NeuralPlayground 0.0.5

The standardised environment for the hippocampus and entorhinal cortex models

NeuralPlayground: The standardised environment for the hippocampus and entorhinal cortex models.

• 1 Introduction
• 2 Installation
• 3 The Project
• 4 How to Contribute
• 5 Cite
• 6 Licence

1. Introduction

The abstract representation of space has been extensively studied in the hippocampus and entorhinal cortex in part due to the easy monitoring of the task and neural recording. A growing variety of theoretical models have been proposed to capture the rich neural and behavioral phenomena associated with these circuits. However, objective comparison of these theories against each other and against empirical data is difficult.

Although the significance of virtuous interaction between experiments and theory is widely recognized, the tools available to facilitate comparison are limited. Some important challenge we aim to solve are:

1. Lack availability and accessibility of data in a standardized, labeled format.

2. Standard or easy ways to compare model output with empirical data.

3. A repository of models and data sets that are relevant for the field of hippocampus and entorhinal cortex.

To address this gap, we present an open-source standardised software framework - NeuralPlayground - to enable comparison between the hippocampus and entorhinal cortex models. This Python software package offers a reproducible way to compare models against a centralised library of published experimental results, including neural recordings and animal behavior. The framework currently contains implementations of three Agents, as well as three Experiments providing simple interfaces to publicly available neural and behavioral data. It also contains a customizable 2-dimensional Arena (continuous and discrete) able to produce common experimental environments in which the agents can move in and interact with. We note that each module can also be used separately, allowing flexible access to influential models and data sets.

We currently rely on visual comparison of a hand-selected number of outputs of the model with neural recordings as shown in github.com/NeuralPlayground/examples/comparison. In the future, a set of quantitative measures and qualitative measures will be added for systematic comparisons from any Agent, Arena, Experiments. We want to restate that this won’t constitute an objective judgment of the quality of an Agent to replicate the brain mechanism. Instead, this allows an objective and complete comparison to the current evidence in the field, as is typically done in publications.

Altogether, we hope our framework, offers a foundation that the community will build upon, working toward a shared, standardized, open, and reproducible computational understanding of the hippocampus and entorhinal cortex.

Try our short tutorial online in Colab.

2. Installation

Create a conda environment

We advise you to install the package in a virtual environment, to avoid conflicts with other packages. For example, using conda:

conda create --name NPG-env python=3.10
conda activate NPG-env
conda install pip

Pip install

You can use pip get the latest release of NeuralPlayground from PyPI.

# install the latest release
pip install NeuralPlayground

# upgrade to the latest release
pip install -U NeuralPlayground

# install a particular release
pip install NeuralPlayground==0.0.1

Note: if you wish to run our implementation of the Tolman-Eichenbaum machine, there are additional dependencies to install. These can be found in the whittington_2020_README.md file.

Install for development

If you want to contribute to the project, get the latest development version from GitHub, and install it in editable mode, including the "dev" dependencies:

Unix (Linux and macOS)

git clone https://github.com/SainsburyWellcomeCentre/NeuralPlayground/
cd NeuralPlayground
pip install -e .[dev]

Note: if you are using the zsh shell (default on macOS), replace the last command with:

pip install -e '.[dev]'

Windows

git clone https://github.com/SainsburyWellcomeCentre/NeuralPlayground/
cd NeuralPlayground
pip install -e .[dev]

3. Project

Try our package! We are gathering opinions to focus our efforts on improving aspects of the code or adding new features, so if you tell us what you would like to have, we might just implement it 😊. This open-source software was built to be collaborative and lasting. We hope that the framework will be simple enough to be adopted by a great number of neuroscientists, eventually guiding the path to the computational understanding of the HEC mechanisms. We follow reproducible, inclusive, and collaborative project design guidelines. All relevant documents can be found in Documents.

How to run a single module

Each module can be used separately to easily explore and analyze experimental data and better understand any implemented model. Additionally, different Arenas can be initialised with artificial architectures or with data from real-life experiments. We provide examples of module instantiation in the detailed jupyter notebooks found in examples_experiment, examples_arena and examples_agents. Check our Tolman-Eichenbaum Machine Implementation this branch (work in progress), you will also need to install pytorch.

How to run interactions between modules

As shown in the jupyter notebooks Examples_agent, the Agent can interact with an Arena in a standard RL framework. The first step is to initialise an Agent and Arena of your choice. The Agent can be thought of as the animal performing the Experiment and the Arena as the experimental setting where the animal navigates and performs a task.

How to run comparisons

We have a backend module to run simulations and compare the results with experimental data, you can find it in simulation_manager notebook. To plot these results, go to comparison_from_manager and examples_comparison. We show visual comparisons between results from agents running with experimental behavior and results from the real experiment.

4. I-want-to-Contribute

There are many ways to contribute to our project.

1. Implement a hippocampal and entorhinal cortex Agent of your choice.

2. Work on improving the Arena.

3. Add an Experimental data set.

4. Implementing metrics to compare the output of the Agent with the experimental data.

5. Refactor the code to improve the readability and efficiency.

All contributions should be submitted through a pull request that we will later access. Before sending a pull request, make sure you have the following:

1. Checked the Licensing frameworks.

2. Use developing tools installing the dev dependencies. Use:

   pre-commit run --all-files

   to format the code and check for errors.

3. Followed the PEP8 and numpy docstring style convention. More details found in Style Guide.

4. Implemented and ran Test.

5. Comment your work.

All contributions to the repository are acknowledged through the all-contributors bot. Refer to the README.md files found in each of the modules for further details on how to contribute to them.

5. Cite

See Citation for the correct citation of this framework.

6. License

More details about the license can be found at Licence.

Contributors ✨

Thanks goes to these wonderful people (emoji key):

Clementine Domine 🎨 🧑‍🏫 💻 🔣	rodrigcd 🎨 🧑‍🏫 💻 🔣	Luke Hollingsworth 📖 💻	Andrew Saxe 🧑‍🏫	DrCaswellBarry 🧑‍🏫	Niko Sirmpilatze 🚇 🚧 🔧	Adam Tyson 🚧 🚇
rhayman 💻	Devon Jarvis 📖 💻

This project follows the all-contributors specification. Contributions of any kind are welcome!