Repository files navigation

This package implements various optimizers with automatic and simultaneousadjustment of the learning rate and the momentum. The AutoOpt package can be usedin a deep learning training instead of the regular optimizers that are availablein the PyTorch framework. The mini-batch flow in a training is shown in the belowfigure.

This package is built and tested in Python 3.6. Createavenv and install the dependencies as follows:

python3 -m venv .envsource .env/bin/activatepip install --upgrade pippip install torch torchvision

Now install the AutoOpt package from its source repository:

pip install [autoopt-path]

Please see the sample code provided in theexamples folder to understandhow this package can be used in training of various ML models.

Please cite theAutoOpt paperif you are using it in a scientific publication.

@inproceedings{9053316,author={T. {Lancewicki} and S. {Kopru}},booktitle={ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},title={Automatic and Simultaneous Adjustment of Learning Rate and Momentum for Stochastic Gradient-based Optimization Methods},year={2020},volume={},number={},pages={3127-3131}}

Copyright 2019 eBay Inc.

Licensed under the Apache License, Version 2.0 (the "License"); you may not use thisfile except in compliance with the License. You may obtain a copy of the License at

https://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under theLicense is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND,either express or implied. See the License for the specific language governing permissionsand limitations under the License.

This software contains code licensed by third parties.See LICENSE.txt.