bamos / HowToTrainYourMAMLPytorch

The original code for the paper "How to train your MAML" along with a replication of the original "Model Agnostic Meta Learning" (MAML) paper in Pytorch.

MAML++ with higher model+optimizer exploration

Here we modify the official MAML++ code to use higher for the model and optimizer so we can ablate across them. We also add hydra for experiment management.

• All of our model definitions are in ./models.py
• The launch command for running the ablation on our cluster is ./launch-all.py
• The model monkey-patching and differentiable optimizer with higher all happens in ./few_shot_learning_system.py
• Our plotting code is available in ./nbs/2019.09.14.plot.ipynb