sane

Educational implementation of neuroevolutionary SANE-algorithm Designed to Adjust the Weight Parameters of a Single-Layer Feedforward Neural Network. Project was made within course of Neuroevolutionary Computing in Tomsk Polytechnic University.

Used packages

---

For working of algorithm and neural net model:

1. numpy=1.16.4
2. matplotlib=3.1.0
3. scikit-learn=0.23.2

Features

---

• Training single-layer network using SANE-algorithm
• Visualisation of training metrics
• Saving the best models of models
• Logging metric values in the learning process, saving hyperparameter values

SANE-Algorithm

---

SANE can be used to evolve single-layer feedforward neural network that consists of a single hidden layer. Used algorithm can be found in this paper.

Set up

---

1. Install required packages.
2. You can use train.py to train model.