一、导包
from collections import namedtuple # 使用具名元组 import tensorflow as tf from tensorflow import keras from tensorflow.keras.layers import * from tensorflow.keras.models import * from tqdm import tqdm from sklearn.model_selection import train_test_split from sklearn.preprocessing import MinMaxScaler,LabelEncoder import pandas as pd import numpy as np
二、读取数据
"""读取数据""" data = pd.read_csv('./data/criteo_sample.txt')
三、获取分类特征和数值特征
"""获取分类特征和数值特征""" columns = data.columns.values dense_features = [feat for feat in columns if 'I' in feat] sparse_features = [feat for feat in columns if 'C' in feat]
四、数据处理
"""数据处理""" def data_process(data, dense_features, sparse_features): # 将数值特征的空值位置填补为0 data[dense_features] = data[dense_features].fillna(0.0) # 调整分布 for f in dense_features: data[f] = data[f].apply(lambda x: np.log(x+1) if x > -1 else -1) # 将分类特征进行编码，由于原数据中的类别都是字符串，所以要使用LabelEncoder编码成数值 data[sparse_features]=data[sparse_features].fillna("0") # 将类别特征进行填补，使用字符串 for f in sparse_features: le = LabelEncoder() data[f]=le.fit_transform(data[f]) return data[dense_features + sparse_features]
train_data = data_process(data, dense_features, sparse_features) train_data['label'] = data['label'] train_data # (200,40)
五、使用具名元组为特征做标记
"""使用具名元组为特征做标记""" SparseFeat = namedtuple('SparseFeat', ['name', 'vocabulary_size', 'embedding_dim']) DenseFeat = namedtuple('DenseFeat', ['name', 'dimension']) dnn_features_columns = [SparseFeat(name=feat, vocabulary_size=data[feat].nunique(), embedding_dim = 4) for feat in sparse_features] + [DenseFeat(name=feat, dimension=1) for feat in dense_features] dnn_features_columns
六、构建模型
6.1 构建输入层
"""构建输入层""" def build_input_layers(dnn_features_columns): dense_input_dict, sparse_input_dict = {}, {} for f in dnn_features_columns: if isinstance(f, SparseFeat): sparse_input_dict[f.name] = Input(shape=(1, ), name=f.name) elif isinstance(f, DenseFeat): dense_input_dict[f.name] = Input(shape=(f.dimension, ), name=f.name) return dense_input_dict, sparse_input_dict
6.2 将类别特征进行embedding
"""将类别特征进行embedding""" def build_embedding_layers(dnn_features_columns, input_layers_dict, is_linear): embedding_layer_dict = {} # 将sparse特征筛选出来 sparse_feature_columns = list(filter(lambda x: isinstance(x,SparseFeat), dnn_features_columns)) if dnn_features_columns else [] # 如果是用于线性部分的embedding层，其维度为1，否则维度就是自己定义的embedding维度 if is_linear: for f in sparse_feature_columns: embedding_layer_dict[f.name] = Embedding(f.vocabulary_size + 1, 1, name='1d_emb_' + f.name) else: for f in sparse_feature_columns: embedding_layer_dict[f.name] = Embedding(f.vocabulary_size + 1, f.embedding_dim, name='kd_emb_' + f.name) return embedding_layer_dict
6.3 将所有的sparse特征embedding进行拼接
"""将所有的sparse特征embedding进行拼接""" def concat_embedding_list(dnn_features_columns, input_layer_dict, embedding_layer_dict, flatten=False): # 筛选sparse特征 sparse_feature_columns = list(filter(lambda x: isinstance(x, SparseFeat), dnn_features_columns)) embedding_list = [] for f in sparse_feature_columns: _input = input_layer_dict[f.name] _embed = embedding_layer_dict[f.name] embed = _embed(_input) if flatten: embed = Flatten()(embed) embedding_list.append(embed) return embedding_list
6.4 构建残差块
"""构建残差块""" class ResidualBlock(Layer): def __init__(self, units): super(ResidualBlock, self).__init__() self.units = units def build(self, input_shape): out_dim = input_shape[-1] self.dnn1 = Dense(self.units, activation='relu') self.dnn2 = Dense(out_dim, activation='relu') def call(self, inputs): x = inputs x = self.dnn1(x) x = self.dnn2(x) x = Activation('relu')(x + inputs) return x
6.5 构建输出层
"""构建输出层""" def get_dnn_logits(dnn_inputs, block_nums=3): dnn_out = dnn_inputs for i in range(block_nums): dnn_out = ResidualBlock(64)(dnn_out) dnn_logits = Dense(1, activation='sigmoid')(dnn_out) return dnn_logits
6.6 构建模型
"""构建模型""" def DeepCrossing(dnn_features_columns): # 1.构建输入层 dense_input_dic, sparse_input_dic = build_input_layers(dnn_features_columns) input_layers = list(dense_input_dic.values()) + list(sparse_input_dic.values()) # 2.将类别特征进行embedding embedding_layer_dict = build_embedding_layers(dnn_features_columns, sparse_input_dic, is_linear=False) # 3.将数值型特征拼接在一起 dense_dnn_list = list(dense_input_dic.values()) dense_dnn_inputs = Concatenate(axis=1)(dense_dnn_list) # 4.将类别Embedding向量进行Flatten sparse_dnn_list = concat_embedding_list(dnn_features_columns, sparse_input_dic, embedding_layer_dict, flatten=True) sparse_dnn_inputs = Concatenate(axis=1)(sparse_dnn_list) # 6.将数值特征和类别特征进行拼接 dnn_inputs = Concatenate(axis=1)([dense_dnn_inputs, sparse_dnn_inputs]) # 7.将所有特征输入到残差模块中 output_layer = get_dnn_logits(dnn_inputs, block_nums=3) # 8.构建模型 model = Model(input_layers, output_layer) return model
七、训练模型
7.1 构建模型
history = DeepCrossing(dnn_features_columns) history.summary()
7.2 编译模型
history.compile(optimizer='adam', loss='binary_crossentropy', metrics=['binary_crossentropy', tf.keras.metrics.AUC(name='auc')])
7.3 准备输入数据
train_model_input = {name: data[name] for name in dense_features + sparse_features}
7.4 模型训练
history.fit(train_model_input, train_data['label'].values, batch_size=64, epochs=5, validation_split=0.2)
