https://github.com/jjerphan/joml
A minimalist numpy-baked Neural Network API
https://github.com/jjerphan/joml
deep-learning fun minimalist neural-network neural-networks numpy python wheel-implementing
Last synced: about 1 month ago
JSON representation
A minimalist numpy-baked Neural Network API
- Host: GitHub
- URL: https://github.com/jjerphan/joml
- Owner: jjerphan
- License: mit
- Created: 2018-08-27T08:16:41.000Z (almost 8 years ago)
- Default Branch: master
- Last Pushed: 2019-01-31T08:47:33.000Z (over 7 years ago)
- Last Synced: 2025-02-08T09:46:51.068Z (over 1 year ago)
- Topics: deep-learning, fun, minimalist, neural-network, neural-networks, numpy, python, wheel-implementing
- Language: Python
- Homepage:
- Size: 221 KB
- Stars: 0
- Watchers: 3
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
JOML: A minimalist `numpy`-baked Neural Network API 🦎
========================================================
## Getting started
*JOML* API is quite similar to *Keras* [`Sequential Model` API](https://keras.io/getting-started/sequential-model-guide/).
It is for now available on [TestPypi](http://test.pypi.org/).
You can install it using this:
```bash
$ pip install --index-url https://test.pypi.org/project/ joml
```
```python
from joml.network import Network
from joml.layer import Layer, SoftMaxCrossEntropyOutputLayer
import numpy as np
# Loading/transforming data into np.ndarray
x_train, y_train, x_test, y_test = my_loader()
# Here those are arrays of respective shape :
# (14, num_examples) for x_train and x_test
# (4, num_examples) for y_train and y_test
# Defining your network
network = Network(input_size=14, name="My really first network")
network.stack(Layer(size=100))
network.stack(Layer(size=40))
network.output(SoftMaxCrossEntropyOutputLayer(size=4))
# Training
network.train(x_train,y_train)
# … wait (a bit) ⏳
# Profit ! 🚀
y_pred, y_hat, accuracy = network.test(x_test,y_test)
```
See [`examples`](./examples) for some examples.
## Features
The API is not definitive yet : *More to come soon !*
## Why does JOML mean?
JOML means "*JOML One More Layer*".
## License
[This project license](./LICENSE) is MIT.