n_epoch = 1, shuffle=True, validation_set=(testX, testY), show_metric=True, batch_size=32)
53. """
54. 模型保存
55. """
56. model.save("cnn.model")
58. """
59. 做测试使用
60. """
61. test=np.linspace(1,101,100).reshape(1,100)
62. print("测试结果：",model.predict(test))

模型训练结果以及模型保存情况:

1. Training Step: 697 | total loss: 0.40838 | time: 79.960s
2. | Adam | epoch: 001 | loss: 0.40838 - acc: 0.8247 -- iter: 22304/22500
3. Training Step: 698 | total loss: 0.39128 | time: 80.112s
4. | Adam | epoch: 001 | loss: 0.39128 - acc: 0.8329 -- iter: 22336/22500
5. Training Step: 699 | total loss: 0.38896 | time: 80.298s
6. | Adam | epoch: 001 | loss: 0.38896 - acc: 0.8402 -- iter: 22368/22500
7. Training Step: 700 | total loss: 0.39468 | time: 80.456s
8. | Adam | epoch: 001 | loss: 0.39468 - acc: 0.8343 -- iter: 22400/22500
9. Training Step: 701 | total loss: 0.39380 | time: 80.640s
10. | Adam | epoch: 001 | loss: 0.39380 - acc: 0.8353 -- iter: 22432/22500
11. Training Step: 702 | total loss: 0.38980 | time: 80.787s
12. | Adam | epoch: 001 | loss: 0.38980 - acc: 0.8392 -- iter: 22464/22500
13. Training Step: 703 | total loss: 0.39020 | time: 80.970s
14. | Adam | epoch: 001 | loss: 0.39020 - acc: 0.8397 -- iter: 22496/22500
15. Training Step: 704 | total loss: 0.38543 | time: 82.891s
16. | Adam | epoch: 001 | loss: 0.38543 - acc: 0.8370 | val_loss: 0.44625 - val_acc: 0.7880 -- iter: 22500/22500
17. --
18. 测试结果： [[ 0.77064246 0.2293576 ]]

加载模型并做预测:

1. import tensorflow as tf
2. import numpy as np
3. import tflearn
4. from tflearn.layers.core import input_data, dropout, fully_connected
5. from tflearn.layers.conv import conv_1d, global_max_pool
6. from tflearn.layers.merge_ops import merge
7. from tflearn.layers.estimator import regression
8. """
9. 跟训练模型的网络结构一样
10. """
11. network = input_data(shape=[None, 100], name='input')
12. network = tflearn.embedding(network, input_dim=10000, output_dim=128)
13. branch1 = conv_1d(network, 128, 3, padding='valid', activation='relu', regularizer="L2")
14. branch2 = conv_1d(network, 128, 4, padding='valid', activation='relu', regularizer="L2")
15. branch3 = conv_1d(network, 128, 5, padding='valid', activation='relu', regularizer="L2")
16. network = merge([branch1, branch2, branch3], mode='concat', axis=1)
17. network = tf.expand_dims(network, 2)
18. network = global_max_pool(network)
19. network = dropout(network, 0.5)
20. network = fully_connected(network, 2, activation='softmax')
21. network = regression(network, optimizer='adam', learning_rate=0.001,
22. loss='categorical_crossentropy', name='target')
23. """
24. 加载模型做预测
25. """
26. model = tflearn.DNN(network)
27. model.load("cnn.model")
28. test=np.linspace(1,101,100).reshape(1,100)
29. # Predict [[ 0.7725634 0.22743654]]
30. prediction = model.predict(test)
31. print("模型预测结果",prediction)

结果：

1. 2017-10-15 19:35:14.940689: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations.
2. 模型预测结果 [[ 0.77064246 0.2293576 ]]
4. Process finished with exit code 0

基于tflearn高阶api怎么做文本分类基本上完成，模型能正确加载也并能做成正确预测，这部署在service上并不是什么难事了，在上一篇博客提到怎么部署成service。