Inverse Online Learning: Understanding Non-Stationary and Reactionary Policies

Alex J. Chan, Alicia Curth, and Mihaela van der Schaar

International Conference on Learning Representations (ICLR) 2022

Last Updated: 14 February 2022

Code Author: Alex J. Chan ([email protected])

This repo contains a PyTorch example implementation of the inverse online learning algorithm presented in our paper. The code is ready to run on a synthetic data example and should be simple to apply to arbitrary datasets.

This repo is pip installable - clone it, optionally create a virtual env, and install it (this will automatically install dependencies):

git clone https://github.com/XanderJC/inverse-online.git

cd inverse-online

pip install -e .

Example usage:

from iol.models import AdaptiveLinearModel
from iol.data_loading import generate_linear_dataset
from iol.constants import HYPERPARAMS

model = AdaptiveLinearModel(**HYPERPARAMS)

training_data   = generate_linear_dataset(10000, 50, seed=41310)
validation_data = generate_linear_dataset(1000,  50, seed=41311).get_whole_batch()
test_data       = generate_linear_dataset(1000,  50, seed=41312).get_whole_batch()

model.fit(
    training_data,
    batch_size=100,
    epochs=5,
    learning_rate=0.01,
    validation_set=validation_data,
)

print(model.validation(test_data))

This example can be run simply from the shell using:

python iol/demo.py

Citing

If you use this software please cite as follows:

@inproceedings{chan2022inverse,
title={Inverse Online Learning: Understanding Non-Stationary and Reactionary Policies},
author={Alex James Chan and Alicia Curth and Mihaela van der Schaar},
booktitle={International Conference on Learning Representations},
year={2022},
url={https://openreview.net/forum?id=DYypjaRdph2}
}