DeepCTR-Torch

DeepCTR的PyTorch版本。

DeepCTR是一个易于使用、模块化且可扩展的深度学习CTR模型库，内置大量核心组件层，方便用户轻松构建自定义模型。您只需通过model.fit()和model.predict()即可使用任何复杂模型。可通过pip install -U deepctr-torch进行安装。

让我们一起开始吧！(中文介绍)

模型列表

模型	论文
卷积点击预测模型	[CIKM 2015]A Convolutional Click Prediction Model
因子分解支持的神经网络	[ECIR 2016]Deep Learning over Multi-field Categorical Data: A Case Study on User Response Prediction
产品基础神经网络	[ICDM 2016]Product-based neural networks for user response prediction
Wide & Deep	[DLRS 2016]Wide & Deep Learning for Recommender Systems
DeepFM	[IJCAI 2017]DeepFM: A Factorization-Machine based Neural Network for CTR Prediction
分段线性模型	[arxiv 2017]Learning Piece-wise Linear Models from Large Scale Data for Ad Click Prediction
深度与交叉网络	[ADKDD 2017]Deep & Cross Network for Ad Click Predictions
注意力因子分解机	[IJCAI 2017]Attentional Factorization Machines: Learning the Weight of Feature Interactions via Attention Networks
神经因子分解机	[SIGIR 2017]Neural Factorization Machines for Sparse Predictive Analytics
xDeepFM	[KDD 2018]xDeepFM: Combining Explicit and Implicit Feature Interactions for Recommender Systems
深度兴趣网络	[KDD 2018]Deep Interest Network for Click-Through Rate Prediction
深度兴趣进化网络	[AAAI 2019]Deep Interest Evolution Network for Click-Through Rate Prediction
AutoInt	[CIKM 2019]AutoInt: Automatic Feature Interaction Learning via Self-Attentive Neural Networks
ONN	[arxiv 2019]Operation-aware Neural Networks for User Response Prediction
FiBiNET	[RecSys 2019]FiBiNET: Combining Feature Importance and Bilinear feature Interaction for Click-Through Rate Prediction
IFM	[IJCAI 2019]An Input-aware Factorization Machine for Sparse Prediction
DCN V2	[arxiv 2020]DCN V2: Improved Deep & Cross Network and Practical Lessons for Web-scale Learning to Rank Systems
DIFM	[IJCAI 2020]A Dual Input-aware Factorization Machine for CTR Prediction
AFN	[AAAI 2020]Adaptive Factorization Network: Learning Adaptive-Order Feature Interactions
SharedBottom	[arxiv 2017]An Overview of Multi-Task Learning in Deep Neural Networks
ESMM	[SIGIR 2018]Entire Space Multi-Task Model: An Effective Approach for Estimating Post-Click Conversion Rate
MMOE	[KDD 2018]Modeling Task Relationships in Multi-task Learning with Multi-gate Mixture-of-Experts
PLE	[RecSys 2020]Progressive Layered Extraction (PLE): A Novel Multi-Task Learning (MTL) Model for Personalized Recommendations

讨论组与相关项目

• Github Discussions
• 微信讨论群

公众号：浅梦学习笔记	微信：deepctrbot	学习小组 加入 主题集合

• 相关项目

  • AlgoNotes
  • DeepCTR
  • DeepMatch
  • GraphEmbedding

主要贡献者(欢迎加入我们!)

​ ​ 沈伟臣 ​ 阿里巴巴集团 ​	​ ​ 赞书勋 阿里巴巴集团 ​	王泽 ​ 美团 ​	​ 张悟桐 腾讯 ​	​ ​ 张跃峰 北京大学 ​
​ ​ 霍俊毅 南安普顿大学 ​	​ ​ 曾凯 ​ 商汤科技 ​	​ ​ 陈K ​ 网易 ​	​ ​ 程伟宇 ​ 上海交通大学 ​	​ ​ 唐 同济大学 ​

DeepCTR-Torch 快速上手指南

DeepCTR-Torch 是一个基于 PyTorch 的深度学习点击率（CTR）预估模型库。它易于使用、模块化且可扩展，内置了多种主流 CTR 模型（如 DeepFM, DIN, DCN 等）的核心组件，支持通过简单的 fit() 和 predict() 接口完成模型训练与预测。

1. 环境准备

在开始之前，请确保您的开发环境满足以下要求：

• 操作系统: Linux, macOS, Windows
• Python 版本: Python 3.6 - 3.9 (推荐 3.8+)
• 核心依赖:
  • torch (PyTorch)
  • scikit-learn
  • pandas
  • numpy

注意：请先安装好对应 CUDA 版本的 PyTorch。如果尚未安装，可访问 PyTorch 官网 获取安装命令。

2. 安装步骤

推荐使用 pip 进行安装。国内用户建议使用清华或阿里镜像源以加速下载。

标准安装：

pip install -U deepctr-torch

使用国内镜像源加速安装（推荐）：

pip install -U deepctr-torch -i https://pypi.tuna.tsinghua.edu.cn/simple

3. 基本使用

DeepCTR-Torch 的使用流程非常简洁，主要分为数据预处理、定义特征列、构建模型、训练与预测五个步骤。以下是一个基于 DeepFM 模型的最简示例：

步骤 1: 导入依赖与准备数据

import pandas as pd
from sklearn.metrics import log_loss, roc_auc_score
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import LabelEncoder, MinMaxScaler

# 读取数据 (以 Criteo 数据集为例)
data = pd.read_csv('criteo_sample.txt')

sparse_features = ['C' + str(i) for i in range(1, 27)]
dense_features = ['I' + str(i) for i in range(1, 14)]

target = ['label']

# 简单预处理：缺失值填充与编码
data[sparse_features] = data[sparse_features].fillna('-1', )
data[dense_features] = data[dense_features].fillna(0, )

for feat in sparse_features:
    lbe = LabelEncoder()
    data[feat] = lbe.fit_transform(data[feat])

mms = MinMaxScaler(feature_range=(0, 1))
data[dense_features] = mms.fit_transform(data[dense_features])

# 划分训练集和测试集
train, test = train_test_split(data, test_size=0.2)

train_model_input = train[sparse_features + dense_features]
test_model_input = test[sparse_features + dense_features]

步骤 2: 定义特征列

需要明确区分稀疏特征（Sparse Features）和稠密特征（Dense Features）。

from deepctr_torch.inputs import SparseFeat, DenseFeat, get_feature_names

fixlen_feature_columns = [SparseFeat(feat, vocabulary_size=data[feat].max() + 1, embedding_dim=4)
                          for feat in sparse_features] + \
                         [DenseFeat(feat, 1, )
                          for feat in dense_features]

feature_names = get_feature_names(fixlen_feature_columns)

步骤 3: 构建、训练与预测模型

from deepctr_torch.models import DeepFM
from deepctr_torch.callbacks import EarlyStopping, ModelCheckpoint
import torch

# 构建设备
device = 'cuda' if torch.cuda.is_available() else 'cpu'

# 初始化模型
model = DeepFM(linear_feature_columns=fixlen_feature_columns, 
               dnn_feature_columns=fixlen_feature_columns, 
               task='binary', 
               device=device)

# 配置回调函数 (可选)
es = EarlyStopping(monitor='val_auc', min_delta=0, verbose=1, patience=3, mode='max')
mdckpt = ModelCheckpoint(filepath='model.ckpt', monitor='val_auc', verbose=1, save_best_only=True, mode='max')

# 训练模型
model.compile("adam", "binary_crossentropy", metrics=['binary_crossentropy', 'auc'])
model.fit(train_model_input, train[target].values, 
          batch_size=256, epochs=10, verbose=2, 
          validation_data=(test_model_input, test[target].values), 
          callbacks=[es, mdckpt])

# 预测结果
pred_ans = model.predict(test_model_input, batch_size=256)

# 评估结果
print("test LogLoss", round(log_loss(test[target].values, pred_ans), 4))
print("test AUC", round(roc_auc_score(test[target].values, pred_ans), 4))

通过以上步骤，您即可快速运行第一个 DeepCTR 模型。如需尝试其他模型（如 DIN, DCN, xDeepFM 等），只需将 DeepFM 替换为对应的模型类名称，并根据模型特性调整输入参数即可。