mrdimosthenis / synapses

A group of neural-network libraries for functional and mainstream languages

Synapses

A group of neural-network libraries for functional and mainstream languages!

Choose a programming language:

• Clojure
• C#
• Elixir
• F#
• Gleam
• Java
• JavaScript
• Python
• Scala

Why Synapses?

It's efficient

The implementation is based on lazy list. The information flows smoothly. Everything is obtained at a single pass.

It's customizable

You can specify the activation function and the weight distribution for the neurons of each layer. If this is not enough, edit the json of a network to be exactly what you have in mind.

It offers visualizations

Get an overview of a neural network by taking a brief look at its svg drawing.

Data preprocessing is simple

By annotating the discrete and continuous attributes, you can create a preprocessor that encodes and decodes the datapoints.

Works for huge datasets

The functions that process big volumes of data, have an Iterable/Stream as argument. RAM should not get full!

It's well tested

Every function is tested for every language. Take a look at the test projects.

It's compatible across languages

The interface is similar across languages. You can transfer a network from one platform to another via its json instance. Create a neural network in Python, train it in Java and get its predictions in JavaScript!

synapses's Issues

Package should not contains references to packages related to tests.

Remove dataclasses dependency?

Hey there, congratulations for this really polished project :)

In the pyproject.toml for your python implementation you require the dataclasses library, which is a python 3.6 backport of the core dataclasses library that was introduced in python 3.7 a couple years ago.

Have you considered upgrading to python 3.7+ (maybe even 3.8) and removing this dependency?