WOMBAT Project

Development of an end-to-end (stim-to-action) neural network trained only with an encoder (stim->sensory representation) and a decoder (motor choice representation -> action). The intermediate mapping can be trained with additional brain representations, or with a mapping between sensory and motor representations.

The goal is to create a testbed model trained on real neural data, to test hypotheses ranging from number of neurons needed for computation, plasticity, timing, and the utility of other brain regions in the stim-to-action behavior. Because the model is never explicitly trained on stim-to-action, the internal representations and computations of the model can be compared to real neural data without any model bias.

Model

To install:

$ cd your_path
$ git clone https://github.com/mgonzal1/wombats-malos.git
$ pip install -e wombats

Importing

import wombats
# or to import specific modules
from wombats import metrics
from wombats import models

# if you change the source files, and want to test the results in a console:
from importlib import reload
metrics = reload(metrics)
models = reload(models)

Check the notebooks directory for initial usage