Skip to main content

Brontes is your helping cyclops for pytorch models training.

Project description

Build Status codecov Updates License: MIT PyPI version Codacy Badge

Brontes

Brontes is your helping cyclops for pytorch models training. It is based on pytorch_lightning and comes with an example script in examples/mnist/run.py that you can adapt for your needs.

Additionally, there is an example in examples/mlflow which describes how to use mlflow with Brontes.

Just do this

Define your dataset_loaders as a dictionary: use train, val and optionally a test split:

dataset_loaders = {
    'train':
        torch.utils.data.DataLoader(
            datasets.MNIST(
                root=DATA_PATH,
                train=True,
                download=True
            ),
            batch_size=BATCH_SIZE,
            shuffle=True,
            num_workers=NUMBER_OF_WORKERS
        ),
    'val':
        torch.utils.data.DataLoader(
            datasets.MNIST(
                root=DATA_PATH,
                train=False,
                download=True
            ),
            batch_size=BATCH_SIZE,
            shuffle=True,
            num_workers=NUMBER_OF_WORKERS
        )
}

define your acrhitecture as a torch.nn.Module (or pick an existing architecture):

base_model = brontes.examples.Net()

and wrap it with Brontes:

brontes_model = Brontes(
    model=base_model,
    loss=torch.nn.NLLLoss(),
    data_loaders=dataset_loaders,
    optimizers=optimizer
)

finally train the model using pytorch_lighning

trainer = pl.Trainer(max_nb_epochs=EPOCHS)
trainer.fit(brontes_model)

Development setup

Setup the conda environment

conda env create -f conda.yml

Activate it:

conda activate brontes

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

brontes-0.0.2.tar.gz (4.9 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page