Awesome

Argus is a lightweight library for training neural networks in PyTorch.

Documentation

https://pytorch-argus.readthedocs.io

Installation

Requirements:

• torch>=2.0.0

From pip:

pip install pytorch-argus

From source:

pip install -U git+https://github.com/lRomul/argus.git@dev

Example

Simple image classification example with create_model from pytorch-image-models:

from torch.utils.data import DataLoader
from torchvision.datasets import MNIST
from torchvision.transforms import Compose, ToTensor, Normalize

import timm

import argus
from argus.callbacks import MonitorCheckpoint, EarlyStopping, ReduceLROnPlateau


def get_data_loaders(batch_size):
    data_transform = Compose([ToTensor(), Normalize((0.1307,), (0.3081,))])
    train_mnist_dataset = MNIST(download=True, root="mnist_data",
                                transform=data_transform, train=True)
    val_mnist_dataset = MNIST(download=False, root="mnist_data",
                              transform=data_transform, train=False)
    train_loader = DataLoader(train_mnist_dataset,
                              batch_size=batch_size, shuffle=True)
    val_loader = DataLoader(val_mnist_dataset,
                            batch_size=batch_size * 2, shuffle=False)
    return train_loader, val_loader


class TimmModel(argus.Model):
    nn_module = timm.create_model


if __name__ == "__main__":
    train_loader, val_loader = get_data_loaders(batch_size=256)

    params = {
        'nn_module': {
            'model_name': 'tf_efficientnet_b0_ns',
            'pretrained': False,
            'num_classes': 10,
            'in_chans': 1,
            'drop_rate': 0.2,
            'drop_path_rate': 0.2
        },
        'optimizer': ('Adam', {'lr': 0.01}),
        'loss': 'CrossEntropyLoss',
        'device': 'cuda'
    }

    model = TimmModel(params)

    callbacks = [
        MonitorCheckpoint(dir_path='mnist', monitor='val_accuracy', max_saves=3),
        EarlyStopping(monitor='val_accuracy', patience=9),
        ReduceLROnPlateau(monitor='val_accuracy', factor=0.5, patience=3)
    ]

    model.fit(train_loader,
              val_loader=val_loader,
              num_epochs=50,
              metrics=['accuracy'],
              callbacks=callbacks,
              metrics_on_train=True)

More examples you can find here. Additional guides on how to customize and use argus component can be found in Guides section.

Why this name, Argus?

The library name is a reference to a planet from World of Warcraft. Argus is the original homeworld of the eredar (a race of supremely talented magic-wielders), now located within the Twisting Nether. It was once described as a utopian world whose inhabitants were both vastly intelligent and highly gifted in magic. It has since been twisted by demonic, chaotic energies and became the stronghold and homeworld of the Burning Legion.