Skip to main content

A lightweight library to help with training neural networks in PyTorch.

Project description

image image imageimage image image
image image image image image
image image image
image image image image
image Twitter facebook numfocus discord
image link

TL;DR

Ignite is a high-level library to help with training and evaluating neural networks in PyTorch flexibly and transparently.

PyTorch-Ignite teaser

Click on the image to see complete code

Features

  • Less code than pure PyTorch while ensuring maximum control and simplicity

  • Library approach and no program's control inversion - Use ignite where and when you need

  • Extensible API for metrics, experiment managers, and other components

Table of Contents

Why Ignite?

Ignite is a library that provides three high-level features:

  • Extremely simple engine and event system
  • Out-of-the-box metrics to easily evaluate models
  • Built-in handlers to compose training pipeline, save artifacts and log parameters and metrics

Simplified training and validation loop

No more coding for/while loops on epochs and iterations. Users instantiate engines and run them.

Example
from ignite.engine import Engine, Events, create_supervised_evaluator
from ignite.metrics import Accuracy


# Setup training engine:
def train_step(engine, batch):
    # Users can do whatever they need on a single iteration
    # Eg. forward/backward pass for any number of models, optimizers, etc
    # ...

trainer = Engine(train_step)

# Setup single model evaluation engine
evaluator = create_supervised_evaluator(model, metrics={"accuracy": Accuracy()})

def validation():
    state = evaluator.run(validation_data_loader)
    # print computed metrics
    print(trainer.state.epoch, state.metrics)

# Run model's validation at the end of each epoch
trainer.add_event_handler(Events.EPOCH_COMPLETED, validation)

# Start the training
trainer.run(training_data_loader, max_epochs=100)

Power of Events & Handlers

The cool thing with handlers is that they offer unparalleled flexibility (compared to, for example, callbacks). Handlers can be any function: e.g. lambda, simple function, class method, etc. Thus, we do not require to inherit from an interface and override its abstract methods which could unnecessarily bulk up your code and its complexity.

Execute any number of functions whenever you wish

Examples
trainer.add_event_handler(Events.STARTED, lambda _: print("Start training"))

# attach handler with args, kwargs
mydata = [1, 2, 3, 4]
logger = ...

def on_training_ended(data):
    print(f"Training is ended. mydata={data}")
    # User can use variables from another scope
    logger.info("Training is ended")


trainer.add_event_handler(Events.COMPLETED, on_training_ended, mydata)
# call any number of functions on a single event
trainer.add_event_handler(Events.COMPLETED, lambda engine: print(engine.state.times))

@trainer.on(Events.ITERATION_COMPLETED)
def log_something(engine):
    print(engine.state.output)

Built-in events filtering

Examples
# run the validation every 5 epochs
@trainer.on(Events.EPOCH_COMPLETED(every=5))
def run_validation():
    # run validation

# change some training variable once on 20th epoch
@trainer.on(Events.EPOCH_STARTED(once=20))
def change_training_variable():
    # ...

# Trigger handler with customly defined frequency
@trainer.on(Events.ITERATION_COMPLETED(event_filter=first_x_iters))
def log_gradients():
    # ...

Stack events to share some actions

Examples

Events can be stacked together to enable multiple calls:

@trainer.on(Events.COMPLETED | Events.EPOCH_COMPLETED(every=10))
def run_validation():
    # ...

Custom events to go beyond standard events

Examples

Custom events related to backward and optimizer step calls:

from ignite.engine import EventEnum


class BackpropEvents(EventEnum):
    BACKWARD_STARTED = 'backward_started'
    BACKWARD_COMPLETED = 'backward_completed'
    OPTIM_STEP_COMPLETED = 'optim_step_completed'

def update(engine, batch):
    # ...
    loss = criterion(y_pred, y)
    engine.fire_event(BackpropEvents.BACKWARD_STARTED)
    loss.backward()
    engine.fire_event(BackpropEvents.BACKWARD_COMPLETED)
    optimizer.step()
    engine.fire_event(BackpropEvents.OPTIM_STEP_COMPLETED)
    # ...

trainer = Engine(update)
trainer.register_events(*BackpropEvents)

@trainer.on(BackpropEvents.BACKWARD_STARTED)
def function_before_backprop(engine):
    # ...

Out-of-the-box metrics

Example
precision = Precision(average=False)
recall = Recall(average=False)
F1_per_class = (precision * recall * 2 / (precision + recall))
F1_mean = F1_per_class.mean()  # torch mean method
F1_mean.attach(engine, "F1")

Installation

From pip:

pip install pytorch-ignite

From conda:

conda install ignite -c pytorch

From source:

pip install git+https://github.com/pytorch/ignite

Nightly releases

From pip:

pip install --pre pytorch-ignite

From conda (this suggests to install pytorch nightly release instead of stable version as dependency):

conda install ignite -c pytorch-nightly

Docker Images

Using pre-built images

Pull a pre-built docker image from our Docker Hub and run it with docker v19.03+.

docker run --gpus all -it -v $PWD:/workspace/project --network=host --shm-size 16G pytorchignite/base:latest /bin/bash
List of available pre-built images

Base

  • pytorchignite/base:latest
  • pytorchignite/apex:latest
  • pytorchignite/hvd-base:latest
  • pytorchignite/hvd-apex:latest
  • pytorchignite/msdp-apex:latest

Vision:

  • pytorchignite/vision:latest
  • pytorchignite/hvd-vision:latest
  • pytorchignite/apex-vision:latest
  • pytorchignite/hvd-apex-vision:latest
  • pytorchignite/msdp-apex-vision:latest

NLP:

  • pytorchignite/nlp:latest
  • pytorchignite/hvd-nlp:latest
  • pytorchignite/apex-nlp:latest
  • pytorchignite/hvd-apex-nlp:latest
  • pytorchignite/msdp-apex-nlp:latest

For more details, see here.

Getting Started

Few pointers to get you started:

Documentation

Additional Materials

Examples

Tutorials

Reproducible Training Examples

Inspired by torchvision/references, we provide several reproducible baselines for vision tasks:

  • ImageNet - logs on Ignite Trains server coming soon ...
  • Pascal VOC2012 - logs on Ignite Trains server coming soon ...

Features:

Code-Generator application

The easiest way to create your training scripts with PyTorch-Ignite:

Communication

User feedback

We have created a form for "user feedback". We appreciate any type of feedback, and this is how we would like to see our community:

  • If you like the project and want to say thanks, this the right place.
  • If you do not like something, please, share it with us, and we can see how to improve it.

Thank you!

Contributing

Please see the contribution guidelines for more information.

As always, PRs are welcome :)

Projects using Ignite

Research papers
Blog articles, tutorials, books
Toolkits
Others

See other projects at "Used by"

If your project implements a paper, represents other use-cases not covered in our official tutorials, Kaggle competition's code, or just your code presents interesting results and uses Ignite. We would like to add your project to this list, so please send a PR with brief description of the project.

Citing Ignite

If you use PyTorch-Ignite in a scientific publication, we would appreciate citations to our project.

@misc{pytorch-ignite,
  author = {V. Fomin and J. Anmol and S. Desroziers and J. Kriss and A. Tejani},
  title = {High-level library to help with training neural networks in PyTorch},
  year = {2020},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/pytorch/ignite}},
}

About the team & Disclaimer

PyTorch-Ignite is a NumFOCUS Affiliated Project, operated and maintained by volunteers in the PyTorch community in their capacities as individuals (and not as representatives of their employers). See the "About us" page for a list of core contributors. For usage questions and issues, please see the various channels here. For all other questions and inquiries, please send an email to contact@pytorch-ignite.ai.

Project details


Release history Release notifications | RSS feed

This version

0.4.9