Artificial Hippocampal Algorithm in Pytorch

ATTN: This is a work in progress.

Pytorch implementation of AHA! an ‘Artificial Hippocampal Algorithm’ for Episodic Machine Learning by Kowadlo, Rawlinson and Ahmed (2019).

Getting Started

Use Pipenv to install dependencies and create compatible virtual environment. (https://thoughtbot.com/blog/how-to-manage-your-python-projects-with-pipenv)

• Requires Python version >= 3.6

Pretraining weights

To pretrain the visual cortex modules, run the following:

python pretrain.py --model=experiments/pretrain --json=experiments/pretrain/params.json

To enable saving of weights after each epoch add the autosave flag:

--autosave or -a

To load pre-trained weights from model path, include:

--load

Sending metrics to wandb.ai:

--wandb

Display image after each epoch:

--showlast

Animate training(mac only)

--animate

Customize relative paths:

weights:

--model

datafolder:

--data

params.json:

--json

Running

Coming...

Weights should already be trained for the VC module, so to run predictions, do the following:

1. python train.py

Logs

Logs are stored in session.log in the project root dir. They are currently set to overwrite (w) and can be set to append with an a in the set_logs

Built With

• Pytorch - Artificial neural network framework
• Pipenv - Dependency Management

Authors

• Jacob Krajewski - Pytorch implementation

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Acknowledgments

• Thanks to Gideon Kowadlo and David Rawlinson for guiding me through some of the more difficult elements of the ANN.
• Thanks to @ptrblk on the Pytorch forums for walking me through some of the more confusing aspects of Pytorch.