moeKedama/dg-magnet-test-script

Python

Prerequisites

Windows11/Ubuntu 22.04 LTS
Python 3.10.11
Pytorch 2.0.1
NVIDIA GPU
Anaconda
CUDA 11.7 /CUDA11.8 /CUDA12.1
Recent GPU driver

Preparation

conda create -n torch2 python=3.10
conda activate torch2
conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia 
pip install scikit-learn
pip install matplotlib 
pip install pandas 
pip install seaborn

Test only

Download the dg-magnet-test-script-model, put the model into pth_file directory.

├── pth_file/10domain
│   ├── ViT_DANN_model_for_valid_7601.pth
├── pth_file/1domain
│   ├── 3C90
│   ├── ├── ViT_3C90_4000.pth
│   ├── 3C94
│   ├── 3E6
│   ├── 3F4
│   ├── 77
│   ├── 78
│   ├── N27
│   ├── N30
│   ├── N49
│   ├── N87
├── pth_file/subdomain
│   ├── 3
│   ├── ├── ViT_sub_3_11401.pth
│   ├── 7
│   ├── N

Download the MagNet Challenge Validation Data, put test datasets into valid_data directory.

├── valid_data
│   ├── 3C90
│   ├── ├── B_waveform.csv
│   ├── ├── Frequency.csv
│   ├── ├── H_Waveform.csv
│   ├── ├── Temperature.csv
│   ├── ├── Volumetric_Loss.csv
│   ├── 3C94
│   ├── 3E6
│   ├── 3F4
│   ├── 77
│   ├── 78
│   ├── N27
│   ├── N30
│   ├── N49
│   ├── N87

Modify the function get_args() .
- modify the --dataset in the get_args() (see the commented code).
Modify the norm_dict and pth_PATH in main .
- norm_dict load the normalization coefficients as a dictionary , the saved coefficients are in var_file directory.
- pth_PATH is the path of saved model.
Run , result will be saved at pred_file/.

├── pred_file/10domain
│   ├── pred_3C90.csv
│   ├── pred_3C94.csv
│   ├── pred_3E6.csv
│   ├── pred_3F4.csv
│   ├── pred_77.csv
│   ├── pred_78.csv
│   ├── pred_N27.csv
│   ├── pred_N30.csv
│   ├── pred_N49.csv
│   ├── pred_N87.csv

Train

Coming soon.

FULL VERSION PRETEST RESULTS

Google Drive Link

Acknowledgments