Learning sophisticated feature interactions behind user behaviors is critical in maximizing CTR for recommender systems. Despite great progress, existing methods seem to have a strong bias towards low- or high-order interactions, or require expertise feature engineering. In this paper, we show that it is possible to derive an end-to-end learning model that emphasizes both low- and high-order feature interactions. The proposed model, DeepFM, combines the power of factorization machines for recommendation and deep learning for feature learning in a new neural network architecture.
[Paper](https://arxiv.org/abs/1703.04247): Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, Xiuqiang He. DeepFM: A Factorization-Machine based Neural Network for CTR Prediction
DeepFM consists of two components. The FM component is a factorization machine, which is proposed in to learn feature interactions for recommendation. The deep component is a feed-forward neural network, which is used to learn high-order feature interactions.
The FM and deep component share the same input raw feature vector, which enables DeepFM to learn low- and high-order feature interactions simultaneously from the input raw features.
- [1] A dataset used in Huifeng Guo, Ruiming Tang, Yunming Ye, Zhenguo Li, Xiuqiang He. DeepFM: A Factorization-Machine based Neural Network for CTR Prediction[J]. 2017.
# [Environment Requirements](#contents)
- Hardware（Ascend/GPU）
- Prepare hardware environment with Ascend or GPU processor. If you want to try Ascend, please send the [application form](https://obs-9be7.obs.cn-east-2.myhuaweicloud.com/file/other/Ascend%20Model%20Zoo%E4%BD%93%E9%AA%8C%E8%B5%84%E6%BA%90%E7%94%B3%E8%AF%B7%E8%A1%A8.docx) to ascend@huawei.com. Once approved, you can get the resources.
Parameters for both training and evaluation can be set in config.py
- train parameters
```
optional arguments:
-h, --help show this help message and exit
--dataset_path DATASET_PATH
Dataset path
--ckpt_path CKPT_PATH
Checkpoint path
--eval_file_name EVAL_FILE_NAME
Auc log file path. Default: "./auc.log"
--loss_file_name LOSS_FILE_NAME
Loss log file path. Default: "./loss.log"
--do_eval DO_EVAL Do evaluation or not. Default: True
--device_target DEVICE_TARGET
Ascend or GPU. Default: Ascend
```
- eval parameters
```
optional arguments:
-h, --help show this help message and exit
--checkpoint_path CHECKPOINT_PATH
Checkpoint file path
--dataset_path DATASET_PATH
Dataset path
--device_target DEVICE_TARGET
Ascend or GPU. Default: Ascend
```
## [Training Process](#contents)
### Training
- running on Ascend
```
python train.py \
--dataset_path='dataset/train' \
--ckpt_path='./checkpoint' \
--eval_file_name='auc.log' \
--loss_file_name='loss.log' \
--device_target='Ascend' \
--do_eval=True > ms_log/output.log 2>&1 &
```
The python command above will run in the background, you can view the results through the file `ms_log/output.log`.
After training, you'll get some checkpoint files under `./checkpoint` folder by default. The loss value are saved in loss.log file.
```
2020-05-27 15:26:29 epoch: 1 step: 41257, loss is 0.498953253030777
2020-05-27 15:32:32 epoch: 2 step: 41257, loss is 0.45545706152915955
...
```
The model checkpoint will be saved in the current directory.
- running on GPU
To do.
### Distributed Training
- running on Ascend
```
sh scripts/run_distribute_train.sh 8 /dataset_path /rank_table_8p.json
```
The above shell script will run distribute training in the background. You can view the results through the file `log[X]/output.log`. The loss value are saved in loss.log file.
- running on GPU
To do.
## [Evaluation Process](#contents)
### Evaluation
- evaluation on dataset when running on Ascend
Before running the command below, please check the checkpoint path used for evaluation.
sh scripts/run_eval.sh 0 Ascend /dataset_path /checkpoint_path/deepfm.ckpt
```
The above python command will run in the background. You can view the results through the file "eval_output.log". The accuracy is saved in auc.log file.
