simpleTorchWrapper

A general framework for PyTorch classification and regression tasks. It is lightweight, easy to use, and currently under development.

Installation

In Google Colab, run the following code to download pyTorchWrapper.

import os
import sys

import os
import sys

try:
    if not os.path.isdir('/content/simpleTorchWrapper'):
        raise FileNotFoundError
except FileNotFoundError:
    print("(◕‿◕✿) Downloading simpleTorchWrapper from GitHub.")
    os.system(f'git clone https://github.com/shuliu2017/simpleTorchWrapper.git')
except Exception as e:
    print(f"(◕‿◕✿) An unexpected error occurred: {e}")

sys.path.append('/content/simpleTorchWrapper')

Install requirements

!pip install -r /content/pyTorchWrapper/requirements.txt

Available Models

efficientNetB2
efficientNetV2S
VIT

Load Modules

# autoreload
# %load_ext autoreload
# %autoreload 2

import simple_torch_wrapper as  stw

Model Training

Example

epochs = 6

stw.pytorch_tools.set_random_seed(seed=0)

model = stw.models.vit_regressor.ViTRegressor()
device = stw.pytorch_tools.get_device()
model = model.to(device)
model = stw.pytorch_tools.enable_multi_gpu(model)

model_name = 'vit_regressor'
loss_fn = torch.nn.MSELoss()
task_type = 'regression'

optimizer = torch.optim.Adam(params=model.parameters(), lr=0.001)
early_stopping = stw.workflow.EarlyStopping(patience=8)
metrics = stw.customized_metrics.regression_metrics
result = stw.workflow.train(model=model,
                                  train_dataloader=train_loader,
                                  validation_dataloader=val_loader,
                                  optimizer=optimizer,
                                  loss_fn=loss_fn,
                                  metrics=metrics,
                                  task_type=task_type,
                                  epochs=epochs,
                                  early_stopping=early_stopping,
                                  save_freq=2,
                                  device=device)

Regression

loss_fn = torch.nn.MSELoss()
task_type = 'regression'
metrics = stw.customized_metrics.regression_metrics # MSE, MAE, R2; evaluated per epoch

Classification

loss_fn = torch.nn.CrossEntropyLoss()
task_type = 'classification'
metrics = stw.customized_metrics.classification_metrics # Accuracy, Recall, Precision, F1; evaluated per epoch

Commonly used optimizers

torch.optim.Adam(params=model.parameters(), lr=0.001)
torch.optim.SGD(params=model.parameters(), lr=0.001)

Model Evaluation

test_model = stw.models.vit_regressor.ViTRegressor().to(device)
stw.pytorch_tools.load_model_state(test_model, target_dir='/content', model_name= f'model_checkpoint.pt')
test_result = stw.workflow.evaluation_step(test_model, test_loader, loss_fn, metrics, task_type, device)

Example Notebooks

regression simple regression on random noise

Team

LYL is an independent research and development team made up of PhDs in computer science, statistics, and physics. We are dedicated to creating innovative applications and conducting cutting-edge research to simplify daily life and make life more enjoyable.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgements

This project is inspired by the code from other open-source projects. We would like to thank the authors of these projects for their contributions:

shuliu2017 / simpleTorchWrapper