- Please first read the requirement.txt, install the required packages, and read the ArgumentParser for each files.
- In order to run
main.py, you are required to modify the PyTorch source code so that the vision transformer can recieve inputs that have more than 3 channels (use thediffcommand to see the modification ofmodification/modified_vision_transformerandmodification/ori_vision_transformer), where the original path of the source code ishttps://github.com/pytorch/vision/blob/main/torchvision/models/vision_transformer.py.
$ python main.py # 棋力模仿
$ python result_collector.py -t=<path_to_dataset *.csv> -m=<path_to_your_model> -o=<path_to_store_the_result>
$ python train_blender.py # 棋風辨識
$ python blending_collector.py -t=<path_to_dataset *.csv> -m=<path_to_your_model> --output=<path_to_store_the_result>
- baseline_model.py: implements the ResNet for blending
- blender.py: the models and training pipeline of blending
- blending_collector.py: collects the result for submission
- GoDataset: classes and functions for reading data and buildindg Pytorch Dataset
- GoEnv: the game of Go leveraging other low-level api
- gogame.py: originally from https://github.com/aigagror/GymGo (with modification), the low-level api for the game of Go
- GoParser: parse the csv files
- goutils.py: a lots of helper functions, such as formatting the moves, so that differents module can communicate.
- govars.py: global variables
- main.py: run and initialize models for task 1
- result_collector.py: collects the result for submission
- state_utils.py:originally from https://github.com/aigagror/GymGo (with modification), the low-level api to calculate the state of Go
- train_blender.py: start the training of blender.py
- trainer.py: the wrapper for training ViT for task 1.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent