https://github.com/ObjectProfile/KerasWrapper

Keras bindings for Pharo
https://github.com/ObjectProfile/KerasWrapper

Last synced: 2 months ago
JSON representation

Keras bindings for Pharo

• Host: GitHub
• URL: https://github.com/ObjectProfile/KerasWrapper
• Owner: ObjectProfile
• License: mit
• Created: 2018-03-16T13:23:45.000Z (over 6 years ago)
• Default Branch: master
• Last Pushed: 2023-12-29T22:41:03.000Z (6 months ago)
• Last Synced: 2024-01-28T00:37:51.396Z (5 months ago)
• Size: 131 KB
• Stars: 9
• Watchers: 8
• Forks: 1
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Lists

• awesome-pharo - Keras Wrapper - Allows to use [Keras](https://keras.io/) functions within Pharo. (Artificial Intelligence and Machine Learning)
• awesome-pharo-ml - Keras bindings - allows to use [Keras](https://keras.io/) functions within Pharo (Deep Learning)

README

        
# KerasWrapper

[Keras](https://keras.io) bindings for Pharo. This binding allows to use Keras (implemented in Python) within Pharo. For the Pharo programmer, the fact that Keras is written in Python is completely transparent.

# Installing KerasWrapper

The following code loads the KerasWrapper in Pharo:

```Smalltalk

Gofer it

    smalltalkhubUser: 'ainfante' project: 'Keras-Wrapper';

    configurationOf: 'KerasWrapper';

    loadDevelopment.

```  

You then needs to install Python, TensorFlow, and Keras. In the help menu, we provide detalled instruction on how to do so.

# Simple Example

Here is the XOR simple example in Pharo

```Smalltalk

inputs := {{0 . 0}.

	{0 . 1}.

	{1 . 0}.

	{1 . 1}}.

expectedOutputs := {0 . 1 . 1 . 0}.

model := KNSequentialModel inputDim: 2.

model add: (KNDenseLayer neurons: 8).

model add: KNTanHActivation new.

model add: (KNDenseLayer neurons: 1).

model add: KNSigmoidActivation new.

model compileLoss: KNBinaryCrossentropyLoss new optimizer: (KNSGDOptimizer new learningRate: 0.1).

fitData := (model

		fit: inputs

		labels: expectedOutputs

		epochs: 200) waitForValue.

```

Evaluating the code produce the following:

![simple xor](img/simpleXOR.png)

The Pharo code given above is inspired from the [pure python version](https://gist.github.com/stewartpark/187895beb89f0a1b3a54)