人工智能实践：Tensorflow笔记：代码总结（6）

1.tf中的循环神经网络&＃xff1a;

2.输入一个字母&＃xff0c;预测下一个字母的例子&＃xff1a;

import numpy as np
import tensorflow as tf
from tensorflow.keras.layers import Dense, SimpleRNN
import matplotlib.pyplot as plt
import os
# 输入。
input_word &＃61; "abcde"
w_to_id &＃61; {&＃39;a&＃39;: 0, &＃39;b&＃39;: 1, &＃39;c&＃39;: 2, &＃39;d&＃39;: 3, &＃39;e&＃39;: 4}
id_to_onehot &＃61; {0: [1., 0., 0., 0., 0.], 1: [0., 1., 0., 0., 0.], 2: [0., 0., 1., 0., 0.], 3: [0., 0., 0., 1., 0.],4: [0., 0., 0., 0., 1.]}
x_train &＃61; [id_to_onehot[w_to_id[&＃39;a&＃39;]], id_to_onehot[w_to_id[&＃39;b&＃39;]], id_to_onehot[w_to_id[&＃39;c&＃39;]],id_to_onehot[w_to_id[&＃39;d&＃39;]], id_to_onehot[w_to_id[&＃39;e&＃39;]]]
y_train &＃61; [w_to_id[&＃39;b&＃39;], w_to_id[&＃39;c&＃39;], w_to_id[&＃39;d&＃39;], w_to_id[&＃39;e&＃39;], w_to_id[&＃39;a&＃39;]]
# 打乱&＃xff1a;
np.random.seed(7)
np.random.shuffle(7)
np.random.seed(7)
np.random.shuffle(7)
tf.random.set_seed(7)
# 数据处理
x_train &＃61; np.reshape(x_train, (len(x_train), 1, 5))
y_train &＃61; np.array(y_train)
# 训练
model &＃61; tf.keras.Sequential([SimpleRNN(3),Dense(5, activation&＃61;&＃39;softmax&＃39;)])model.compile(optimizer&＃61;tf.keras.optimizers.Adam(0.01),loss&＃61;tf.keras.losses.SparseCategoricalCrossentropy(from_logits&＃61;False),metrics&＃61;[&＃39;sparse_categorical_accuracy&＃39;])checkpoint_save_path &＃61; "./checkpoint/rnn_onehot_1pre1.ckpt"if os.path.exists(checkpoint_save_path &＃43; &＃39;.index&＃39;):print(&＃39;-------------load the model-----------------&＃39;)model.load_weights(checkpoint_save_path)cp_callback &＃61; tf.keras.callbacks.ModelCheckpoint(filepath&＃61;checkpoint_save_path,save_weights_only&＃61;True,save_best_only&＃61;True,monitor&＃61;&＃39;loss&＃39;) # 由于fit没有给出测试集&＃xff0c;不计算测试集准确率&＃xff0c;根据loss&＃xff0c;保存最优模型history &＃61; model.fit(x_train, y_train, batch_size&＃61;32, epochs&＃61;100, callbacks&＃61;[cp_callback])model.summary()# print(model.trainable_variables)
file &＃61; open(&＃39;./weights.txt&＃39;, &＃39;w&＃39;) # 参数提取
for v in model.trainable_variables:file.write(str(v.name) &＃43; &＃39;\n&＃39;)file.write(str(v.shape) &＃43; &＃39;\n&＃39;)file.write(str(v.numpy()) &＃43; &＃39;\n&＃39;)
file.close()############################################### show ################################################ 显示训练集和验证集的acc和loss曲线
acc &＃61; history.history[&＃39;sparse_categorical_accuracy&＃39;]
loss &＃61; history.history[&＃39;loss&＃39;]plt.subplot(1, 2, 1)
plt.plot(acc, label&＃61;&＃39;Training Accuracy&＃39;)
plt.title(&＃39;Training Accuracy&＃39;)
plt.legend()plt.subplot(1, 2, 2)
plt.plot(loss, label&＃61;&＃39;Training Loss&＃39;)
plt.title(&＃39;Training Loss&＃39;)
plt.legend()
plt.show()############### predict #############preNum &＃61; int(input("input the number of test alphabet:"))
for i in range(preNum):alphabet1 &＃61; input("input test alphabet:")alphabet &＃61; [id_to_onehot[w_to_id[alphabet1]]]# 使alphabet符合SimpleRNN输入要求&＃xff1a;[送入样本数&＃xff0c; 循环核时间展开步数&＃xff0c; 每个时间步输入特征个数]。此处验证效果送入了1个样本&＃xff0c;送入样本数为1&＃xff1b;输入1个字母出结果&＃xff0c;所以循环核时间展开步数为1; 表示为独热码有5个输入特征&＃xff0c;每个时间步输入特征个数为5alphabet &＃61; np.reshape(alphabet, (1, 1, 5))result &＃61; model.predict([alphabet])pred &＃61; tf.argmax(result, axis&＃61;1)pred &＃61; int(pred)tf.print(alphabet1 &＃43; &＃39;->&＃39; &＃43; input_word[pred])

3.输入4个字母&＃xff0c;预测下一个字母的例子&＃xff1a;

注意此时&＃xff1a;循环核时间步展开 的大小为4.因为要输入4个了。

再注意&＃xff1a;4个时间步不同的仅仅是ht。

import numpy as np
import tensorflow as tf
from tensorflow.keras.layers import Dense, SimpleRNN
import matplotlib.pyplot as plt
import osinput_word &＃61; "abcde"
w_to_id &＃61; {&＃39;a&＃39;: 0, &＃39;b&＃39;: 1, &＃39;c&＃39;: 2, &＃39;d&＃39;: 3, &＃39;e&＃39;: 4} # 单词映射到数值id的词典
id_to_onehot &＃61; {0: [1., 0., 0., 0., 0.], 1: [0., 1., 0., 0., 0.], 2: [0., 0., 1., 0., 0.], 3: [0., 0., 0., 1., 0.],4: [0., 0., 0., 0., 1.]} # id编码为one-hotx_train &＃61; [[id_to_onehot[w_to_id[&＃39;a&＃39;]], id_to_onehot[w_to_id[&＃39;b&＃39;]], id_to_onehot[w_to_id[&＃39;c&＃39;]], id_to_onehot[w_to_id[&＃39;d&＃39;]]],[id_to_onehot[w_to_id[&＃39;b&＃39;]], id_to_onehot[w_to_id[&＃39;c&＃39;]], id_to_onehot[w_to_id[&＃39;d&＃39;]], id_to_onehot[w_to_id[&＃39;e&＃39;]]],[id_to_onehot[w_to_id[&＃39;c&＃39;]], id_to_onehot[w_to_id[&＃39;d&＃39;]], id_to_onehot[w_to_id[&＃39;e&＃39;]], id_to_onehot[w_to_id[&＃39;a&＃39;]]],[id_to_onehot[w_to_id[&＃39;d&＃39;]], id_to_onehot[w_to_id[&＃39;e&＃39;]], id_to_onehot[w_to_id[&＃39;a&＃39;]], id_to_onehot[w_to_id[&＃39;b&＃39;]]],[id_to_onehot[w_to_id[&＃39;e&＃39;]], id_to_onehot[w_to_id[&＃39;a&＃39;]], id_to_onehot[w_to_id[&＃39;b&＃39;]], id_to_onehot[w_to_id[&＃39;c&＃39;]]],
]
y_train &＃61; [w_to_id[&＃39;e&＃39;], w_to_id[&＃39;a&＃39;], w_to_id[&＃39;b&＃39;], w_to_id[&＃39;c&＃39;], w_to_id[&＃39;d&＃39;]]np.random.seed(7)
np.random.shuffle(x_train)
np.random.seed(7)
np.random.shuffle(y_train)
tf.random.set_seed(7)# 使x_train符合SimpleRNN输入要求&＃xff1a;[送入样本数&＃xff0c; 循环核时间展开步数&＃xff0c; 每个时间步输入特征个数]。
# 此处整个数据集送入&＃xff0c;送入样本数为len(x_train)&＃xff1b;输入4个字母出结果&＃xff0c;循环核时间展开步数为4; 表示为独热码有5个输入特征&＃xff0c;每个时间步输入特征个数为5
x_train &＃61; np.reshape(x_train, (len(x_train), 4, 5))
y_train &＃61; np.array(y_train)model &＃61; tf.keras.Sequential([SimpleRNN(3),Dense(5, activation&＃61;&＃39;softmax&＃39;)
])model.compile(optimizer&＃61;tf.keras.optimizers.Adam(0.01),loss&＃61;tf.keras.losses.SparseCategoricalCrossentropy(from_logits&＃61;False),metrics&＃61;[&＃39;sparse_categorical_accuracy&＃39;])checkpoint_save_path &＃61; "./checkpoint/rnn_onehot_4pre1.ckpt"if os.path.exists(checkpoint_save_path &＃43; &＃39;.index&＃39;):print(&＃39;-------------load the model-----------------&＃39;)model.load_weights(checkpoint_save_path)cp_callback &＃61; tf.keras.callbacks.ModelCheckpoint(filepath&＃61;checkpoint_save_path,save_weights_only&＃61;True,save_best_only&＃61;True,monitor&＃61;&＃39;loss&＃39;) # 由于fit没有给出测试集&＃xff0c;不计算测试集准确率&＃xff0c;根据loss&＃xff0c;保存最优模型history &＃61; model.fit(x_train, y_train, batch_size&＃61;32, epochs&＃61;100, callbacks&＃61;[cp_callback])model.summary()# print(model.trainable_variables)
file &＃61; open(&＃39;./weights.txt&＃39;, &＃39;w&＃39;) # 参数提取
for v in model.trainable_variables:file.write(str(v.name) &＃43; &＃39;\n&＃39;)file.write(str(v.shape) &＃43; &＃39;\n&＃39;)file.write(str(v.numpy()) &＃43; &＃39;\n&＃39;)
file.close()############################################### show ################################################ 显示训练集和验证集的acc和loss曲线
acc &＃61; history.history[&＃39;sparse_categorical_accuracy&＃39;]
loss &＃61; history.history[&＃39;loss&＃39;]plt.subplot(1, 2, 1)
plt.plot(acc, label&＃61;&＃39;Training Accuracy&＃39;)
plt.title(&＃39;Training Accuracy&＃39;)
plt.legend()plt.subplot(1, 2, 2)
plt.plot(loss, label&＃61;&＃39;Training Loss&＃39;)
plt.title(&＃39;Training Loss&＃39;)
plt.legend()
plt.show()############### predict #############preNum &＃61; int(input("input the number of test alphabet:"))
for i in range(preNum):alphabet1 &＃61; input("input test alphabet:")alphabet &＃61; [id_to_onehot[w_to_id[a]] for a in alphabet1]# 使alphabet符合SimpleRNN输入要求&＃xff1a;[送入样本数&＃xff0c; 循环核时间展开步数&＃xff0c; 每个时间步输入特征个数]。此处验证效果送入了1个样本&＃xff0c;送入样本数为1&＃xff1b;输入4个字母出结果&＃xff0c;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&＃xff0c;每个时间步输入特征个数为5alphabet &＃61; np.reshape(alphabet, (1, 4, 5))result &＃61; model.predict([alphabet])pred &＃61; tf.argmax(result, axis&＃61;1)pred &＃61; int(pred)tf.print(alphabet1 &＃43; &＃39;->&＃39; &＃43; input_word[pred])


4.

5.

6.

7.

8.

9.

10.