Welcome to NeuralLib Documentation

NeuralLib Documentation

Overview

NeuralLib is a Python library designed for advanced biosignal processing using neural networks. The core concept revolves around creating, training, and managing neural network models and leveraging their components for transfer learning (TL). This allows for the reusability of pre-trained models or parts of them to create new models and adapt them to different tasks or datasets efficiently.

The library supports:

Training and testing Architectures from scratch for specific biosignals processing tasks.
Adding tested models to hugging face repositories to create ProductionModels and share them with the community for public usage.
Extracting trained components from production models using TLFactory.
Combining, freezing, or further fine-tuning pre-trained components to trainTLModels.

End users

NeuralLib is designed to support two types of users:

General Users (pip package users)

These users do not need to train models from scratch. Instead, they can install NeuralLib via pip and use pre-trained models available in the model_hub module. This allows them to apply biosignal processing models from NeuralLib collection in hugging face (Collection) without requiring deep knowledge of neural networks.

Install with:
```
pip install NeuralLib
```
Use pre-trained models with minimal setup:
```
from NeuralLib.model_hub import ProductionModel
model = ProductionModel.load_from_huggingface("model-name")
predictions = model.predict(input_signal)
```
Developers & Researchers (GitHub users)

These users need full access to the library's source code for model development, training, and fine-tuning. They can:
- Train new models from scratch using the architectures module
- Test and validate models before sharing them with the community
- Extend NeuralLib with custom architectures and transfer learning workflows
Clone the repository for development:
```
git clone https://github.com/marianaagdias/NeuralLib.git
cd NeuralLib
pip install -e .
```
Train a model from scratch: (for more detailed examples, check the Tutorials folder)
```
from NeuralLib.architectures import GRUseq2seq
model = GRUseq2seq(model_name="test", n_features=10, hid_dim=64, n_layers=2, dropout=0.1, learning_rate=0.001)
model.train_from_scratch(path_x="data/train_x.npy", path_y="data/train_y.npy", batch_size=32, epochs)
```

Functionalities

Two main modules:

Utils and config: utils config