Python实现FM【tensorflow2.0】

糖逗

1. import tensorflow as tf
   
2. import numpy as np
   
3. from sklearn.preprocessing import LabelEncoder
   
4. from sklearn.model_selection import train_test_split
   
5. 
   
6. #将数据划分为测试集和训练集
   
7. def preprocess(x, y):
   
8.     x = tf.cast(x, dtype = tf.float64)
   
9.     #x = x / tf.reduce_max(x)
   
10.     y = tf.cast(y, dtype = tf.int64)
    
11.     return x, y
    
12. 
    
13. from sklearn.datasets import load_breast_cancer
    
14. from sklearn.model_selection import train_test_split
    
15. data = load_breast_cancer()
    
16. x_train, x_test, y_train, y_test = train_test_split(data.data, data.target, test_size=0.2,
    
17.                                                         random_state = 11, stratify = data.target)
    
18. print(x_train.shape, y_train.shape, x_test.shape, y_test.shape)
    
19. 
    
20. train_db = tf.data.Dataset.from_tensor_slices((np.array(x_train), y_train))
    
21. train_db = train_db.shuffle(123).map(preprocess).batch(20)
    
22. 
    
23. test_db = tf.data.Dataset.from_tensor_slices((np.array(x_test), y_test))
    
24. test_db = test_db.map(preprocess).batch(20)
    
25. 
    
26. sample = next(iter(train_db))
    
27. print('sample:', sample[0].shape, sample[1].shape,
    
28.       tf.reduce_min(sample[0]), tf.reduce_max(sample[0]))

复制代码

1. from tensorflow.keras import layers, optimizers
   
2. from tensorflow import keras
   
3. 
   
4. class FM(keras.Model):
   
5.     def __init__(self, k = 4):
   
6.         super(FM, self).__init__()
   
7.         self.k = k
   
8.         
   
9.     def build(self,input_shape):
   
10.         self.fc = tf.keras.layers.Dense(units = 1,
    
11.                                   bias_regularizer = tf.keras.regularizers.l2(0.01),
    
12.                                   kernel_regularizer = tf.keras.regularizers.l1(0.02))
    
13. 
    
14.         self.v = self.add_weight(shape = (input_shape[-1], self.k),
    
15.                                       initializer = 'glorot_uniform',
    
16.                                       trainable = True)
    
17.         super(FM, self).build(input_shape)
    
18.         
    
19.     def call(self, x, training=True):
    
20.         #[1, dim]@[dim, k] = [1, k]
    
21.         a = tf.pow(tf.matmul(x, self.v), 2)
    
22.         #[1, dim] @[dim, k] = [1, k]
    
23.         b = tf.matmul(tf.pow(x, 2), tf.pow(self.v, 2))
    
24.         #[1, dim] @[dim, 1] + reduce_mean([1, k] - [1, k])
    
25.         linear = self.fc(x)
    
26.         add = tf.keras.layers.Add()([linear, tf.reduce_sum(a - b, axis = 1, keepdims = True)*0.5])
    
27.         return tf.sigmoid(add)
    
28.    
    
29. 
    
30. model = FM()
    
31. model.build((None, 30))
    
32. model.summary()

复制代码

1. def main():
   
2.     model = FM()
   
3.     model.compile(optimizer = keras.optimizers.Adam(1e-3),
   
4.                   loss = tf.keras.losses.binary_crossentropy,
   
5.                   metrics = [tf.keras.metrics.binary_accuracy])
   
6.     model.fit(train_db, epochs = 200, validation_data = test_db)
   
7. 
   
8.     model.evaluate(test_db)
   
9. 
   
10. if __name__ == '__main__':
    
11.     main()

复制代码

注意事项：
1.“super(FM, self).build(input_shape)”必须写，且要写在def的最后一行。

糖逗

糖逗

参考链接：https://blog.csdn.net/ganghaodream/article/details/98964903

糖逗

论文：http://citeseerx.ist.psu.edu/vie ... p=rep1&type=pdf

代码由网络资源整合学习后写成，感谢网络中乐于进行学习分享的大家

诸己h

糖逗 发表于 2021-1-7 11:57
论文：http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.393.8529&rep=rep1&type=pdf

代码由 ...

感谢感谢