A comprehensive toolkit package designed to help you accurately predict key metrics such as Click-Through Rates (CTR), Conversion Rates (CVR), uplift, and pricing strategies. Built with state-of-the-art algorithms and user-friendly interfaces, our package streamlines the process of forecasting and decision-making, allowing you to make data-driven choices with confidence. Whether you're looking to optimize your marketing campaigns, boost sales conversions, or fine-tune your pricing model, our package provides the insights you need to succeed in today's competitive market.
You can learn to use the package by referring to the examples in the directory ./example
More solution examples will be released soon~
The following eval matrix has been implemented:
# | Eval Matrix | Explanation | Note |
---|---|---|---|
1 | AUC | Area Under the ROC Curve | For Classification |
2 | Confusion_Matrix | Confusion Matrix is a performance measurement for classification | For Classification |
3 | ACC_F1_score | Accuracy, Macro-F1 and Weighted-F1 | For Classification |
4 | Top_K_Acc | top_k_accuracy_score | For Classification |
5 | Multi_Class_RP | Multi Class precision, recall and F-beta | For Classification |
6 | r2_score | R2_score | For Classification |
7 | MAE | Mean Absolute Error | For Regression |
8 | MSE | Mean Square Error | For Regression |
9 | MAPE | Mean Absolute Percentage Error | For Regression |
10 | tsne | t-distributed stochastic neighbor embedding | For Manifold |
11 | sp_emb | spectral decomposition to the corresponding graph laplacian | For Manifold |
The models currently implemented in recommendation algorithms:
# | Model Name | model | Note |
---|---|---|---|
1 | Wide and Deep | WndModel | Traditional recommendations |
2 | DNN | DNNModel | Traditional recommendations |
3 | DeepFM | DeepFMModel | Traditional recommendations |
4 | Deep and Cross | DCNModel | Traditional recommendations |
5 | NFM | NFMModel | Traditional recommendations |
6 | Tower | TowerModel | Traditional recommendations |
7 | FLEN | FLENModel | Traditional recommendations |
8 | Fibinet | FiBiNetModel | Traditional recommendations |
9 | InterHAt | InterHAtModel | Traditional recommendations |
10 | CAN | CANModel | Traditional recommendations |
11 | MaskNet | MaskNetModel | Traditional recommendations |
12 | ContextNet | ContextNetModel | Traditional recommendations |
13 | EDCN | EDCNModel | Traditional recommendations |
14 | BertSeq | Bert4RecModel | Sequence recommendation |
15 | GRU4Rec | GRU4RecModel | Sequence recommendation |
16 | DIN | DINModel | Sequence recommendation |
17 | DCAP | DCAPModel | Sequence recommendation |
18 | FBAS | FBASModel | Sequence recommendation |
19 | ESMM | ESMMModel | Multi objective recommendation |
20 | MMoE | GeneralMMoEModel | Multi objective recommendation |
21 | Hard Sharing | HardSharingModel | Multi objective recommendation |
22 | Cross Sharing | CrossSharingModel | Multi objective recommendation |
23 | Cross Stitch | CrossStitchModel | Multi objective recommendation |
24 | PLE | PLEModel | Multi objective recommendation |
In the consolidated algorithms, the following Layer networks have been implemented, which can be conveniently called by higher-level models, or users can directly call the Layer layers to assemble their own models.
# | Graph-based Layer | Note |
---|---|---|
1 | HOMOGNNLayer | General GNN layers for Homogeneity Graph (GCNConv, GATConv, SAGEConv, TransformerConv, ARMAConv) |
2 | HETEGNNLayer | General GNN layers for heterogeneous Graph (HGTConv,HANConv) |
# | Layer | Note |
---|---|---|
1 | DNNLayer | DNN Net |
2 | FMLayer | FM Net in DeepFM, NFM |
3 | CrossLayer | Cross Net in Deep and Cross |
4 | CINLayer | CIN Net in XDeepFM |
5 | MultiHeadAttentionLayer | multi head attention in Bert |
6 | SelfAttentionLayer | scaled dot self attention in Bert |
7 | LayerNorm | Layer Normalization in Bert |
8 | PositionWiseFeedForwardLayer | Position wise feed forward in Bert |
9 | TransformerLayer | Transformer(including multi head attention and LayerNorm) in Bert |
10 | TransformerEncoder | Multi-Transformer in Bert |
11 | AutoIntLayer | Similar with TransformerLayer |
12 | FuseLayer | Local Activation Unit in DIN |
13 | SENETLayer | Squeeze and Excitation Layer |
14 | FieldWiseBiInteractionLayer | FM and MF layer in FLEN |
15 | CrossStitchLayer | Cross-stitch Networks for Multi-task Learning |
16 | GeneralMMoELayer | Modeling Task Relationships in Multi-task Learning with Multi-gate Mixture-of-Experts |
17 | Dice | Dice activation function |
18 | PositionEncodingLayer | Positional Encoding Layer in Transformer |
19 | CGCGatingNetworkLayer | task and expert Net in PLE |
20 | BiLinearInteractionLayer | Last feature net in Fibinet |
21 | CoActionLayer | co-action unit layer in CAN |
22 | MaskBlockLayer | MaskBlockLayer in MaskNet |
If you find this code useful in your research, please cite it using the following BibTeX:
@software{
Wang_gbiz_torch_A_comprehensive_2023,
author = {Wang, Haowen},
doi = {10.5281/zenodo.10222799},
month = nov,
title = {{gbiz_torch: A comprehensive toolkit for predicting key metrics in e-commercial fields}},
url = {https://github.com/whw199833/gbiz_torch},
version = {2.0.4},
year = {2023}
}
or following APA:
Wang, H. (2023). gbiz_torch: A comprehensive toolkit for predicting key metrics in e-commercial fields (Version 2.0.4) [Computer software]. https://doi.org/10.5281/zenodo.10222799
If you have some questions or some advice, or want to contribute to this repo, do not hesitate to contact me:
mail: wanghw@zju.edu.cn
wechat: whw199833