https://github.com/CyberZHG/keras-radam

RAdam implemented in Keras & TensorFlow
https://github.com/CyberZHG/keras-radam

adam keras optimizers radam rectified-adam tensorflow

Last synced: 5 days ago
JSON representation

RAdam implemented in Keras & TensorFlow

• Host: GitHub
• URL: https://github.com/CyberZHG/keras-radam
• Owner: CyberZHG
• License: mit
• Archived: true
• Created: 2019-08-16T08:14:46.000Z (about 5 years ago)
• Default Branch: master
• Last Pushed: 2022-01-22T10:40:58.000Z (almost 3 years ago)
• Last Synced: 2024-10-31T08:05:30.136Z (10 days ago)
• Topics: adam, keras, optimizers, radam, rectified-adam, tensorflow
• Language: Python
• Homepage: https://pypi.org/project/keras-rectified-adam/
• Size: 54.7 KB
• Stars: 325
• Watchers: 15
• Forks: 46
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Keras RAdam

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

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

\[[中文](https://github.com/CyberZHG/keras-radam/blob/master/README.zh-CN.md)|[English](https://github.com/CyberZHG/keras-radam/blob/master/README.md)\]

Unofficial implementation of [RAdam](https://arxiv.org/pdf/1908.03265v1.pdf) in Keras. 

## Install

```bash

pip install keras-rectified-adam

```

## External Link

- [tensorflow/addons:RectifiedAdam](https://github.com/tensorflow/addons/blob/master/tensorflow_addons/optimizers/rectified_adam.py)

## Usage

```python

from tensorflow import keras

import numpy as np

from keras_radam import RAdam

# Build toy model with RAdam optimizer

model = keras.models.Sequential()

model.add(keras.layers.Dense(input_shape=(17,), units=3))

model.compile(RAdam(), loss='mse')

# Generate toy data

x = np.random.standard_normal((4096 * 30, 17))

w = np.random.standard_normal((17, 3))

y = np.dot(x, w)

# Fit

model.fit(x, y, epochs=5)

```

### Use Warmup

```python

from keras_radam import RAdam

RAdam(total_steps=10000, warmup_proportion=0.1, min_lr=1e-5)

```