facing issues with fearec

[thigazholi]

1. how to choose seq_len in config for fearec?
2. The training job doesn't end on custom interaction data. How to train FEARec sequential model on custom data?

[Yilu114]

FEARec is a model designed for sequential recommendation, with strong modeling capabilities in both the time and frequency domains. If you want to train the FEARec model on custom data, here are some steps and recommendations:

1. Choosing seq_len:

   • seq_len refers to the maximum length of user behavior sequences. In your configuration file, you can set the value of seq_len to fit your dataset. Typically, you can choose an appropriate value based on the average behavior sequence length in your dataset. If your dataset has varying sequence lengths, consider selecting a slightly larger value to avoid truncating longer sequences. However, keep in mind that excessively large seq_len values may lead to high memory consumption, so it's essential to strike a balance.

2. Training the FEARec Model:

   • First, ensure that you have installed RecBole and downloaded the source code.
   • Next, you can use the run_hyper.py script to tune the model's hyperparameters. Here's an example command (modify it according to your dataset and requirements):

     python run_hyper.py --model=FEARec --dataset=your_dataset_name --config_files=your_config_files_path --params_file=hyper.test

     Replace your_dataset_name with your dataset's name and your_config_files_path with the path to your configuration files. You can set the hyperparameter search space in the hyper.test file.

3. Custom Dataset:

   • If your dataset is custom, you'll need to convert it to a format supported by RecBole. Refer to the dataset section in the official RecBole documentation for specific steps.

I hope this information helps!