Artificial Neural Network (hibou.nn)#

Description#

This module provides some tools to construct an Artificial Neural Network (ANN). Differents configuration of ANN are allowed. An ensemblist vision of ANN is made possible by the class ModelSet.

The default ANN model can be access by default_model and you can tune the hyperparameters via default_model_tuner. The tunable hyperparameters are:

  • the number of Dense layers;

  • the activation function for each layer;

  • the number of units per layer;

  • the learning rate of the Dropout layer;

  • the learning rate of the optimizer (AdamW with amsgrad);

The architecture of the ANN provided by default_model is:

Input layer
─────────────  ┐
Dense layer 1  │
Dense layer 2  │ variable number
...            │ of Dense layers
Dense layer n  │
─────────────  ┘
Flatten
Output layer

Another type of model (gradient boosted trees) is available with gbt_model.

License#

This project is licensed under the GNU Affero General Public Licence v3.0. For more information please see the LICENSE file.

Class#

Functions#