Biolearn

Unsupervised neural networks with biological-inspired learning rules

• Prerequisites
• Installation
• Usage
• Authors

Prerequisites

The only requirements to successfully install, test, and use the biolearn package are the following:

• numpy
• matplotlib
• scikit-learn
• tqdm
• pytest
• hypothesis

This Python version guarantees a good integration with the other common Machine Learning tools provided by scikit-learn package.
In this way you can use the biolearn as an equivalent alternative in other pipelines.
Like other Machine Learning algorithms also the biolearn ones depend on many hyper-parameters, which have to be tuned according to the given problem.

Installation

To install the biolearn package, first you have to clone the repository:

git clone https://github.com/SimoneGasperini/biolearn.git

Then, install the prerequisites using pip:

cd biolearn
pip install -r requirements.txt

Finally, simply run the setup.py script to complete the installation:

python setup.py install

Usage

The biolearn classes provide the member function fit to train your model on data.
Here, a simple example on MNIST digits dataset is shown.

from sklearn.datasets import fetch_openml
# Download and normalize MNIST dataset
X, y = fetch_openml(name='mnist_784', version=1, data_id=None, return_X_y=True)
X *= 1. / 255
from biolearn.model.bcm import BCM
from biolearn.utils.activations import Relu
from biolearn.utils.optimizer import Adam
# Build and fit the model
model = BCM(outputs=100, num_epochs=20, batch_size=1000, activation=Relu(), optimizer=Adam())
model.fit(X)

You can also visualize the synaptic weights using the utility function provided by the package:

from biolearn.utils.misc import view_weights
view_weights(model.weights, dims=(28, 28))

In the examples directory, some simple Jupyter notebooks are provided as quick tutorials, and the basic features about BCM model training are shown.
Finally, in the online documentation, you can find a general presentation and the mathematical characterization of the models, together with classes API description and more complete examples.

Authors

• Nico Curti git, unibo

• Simone Gasperini git

• Mattia Ceccarelli git, unibo

See also the original complete repository plasticity (by Nico Curti), which provides a Cython wrap and the C++ implementation of the same models.