Python实现FM【tensorflow2.0】

糖逗

import tensorflow as tf
import numpy as np
from sklearn.preprocessing import LabelEncoder
from sklearn.model_selection import train_test_split

#将数据划分为测试集和训练集
def preprocess(x, y):
    x = tf.cast(x, dtype = tf.float64) 
    #x = x / tf.reduce_max(x)
    y = tf.cast(y, dtype = tf.int64)
    return x, y

from sklearn.datasets import load_breast_cancer
from sklearn.model_selection import train_test_split
data = load_breast_cancer()
x_train, x_test, y_train, y_test = train_test_split(data.data, data.target, test_size=0.2,
                                                        random_state = 11, stratify = data.target)
print(x_train.shape, y_train.shape, x_test.shape, y_test.shape)

train_db = tf.data.Dataset.from_tensor_slices((np.array(x_train), y_train))
train_db = train_db.shuffle(123).map(preprocess).batch(20)

test_db = tf.data.Dataset.from_tensor_slices((np.array(x_test), y_test))
test_db = test_db.map(preprocess).batch(20)

sample = next(iter(train_db))
print('sample:', sample[0].shape, sample[1].shape,
      tf.reduce_min(sample[0]), tf.reduce_max(sample[0]))

from tensorflow.keras import layers, optimizers
from tensorflow import keras

class FM(keras.Model):
    def __init__(self, k = 4):
        super(FM, self).__init__()
        self.k = k
        
    def build(self,input_shape):
        self.fc = tf.keras.layers.Dense(units = 1,
                                  bias_regularizer = tf.keras.regularizers.l2(0.01),
                                  kernel_regularizer = tf.keras.regularizers.l1(0.02))

        self.v = self.add_weight(shape = (input_shape[-1], self.k),
                                      initializer = 'glorot_uniform',
                                      trainable = True)
        super(FM, self).build(input_shape)
        
    def call(self, x, training=True):
        #[1, dim]@[dim, k] = [1, k]
        a = tf.pow(tf.matmul(x, self.v), 2)
        #[1, dim] @[dim, k] = [1, k]
        b = tf.matmul(tf.pow(x, 2), tf.pow(self.v, 2))
        #[1, dim] @[dim, 1] + reduce_mean([1, k] - [1, k])
        linear = self.fc(x)
        add = tf.keras.layers.Add()([linear, tf.reduce_sum(a - b, axis = 1, keepdims = True)*0.5])
        return tf.sigmoid(add) 
    

model = FM()
model.build((None, 30))
model.summary()

def main():
    model = FM()
    model.compile(optimizer = keras.optimizers.Adam(1e-3),
                  loss = tf.keras.losses.binary_crossentropy,
                  metrics = [tf.keras.metrics.binary_accuracy])
    model.fit(train_db, epochs = 200, validation_data = test_db)

    model.evaluate(test_db)

if __name__ == '__main__':
    main()

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