pytorch_CTR, all models implemented by pytorch

DIN， DIEN, CAN(comming soon) belong to user-behavior sequence model，you should run with sequence_model.py

模型	创新点
FM	将二阶交叉特征考虑进来，提高模型的表达能力；引入隐向量，缓解了数据稀疏带来的参数难训练问题；模型复杂度为线性
FFM	FFM在FM的基础上进一步改进，在模型中引入类别（field）的概念，将同一个field的特征单独进行one-hot
DCN/DCNV2
W&D
DeepFM
DCAP(CIKM2021)
DIN/DIEN	user-behavior sequence model

1th step: pip install -r requirements.txt

2th step: put the dataset to the /torchfm/dataset

3th step: Create a new folder to save the model，please input the command: mkdir chkpt

4th step: if you want train your model, please run the example/main.py or example/sequence_model.py

5th step: If you want see the loss and acc, please input the command: mkdir logs, and then tensorboard --logdir=logs

worth that: I only use the ml-1m dataset for training and testing

m1-1m download link: https://pan.baidu.com/s/12_LdFDwAp7qpbN9Pl4odHw?pwd=1234 提取码：1234

m1-20m download link:：https://pan.baidu.com/s/1FvCB0oZs0kF-vksNzFiczQ?pwd=1234 提取码：1234

Criteo download link：https://pan.baidu.com/s/1DCzyUTqNUZuplcIrIeC71g?pwd=1234 提取码：1234

fixed by https://github.com/zachstarkk/DCAP and https://github.com/rixwew/pytorch-fm