1 import numpy as np
2 import tensorflow as tf
3 from tensorflow.keras.layers import Dense, SimpleRNN
4 import matplotlib.pyplot as plt
5 import os
6
7
8 input_word = 'abcde'
9 w_to_id = {'a':0, 'b':1, 'c':2, 'd':3, 'e':4}
10 id_to_onehot = {0:[1., 0., 0., 0., 0.], 1:[0., 1., 0., 0., 0.], 2:[0., 0., 1., 0., 0., ], 3:[0., 0., 0., 1., 0.],
11 4:[0., 0., 0., 0., 1.]}
12
13
14 x_train = [[id_to_onehot[w_to_id['a']], id_to_onehot[w_to_id['b']], id_to_onehot[w_to_id['c']], id_to_onehot[w_to_id['d']]],
15 [id_to_onehot[w_to_id['b']], id_to
16 [id_to_onehot[w_to_id['d']], id__onehot[w_to_id['c']], id_to_onehot[w_to_id['d']], id_to_onehot[w_to_id['e']]],
17 [id_to_onehot[w_to_id['c']], id_to_onehot[w_to_id['d']], id_to_onehot[w_to_id['e']], id_to_onehot[w_to_id['a']]],to_onehot[w_to_id['e']], id_to_onehot[w_to_id['a']], id_to_onehot[w_to_id['b']]],
18 [id_to_onehot[w_to_id['e']], id_to_onehot[w_to_id['a']], id_to_onehot[w_to_id['b']], id_to_onehot[w_to_id['c']]]]
19 y_train = [w_to_id['e'], w_to_id['a'], w_to_id['b'], w_to_id['c'], w_to_id['d']]
20
21
22 print(x_train)
23 print(y_train)
24
25
26 np.random.seed(7)
27 np.random.shuffle(x_train)
28 np.random.seed(7)
29 np.random.shuffle(y_train)
30 tf.random.set_seed(7)
31
32
33 # 使x_train符合SimpleRNN输入要求：[送入样本数， 循环核时间展开步数， 每个时间步输入特征个数]。
34 # 此处整个数据集送入，送入样本数为len(x_train)；输入4个字母出结果，循环核时间展开步数为4; 表示为独热码有5个输入特征，每个时间步输入特征个数为5
35 x_train = np.reshape(x_train, (len(x_train), 4, 5))
36 y_train = np.array(y_train)
37
38
39 model = tf.keras.models.Sequential([
40 SimpleRNN(3),
41 Dense(5, activation='softmax')
42 ])
43
44 model.compile(optimizer=tf.keras.optimizers.Adam(0.01),
45 loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
46 metrics=['sparse_categorical_accuracy'])
47
48 checkpoint_save_path = './checkpoint/rnn_onehot_4pre1.ckpt'
49
50 if os.path.exists(checkpoint_save_path + '.index'):
51 print('-----------load the model-------------------')
52 model.load_weigts(checkpoint_save_path)
53
54 cp_callback = tf.keras.callbacks.ModelCheckpoint(filepath=checkpoint_save_path,
55 save_weights_only=True,
56 save_best_only=True,
57 monitor='loss')
58
59 history = model.fit(x_train, y_train, batch_size=32, epochs=100, callbacks=[cp_callback])
60
61 model.summary()
62
63
64 with open('./weights.txt', 'w') as f:
65 for v in model.trainable_variables:
66 f.write(str(v.name) +'\n')
67 f.write(str(v.shape) + '\n')
68 f.write(str(v.numpy()) + '\n')
69
70
71
72 acc = history.history['sparse_categorical_accuracy']
73 loss = history.history['loss']
74
75 plt.subplot(1, 2, 1)
76 plt.plot(acc, label='Training Accuracy')
77 plt.title('Training Accuracy')
78 plt.legend()
79
80 plt.subplot(1, 2, 2)
81 plt.plot(loss, label='Training Loss')
82 plt.title('Training Loss')
83 plt.legend()
84 plt.show()
85
86
87
88 preNum = int(input("input the number of test alphabet:"))
89 for i in range(preNum):
90 alphabet1 = input("input test alphabet:")
91 alphabet = [id_to_onehot[w_to_id[a]] for a in alphabet1]
92 #使alphabet符合SimpleRNN输入要求：[送入样本数， 循环核时间展开步数，
93 #每个时间步输入特征个数]。此处验证效果送入了1个样本，送入样本数为1；输入4个字母出结果，
94 #所以循环核时间展开步数为4; 表示为独热码有5个输入特征，每个时间步输入特征个数为5
95 alphabet = np.reshape(alphabet, (1, 4, 5))
96 result = model.predict([alphabet])
97 pred = tf.argmax(result, axis=1)
98 pred = int(pred)
99 tf.print(alphabet1 + '->' + input_word[pred])
