https://github.com/cyberzhg/keras-targeted-dropout

Targeted dropout implemented in Keras
https://github.com/cyberzhg/keras-targeted-dropout

dropout keras regularization

Last synced: 7 days ago
JSON representation

Targeted dropout implemented in Keras

• Host: GitHub
• URL: https://github.com/cyberzhg/keras-targeted-dropout
• Owner: CyberZHG
• License: mit
• Archived: true
• Created: 2018-11-26T02:36:00.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2019-05-27T10:20:20.000Z (over 5 years ago)
• Last Synced: 2024-11-15T04:13:26.326Z (3 months ago)
• Topics: dropout, keras, regularization
• Language: Python
• Homepage: https://pypi.org/project/keras-targeted-dropout/
• Size: 15.6 KB
• Stars: 6
• Watchers: 3
• Forks: 4
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Keras Targeted Dropout

[![Travis](https://travis-ci.org/CyberZHG/keras-targeted-dropout.svg)](https://travis-ci.org/CyberZHG/keras-targeted-dropout)

[![Coverage](https://coveralls.io/repos/github/CyberZHG/keras-targeted-dropout/badge.svg?branch=master)](https://coveralls.io/github/CyberZHG/keras-targeted-dropout)

[![Version](https://img.shields.io/pypi/v/keras-targeted-dropout.svg)](https://pypi.org/project/keras-targeted-dropout/)

![Downloads](https://img.shields.io/pypi/dm/keras-targeted-dropout.svg)

![License](https://img.shields.io/pypi/l/keras-targeted-dropout.svg)

Unofficial implementation of [Targeted Dropout](https://openreview.net/pdf?id=HkghWScuoQ) with tensorflow backend.

Note that there is no model compression in this implementation.

## Install

```bash

pip install keras-targeted-dropout

```

## Usage

```python

import keras

from keras_targeted_dropout import TargetedDropout

model = keras.models.Sequential()

model.add(TargetedDropout(

    layer=keras.layers.Dense(units=2, activation='softmax'),

    drop_rate=0.8,

    target_rate=0.2,

    drop_patterns=['kernel'],

    mode=TargetedDropout.MODE_UNIT,

    input_shape=(5,),

))

model.compile(optimizer='adam', loss='mse')

model.summary()

```

* `drop_rate`: Dropout rate for each pixel.

* `target_rate`: The proportion of bottom weights selected as candidates

* `drop_patterns`: A list of names of weights to be dropped.

* `mode`: `TargetedDropout.MODE_UNIT` or `TargetedDropout.MODE_WEIGHT`.

The final dropout rate will be `drop_rate` times `target_rate`.