0.0.1-beta5

Adds:

• add the option to use hooks from torch 2.0.0 instead of decorators for TBPTTs classes.
• update the Trainer to add the possibility to pass hidden through dataset. This feature must be improve.

Fixes:

• minor bugs with checkpoint manager
• fix the curbd (RLS with outputs strategy) learning algorithm

0.0.1-beta4

Fixes:

• Minor bugs;

• RLS output strategy according to what is publish in CURBD.

Changes:

• Forward method of WilsonCowanCURBDLayer according to what is publish in CURBD;
• Reindent files from tab to 4 spaces.

Features:

• LinearRNN layer.

0.0.1-beta3

This branch was used to implement the Eligibility propagation (E-Prop) learning algorithm. Here is a list of the different parameters brought this branch:

• Addition of the E-prop learning algorithm
• Bug correction in the sequential model
• Addition of the Spiking layer which is essentially the addition of low pass filter to improve the learning
• The layers.py document has been converted to a file
• Modification in the data visualization for heatmaps (It is now easy to use the same permutation for two heatmaps)
• Fixed bug in the animation of time series with NetworkX
• The LR Schedulers has been modified. The LR Schedulers is now working with multiples parameter groups
• Multiple functions were added to utils.py
• Documentation has been added

0.0.1-beta2

• Fix inplace operation in Lp Loss
• Add colab feature to tutorials

0.0.1-beta1

Adds:

• Reinforcement learning pipeline
• PPO learning algorithm
• Support for torch.nn.Module in Sequential model
• Linear layer

Bug to fix:

• Convergence of PPO

v0.0.1-beta0

Add:

• TBPTT
• RLS (inputs and outputs versions)