apcamargo / Gradient-Centralization-TensorFlow

Instantly improve your training performance of TensorFlow models with just 2 lines of code!

Gradient Centralization TensorFlow

This Python package implements Gradient Centralization in TensorFlow, a simple and effective optimization technique for Deep Neural Networks as suggested by Yong et al. in the paper Gradient Centralization: A New Optimization Technique for Deep Neural Networks. It can both speedup training process and improve the final generalization performance of DNNs.

Installation

Run the following to install:

pip install gradient-centralization-tf

About the Examples

gctf_mnist.ipynb

This notebook shows the the process of using the gradient-centralization-tf Python package to train on the Fashion MNIST dataset availaible from tf.keras.datasets. It further also compares using gctf and performance without using gctf.

gctf_horses_v_humans.ipynb

This notebook shows the the process of using the gradient-centralization-tf Python package to train on the Horses vs Humans dataset by Laurence Moroney. It further also compares using gctf and performance without using gctf.

Usage

gctf.centralized_gradients_for_optimizer

Create a centralized gradients functions for a specified optimizer.

Arguments:

• optimizer: a tf.keras.optimizers.Optimizer object. The optimizer you are using.

Example:

>>> opt = tf.keras.optimizers.Adam(learning_rate=0.1)
>>> opt.get_gradients = gctf.centralized_gradients_for_optimizer(opt)
>>> model.compile(optimizer = opt, ...)

Returns:

A tf.keras.optimizers.Optimizer object.

gctf.get_centralized_gradients

Computes the centralized gradients.

This function is ideally not meant to be used directly unless you are building a custom optimizer, in which case you could point get_gradients to this function. This is a modified version of tf.keras.optimizers.Optimizer.get_gradients.

Arguments:

• optimizer: a tf.keras.optimizers.Optimizer object. The optimizer you are using.
• loss: Scalar tensor to minimize.
• params: List of variables.

Returns:

A gradients tensor.

gctf.optimizers

Pre built updated optimizers implementing GC.

This module is speciially built for testing out GC and in most cases you would be using gctf.centralized_gradients_for_optimizer though this module implements gctf.centralized_gradients_for_optimizer. You can directly use all optimizers with tf.keras.optimizers updated for GC.

Example:

>>> model.compile(optimizer = gctf.optimizers.adam(learning_rate = 0.01), ...)
>>> model.compile(optimizer = gctf.optimizers.rmsprop(learning_rate = 0.01, rho = 0.91), ...)
>>> model.compile(optimizer = gctf.optimizers.sgd(), ...)

Returns:

A tf.keras.optimizers.Optimizer object.

Developing gctf

To install gradient-centralization-tf, along with tools you need to develop and test, run the following in your virtualenv:

git clone https://github.com/Rishit-dagli/Gradient-Centralization-TensorFlow
# or clone your own fork

pip install -e .[dev]

Want to Contribute 🙋‍♂️?

Awesome! If you want to contribute to this project, you're always welcome! See Contributing Guidelines. You can also take a look at open issues for getting more information about current or upcoming tasks.

Want to discuss? 💬

Have any questions, doubts or want to present your opinions, views? You're always welcome. You can start discussions.

License

Copyright 2020 Rishit Dagli

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

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

Unless required by applicable law or agreed to in writing, software
distributed under the License 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 permissions and
limitations under the License.