Skeleton Codes for Deep Learning Projects (Pytorch)

This repo contains skeleton codes of deep learning projects.
It may contain bugs or unreasonable designs, which will be fixed or improved when I find them.
Contain a git precommit config file which is set for black formating (install the precommit use pre-commit install).

Training

Execute train.py directly, using the parameters set in config.py.
You can set multiple groups of parameters in the list of dictionaries in the file.
Before training it will create a folder .\logs(can be changed in config.py), and a subfolder with name expriment_name+timestamp.
The config used in this training process will be stored in the subfolder as file config.txt
The best model will be stored in the subfolder as file model.pth.
The training style is leave-one-out validation and contains an early-stopping mechanism.

Implement the Inference class in ./infer/inference.py
It will load model.pth from the log directory and use the parameters saved in config.txt in the same folder.
Implement ./test.py to test the model

If the model contains a lot of components, add Python files for these components in the directory./model.
If a customized iterator that is different than the test dataloader is needed, please put these files into the directory ./inference.
If some Python scripts are not expected to be committed, rename them with the string utr_ at the beginning, e.g., utr_temp.py.