import tensorflow as tf
from tensorflow.keras.layers import Layer
from tensorflow.keras.layers import Input, Dense
class FM_layer(Layer):
def __init__(self, k, w_reg, v_reg):
super().__init__()
self.k = k
self.w_reg = w_reg
self.v_reg = v_reg
def build(self, input_shape):
self.w0 = self.add_weight(name='w0', shape=(1,),
initializer=tf.zeros_initializer(),
trainable=True,)
self.w = self.add_weight(name='w', shape=(input_shape[-1], 1),
initializer=tf.random_normal_initializer(),
trainable=True,
regularizer=tf.keras.regularizers.l2(self.w_reg))
self.v = self.add_weight(name='v', shape=(input_shape[-1], self.k),
initializer=tf.random_normal_initializer(),
trainable=True,
regularizer=tf.keras.regularizers.l2(self.v_reg))
def call(self, inputs, **kwargs):
linear_part = tf.matmul(inputs, self.w) + self.w0 #shape:(batchsize, 1)
inter_part1 = tf.pow(tf.matmul(inputs, self.v), 2) #shape:(batchsize, self.k)
inter_part2 = tf.matmul(tf.pow(inputs, 2), tf.pow(self.v, 2)) #shape:(batchsize, self.k)
inter_part = 0.5*tf.reduce_sum(inter_part1 - inter_part2, axis=-1, keepdims=True) #shape:(batchsize, 1)
output = linear_part + inter_part
return output
class Dense_layer(Layer):
def __init__(self, hidden_units, output_dim, activation):
super().__init__()
self.hidden_units = hidden_units
self.output_dim = output_dim
self.activation = activation
self.hidden_layer = [Dense(i, activation=self.activation) for i in self.hidden_units]
self.output_layer = Dense(self.output_dim, activation=None)
def call(self, inputs):
x = inputs
for layer in self.hidden_layer:
x = layer(x)
output = self.output_layer(x)
return output
from layer import FM_layer, Dense_layer
import tensorflow as tf
from tensorflow.keras import Model
from tensorflow.keras.layers import Embedding
class DeepFM(Model):
def __init__(self, feature_columns, k, w_reg, v_reg, hidden_units, output_dim, activation):
super().__init__()
self.dense_feature_columns, self.sparse_feature_columns = feature_columns
# 将类别特征从onehot_dim转换成embed_dim
self.embed_layers = {
'embed_' + str(i): Embedding(feat['feat_onehot_dim'], feat['embed_dim'])
for i, feat in enumerate(self.sparse_feature_columns)
}
self.FM = FM_layer(k, w_reg, v_reg)
self.Dense = Dense_layer(hidden_units, output_dim, activation)
def call(self, inputs):
dense_inputs, sparse_inputs = inputs[:, :13], inputs[:, 13:]
# embedding
sparse_embed = tf.concat([self.embed_layers['embed_{}'.format(i)](sparse_inputs[:, i])
for i in range(sparse_inputs.shape[1])], axis=1)
x = tf.concat([dense_inputs, sparse_embed], axis=-1)
fm_output = self.FM(x)
dense_output = self.Dense(x)
output = tf.nn.sigmoid(0.5*(fm_output + dense_output))
return output
#pandas 是基于NumPy的一种工具，该工具是为解决数据分析任务而创建的。
import pandas as pd
#读取数据集文件
#因为没有表头，若想要把第一行作为数据需要加上header=None
data = pd.read_csv("./credit-a.csv",header=None)
#data.head() #可以输出前5行数据查看
data.iloc[:,-1].value_counts()
————————————————
版权声明：本文为CSDN博主「有梦想的Programmer」的原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接及本声明。
原文链接：https://blog.csdn.net/qq_45703999/article/details/113919122
import tensorflow as tf
#构建顺序模型并为其添加层
model=tf.keras.Sequential(
[
#第二层：Dense隐藏层，第1个隐藏层需要指定输入数据维度为15，单元个数可自定义
tf.keras.layers.Dense(4,input_shape=(15,),activation="relu"),
#第三层：Dense隐藏层，无需指定输入数据维度
tf.keras.layers.Dense(4,activation="relu"),
#第四层：Dense输出层，单元个数为1(因为目标值就是1维),用sigmoid函数计算出最终概率
tf.keras.layers.Dense(1,activation="sigmoid")
]
)
info = model.summary() #可查看当前模型信息
print(info)
# 为指定模型训练用的优化算法和损失函数
model.compile(
optimizer="adam", #利用adam优化算法
loss="binary_crossentropy", #利用binary_crossentropy作损失函数
metrics=["acc"] #显示在运行过程中的正确率情况
)
# 开始训练，指定数据集的input_x、output_y、训练次数epochs=100
his = model.fit(x=x,y=y,epochs=100)
#his 变量记录了训练好的模型的信息
his.history.keys() #history是一个字典数据
import matplotlib.pyplot as plt
# 绘制loss曲线
plt.plot(his.epoch,his.history.get('loss'))
# 绘制acc曲线
plt.plot(his.epoch,his.history.get('acc'))
# 用训练好的模型做一次测试（这里取训练数据集中前10行+中间15列数据作为测试数据）
test = data.iloc[:10,:-1]
model.predict(test)
