Skip to content

An Implementation of Model-Agnostic Meta-Learning in PyTorch with Torchmeta

License

Notifications You must be signed in to change notification settings

nmi-lab/pytorch-maml

 
 

Repository files navigation

Model-Agnostic Meta-Learning (METAL Challenge)

run train 'data' --output-folder='logs/run1_kshot1_seq' --dataset='doublenmnistsequence' --use-cuda --batch-size=5 --verbose --meta-lr=1.2e-4 --step-size=0.1 --num-steps=1 --num-workers=8

Original Documentation

Documentation

An implementation of Model-Agnostic Meta-Learning (MAML) in PyTorch with Torchmeta.

Getting started

To avoid any conflict with your existing Python setup, it is suggested to work in a virtual environment with virtualenv. To install virtualenv:

pip install --upgrade virtualenv

Create a virtual environment, activate it and install the requirements in requirements.txt.

virtualenv venv
source venv/bin/activate
pip install -r requirements.txt

Requirements

  • Python 3.6 or above
  • PyTorch 1.5
  • Torchvision 0.6
  • Torchmeta 1.4.6

Usage

You can use train.py to meta-train your model with MAML. For example, to run MAML on Omniglot 1-shot 5-way with default parameters from the original paper:

python train.py /path/to/data --dataset omniglot --num-ways 5 --num-shots 1 --use-cuda --step-size 0.4 --batch-size 32 --num-workers 8 --num-epochs 600 --output-folder /path/to/results

The meta-training script creates a configuration file you can use to meta-test your model. You can use test.py to meta-test your model:

python test.py /path/to/results/config.json

References

The code available in this repository is mainly based on the paper

Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation of deep networks. International Conference on Machine Learning (ICML), 2017 [ArXiv]

If you want to cite this paper

@article{finn17maml,
  author    = {Chelsea Finn and Pieter Abbeel and Sergey Levine},
  title     = {Model-{A}gnostic {M}eta-{L}earning for {F}ast {A}daptation of {D}eep {N}etworks},
  journal   = {International Conference on Machine Learning (ICML)},
  year      = {2017},
  url       = {http://arxiv.org/abs/1703.03400}
}

About

An Implementation of Model-Agnostic Meta-Learning in PyTorch with Torchmeta

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%