本篇深入分析郭大nffm的代码

TensorFlow计算图

计算图的构建

ones = tf.ones_like(emb_inp_v2)

mask_a = tf.matrix_band_part(ones, 0, -1) # Upper triangular matrix of 0s and 1s

mask_b = tf.matrix_band_part(ones, 0, 0)  # Diagonal matrix of 0s and 1s

mask = tf.cast(mask_a - mask_b, dtype=tf.bool) # Make a bool mask

'''

True - True = 0 and 0 = False

'''

#DNN

dnn_input = tf.boolean_mask(emb_inp_v2, mask) # 把上三角减去对角线的部分取出来

dnn_input = tf.reshape(dnn_input,[tf.shape(emb_inp_v2)[0],hparams.feature_nums*(hparams.feature_nums-1)//2])

input_size=int(dnn_input.shape[-1])

for idx in range(len(hparams.hidden_size)):

    glorot = np.sqrt(2.0 / (input_size + hparams.hidden_size[idx]))

    W = tf.Variable(np.random.normal(loc=0, scale=glorot, size=(input_size, hparams.hidden_size[idx])), dtype=np.float32)

    dnn_input=tf.tensordot(dnn_input,W,[[-1],[0]])

    if hparams.norm is True:

        dnn_input=self.batch_norm_layer(dnn_input,\

                                   self.use_norm,'norm_'+str(idx))

    dnn_input=tf.nn.relu(dnn_input)

    input_size=hparams.hidden_size[idx]

glorot = np.sqrt(2.0 / (hparams.hidden_size[-1] + 1))

W = tf.Variable(np.random.normal(loc=0, scale=glorot, size=(hparams.hidden_size[-1], 1)), dtype=np.float32)

b = tf.Variable(tf.constant(-3.5), dtype=np.float32)

w3=tf.tensordot(dnn_input,W,[[-1],[0]])+b

为什么没有把单项特征算进去？

巴特西

NFFM的原理与代码

TensorFlow计算图

计算图的构建

最新文章

热门文章