songdark/deepfm_keras

A simple DeepFM.

DeepFM in Keras

Introduction

A simple DeepFM. See details at here.

Environments

• Keras 2.0.8
• TensorFlow 1.7

Usage

    python DeepFM.py

Reference

DeepFM: A Factorization-Machine based Neural Network for CTR Prediction paper address

Notes

If you meet errors about Embedding Layer, try fix the compute_mask function.

    def compute_mask(self, inputs, mask=None):
        if not self.mask_zero:
            return None
        else:
            # return K.not_equal(inputs, 0)
            mask = K.repeat(K.not_equal(inputs, 0), self.output_dim)
            mask = tf.transpose(mask, [0,2,1])
            return mask

deepfm_keras's Issues

关于多值特征处理的问题

in_interest = Input(shape=[3], name="interest") # None*3, 最长长度3
这里的3是如何确定的呢

能提供测试数据集吗？

MyFlatten这个总是返回none导致错误

看起来在应用的时候mask就是等于None的，这导致返回就是None，程序报错（model部分）。现在我直接用了kares自己的flatten，请问这个是哪里错了么？

    class MyFlatten(Layer):
        def __init__(self, **kwargs):
            self.supports_masking = True
            super(MyFlatten, self).__init__(**kwargs)

    # def compute_mask(self, inputs, mask=None):
    #     if mask==None:
    #         return mask
    # 	return k.batch_flatten(mask)
        def compute_mask(self, inputs, mask=None):
            if mask == None:
                return mask
            else:
                # return K.not_equal(inputs, 0)
                mask = k.repeat(k.not_equal(inputs, 0), self.output_dim)
                mask = tf.transpose(mask, [0, 2, 1])
                return mask
    
        def call(self, inputs, mask=None):
            return k.batch_flatten(inputs)
    
        def compute_output_shape(self, input_shape):
            return (input_shape[0], np.prod(input_shape[1:]))