1 import tensorflow as tf
 2 import numpy as np
 3 from tensorflow.keras.layers import Dense, SimpleRNN, Embedding
 4 import matplotlib.pyplot as plt
 5 import os
 6 
 7 
 8 input_word = "abcdefghijklmnopqrstuvwxyz"
 9 w_to_id = {'a':0, 'b':1, 'c':2, 'd':3, 'e':4,
10            'f':5, 'g':6, 'h':7, 'i':8, 'j':9,
11            'k':10, 'l':11, 'm':12, 'n':13, 'o':14, 'p':15,
12            'q':16, 'r':17, 's':18, 't':19, 
13            'u':20, 'v':21, 'w':22, 'x':23, 'y':24, 'z':25}
14 
15 training_set_scaled = [x for x in range(26)]
16 
17 
18 x_train = []
19 y_train = []
20 
21 for i in range(4, 26):
22   x_train.append(training_set_scaled[i-4 : i])
23   y_train.append(training_set_scaled[i])
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Embedding输入要求：[送入样本数， 循环核时间展开步数] ，
34 # 此处整个数据集送入所以送入，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。
35 x_train = np.reshape(x_train, (len(x_train), 4))
36 y_train = np.array(y_train)
37 
38 
39 model = tf.keras.Sequential([
40         Embedding(26, 2),
41         SimpleRNN(10),
42         Dense(26, activation='softmax')
43 ])
44 
45 model.compile(optimizer=tf.keras.optimizers.Adam(0.01),
46               loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
47               metrics=['sparse_categorical_accuracy'])
48 
49 checkpoint_save_path = "./checkpoint/rnn_embedding_epred1.ckpt"
50 
51 if os.path.exists(checkpoint_save_path + '.index'):
52   print('-----------load the model----------')
53   model.load_weights(checkpoint_save_path)
54 
55 cp_callback = tf.keras.callbacks.ModelCheckpoint(
56     filepath=checkpoint_save_path,
57     save_weights_only=True,
58     save_best_only=True,
59     monitor='loss')
60 
61 history = model.fit(x_train, y_train, batch_size=32, epochs=100, callbacks=[cp_callback])
62 
63 model.summary()
64 
65 
66 with open('./weights.txt', 'w') as f:
67   for v in model.trainable_variables:
68     f.write(str(v.name) + '\n')
69     f.write(str(v.shape) + '\n')
70     f.write(str(v.numpy()) + '\n')
71 
72 
73 acc = history.history['sparse_categorical_accuracy']
74 loss = history.history['loss']
75 
76 plt.subplot(1, 2, 1)
77 plt.plot(acc, label='Training Accuracy')
78 plt.title('Training Accuracy')
79 plt.legend()
80 
81 plt.subplot(1, 2, 2)
82 plt.plot(loss, label='Training Loss')
83 plt.title('Training Loss')
84 plt.legend()
85 plt.show()
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 = [w_to_id[a] for a in alphabet1]
92   # 使alphabet符合Embedding输入要求：[送入样本数， 时间展开步数]。
93   # 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，循环核时间展开步数为4。
94   alphabet = np.reshape(alphabet, (1, 4))
95   result = model.predict([alphabet])
96   pred = tf.argmax(result, axis=1)
97   pred =int(pred)
98   tf.print(alphabet1 + '->' + input_word[pred])