torchfuel 0.1.2

Library build on top of pytorch to fuel productivity

torchfuel

Build on top of pytorch to fuel productivity.

Features

• Generic Trainer
• Classification Trainer (with cross-entropy loss)
• MSE Trainer
• Additional utility layers
• Better dataloaders (currently only for image datasets)

Classification Example

import os
import time
from collections import namedtuple

import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
from torchvision import datasets, models, transforms

from torchfuel.data_loaders.image import ImageDataLoader
from torchfuel.trainers.classification import ClassificationTrainer
from torchfuel.transforms.noise import DropPixelNoiser


dl = ImageDataLoader(
    train_data_folder='imgs/train',
    eval_data_folder='imgs/eval',
    pil_transformations=[transforms.RandomHorizontalFlip()]
    tensor_transformations=[DropPixelNoiser()],
    batch_size=64,
    imagenet_format=True,
)

train_dataloader, eval_dataloader, n_classes = dl.prepare()

device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')

model = Model(...).to(device)

optimiser = optim.SGD(model.parameters(), lr=0.01, momentum=0.9)

scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimiser, 'min', patience=20)

trainer = ClassificationTrainer(device, model, optimiser, scheduler)

fitted_model = trainer.fit(epochs, train_dataloader, eval_dataloader)

How to install

Clone repository and run:

pip install .

Optionally (not up to date):

pip install torchfuel