This is the repo for the paper: Modeling Multi-Relational Connectivity for Personalized Fashion Matching

Requirements

1. OS: Ubuntu 22.04
2. python 3.8.16
3. Supported (tested) CUDA Versions: V11.1
4. python modules: refer to the modules in requirements.txt

Code Structure

1. The entry script for training and evaluation is: main.py
2. The config file is: train_model_config.yaml
3. The script for data preprocess and dataloader: utility.py, dataloader.py
4. The CP_TransMatch model: CP_TransMatch.py
5. Utility codes for evaluation: eval_util.py
6. Utility codes for knowledge graph-relevant functions: eval_util.py
7. The model folder: ./model/
8. The recommendation results in the evaluation are recorded in ./results/
9. The evaluation performance during training is saved in ./performance/
10. The ./model, ./performance, ./results files will be generated automatically when first time runing the codes.

How to Run

1. We prepare the processed iqon_s dataset as an example to illustrate the CP_TransMatch method, which you can find here. Download and unzip the data file, put it into the roo_datapath which you have assign in data_config.yaml. iqon_s is processed based on the dataset provided by https://github.com/SparkJiao/MG-PFCM_outfit_rec related to the paper. Codes for the pre-processing can be found in utility.py, you can also download the original data and try to process it by yourself.

2. Settings in the configure file train_model_config.yaml are all experimental settings mainly for the training process and the model. Before getting started, you need to revise the root_path for saving your experimental rsults. Settings related to dataset is in data_config.yaml, in which you also need to revise the root_datapath into the path of your data. We offer one hyper-parameters feasibly to change with the command line, which is the dataset (-d) for the experiment. 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 main.py -d=iqon_s -g=0
   

4. 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 for specific dataset (say "iqon") in a table format on the terminal screen by running:

   python get_performance.py -d=iqon_s