This is the repo for the paper: Leveraging Two Types of Global Graph for Sequential Fashion Recommendation

Requirements

1. OS: Ubuntu 16.04 or higher version
2. python3.7
3. Supported (tested) CUDA Versions: V10.2
4. python modules: refer to the modules in requirements.txt

Code Structure

1. The entry script for training and evaluation is: train.py
2. The config file is: config.yaml
3. The script for data preprocess and dataloader: utility.py
4. The model folder: ./model/.
5. The experimental logs in tensorboard-format are saved in ./logs.
6. The experimental logs in txt-format are saved in ./performance.
7. The best model for each experimental setting is saved in ./model_saves.
8. The recommendation results in the evaluation are recorded in ./results.
9. The ./logs, ./performance, ./model_saves, ./results files will be generated automatically when first time runing the codes.
10. The script get_all_the_res.py is used to print the performance of all the trained and tested models on the screen.

How to Run

1. Download the dataset, decompress it and put it in the top directory with the following command. Note that the downloaded files include two datasets ulilized in the paper: iFashion and amazon_fashion.

   tar zxvf dgsr_dataset.tar.gz. 
   

2. Settings in the configure file config.yaml are basic experimental settings, which are usually fixed in the experiments. To tune other hyper-parameters, you can use command line to pass the parameters. The command line supported hyper-parameters including: the dataset (-d), sequence length (-l) and embedding size (-e). You can also specify which gpu device (-g) to use in the experiments.

3. Run the training and evaluation with the specified hyper-parameters by the command:

   python train.py -d=ifashion -l=5 -e=50 -g=0. 
   

4. During the training, you can monitor the training loss and the evaluation performance by Tensorboard. You can get into ./logs to track the curves of your training and evaluation with the following command:

   tensorboard --host="your host ip" --logdir=./
   

5. The performance of the model is saved in ./performance. You can get into the folder and check the detailed training process of any finished experiments (Compared with the tensorboard log save in ./logs, it is just the txt-version human-readable training log). To quickly check the results for all implemented experiments, you can also print the results of all experiments in a table format on the terminal screen by running:

   python get_all_the_res.py
   

6. The best model will be saved in ./model_saves.