A Lighting Pytorch Framework for Recommendation System, Easy-to-use and Easy-to-extend.

Last update: Jan 04, 2023

Overview

Torch-RecHub

A Lighting Pytorch Framework for Recommendation Models, Easy-to-use and Easy-to-extend.

安装

pip install torch-rechub

主要特性

scikit-learn风格易用的API（fit、predict），即插即用
训练过程与模型定义解耦，易拓展，可针对不同类型的模型设置不同的训练机制
使用Pytorch原生Dataset、DataLoader，易修改，自定义数据
高度模块化，支持常见Layer（MLP、FM、FFM、target-attention、self-attention、transformer等），容易调用组装成新模型
支持常见排序模型（WideDeep、DeepFM、DIN、DCN、xDeepFM等）
支持常见召回模型（DSSM、YoutubeDNN、MIND、SARSRec等）
丰富的多任务学习支持
- SharedBottom、ESMM、MMOE、PLE、AITM等模型
- GradNorm、UWL等动态loss加权机制
聚焦更生态化的推荐场景
- 冷启动
- 延迟反馈
- 去偏
支持丰富的训练机制（对比学习、蒸馏学习等）
第三方高性能开源Trainer支持（Pytorch Lighting等）
更多模型正在开发中

快速使用

from torch_rechub.rmodels.ranking import WideDeep, DeepFM, DIN
from torch_rechub.trainers import CTRTrainer
from torch_rechub.basic.utils import DataGenerator

dg = DataGenerator(x, y)
train_dataloader, val_dataloader, test_dataloader = dg.generate_dataloader()

model = DeepFM(deep_features=deep_features, fm_features=fm_features, mlp_params={"dims": [256, 128], "dropout": 0.2, "activation": "relu"})

ctr_trainer = CTRTrainer(model)
ctr_trainer.fit(train_dataloader, val_dataloader)
auc = ctr_trainer.evaluate(ctr_trainer.model, test_dataloader)

多任务学习

from torch_rechub.models.multi_task import SharedBottom, ESMM, MMOE, PLE, AITM
from torch_rechub.trainers import MTLTrainer

model = MMOE(features, task_types, n_expert=3, expert_params={"dims": [64,32,16]}, tower_params_list=[{"dims": [8]}, {"dims": [8]}])

ctr_trainer = MTLTrainer(model)
ctr_trainer.fit(train_dataloader, val_dataloader)
auc = ctr_trainer.evaluate(ctr_trainer.model, test_dataloader)

Note:

所有模型均在大多数论文提及的多个知名公开数据集中测试，达到或者接近论文性能。

使用案例：Examples

每个数据集将会提供

一个使用脚本，包含样本生成、模型训练与测试，并提供一套测评所用参数。

一个预处理脚本，将原始数据进行预处理，转化成csv。

数据格式参考文件（100条）。

全量数据，统一的csv文件，提供高速网盘下载链接和原始数据链接。

初步规划TODO清单

A Lighting Pytorch Framework for Recommendation System, Easy-to-use and Easy-to-extend.

Related tags

Overview

Torch-RecHub

安装

主要特性

快速使用

Owner

Mincai Lai

Prevent `CUDA error: out of memory` in just 1 line of code.

PyTorch implementation of Soft-DTW: a Differentiable Loss Function for Time-Series in CUDA

A python implementation of Deep-Image-Analogy based on pytorch.

Code and description for my BSc Project, September 2021

Hub is a dataset format with a simple API for creating, storing, and collaborating on AI datasets of any size.

A free, multiplatform SDK for real-time facial motion capture using blendshapes, and rigid head pose in 3D space from any RGB camera, photo, or video.

UMPNet: Universal Manipulation Policy Network for Articulated Objects

Official implementation of NPMs: Neural Parametric Models for 3D Deformable Shapes - ICCV 2021

KinectFusion implemented in Python with PyTorch

Neural Scene Flow Fields for Space-Time View Synthesis of Dynamic Scenes

for taichi voxel-challange event

Official Python implementation of the FuzionCoin protocol

Large-scale language modeling tutorials with PyTorch

SHIFT15M: multiobjective large-scale fashion dataset with distributional shifts

The Turing Change Point Detection Benchmark: An Extensive Benchmark Evaluation of Change Point Detection Algorithms on real-world data

Code for the paper "Next Generation Reservoir Computing"

Deep Anomaly Detection with Outlier Exposure (ICLR 2019)

Membership Inference Attack against Graph Neural Networks

We envision models that are pre-trained on a vast range of domain-relevant tasks to become key for molecule property prediction

RM Operation can equivalently convert ResNet to VGG, which is better for pruning; and can help RepVGG perform better when the depth is large.