(WIP) Template for new Neural Reckoning research projects

Use this repository to create new projects or just copy/paste bits you need.

Start by reading through Patrick Mineault's Good Research Code Handbook.

Install

• Miniconda. This is a small version of Anaconda with just the basics. Quicker to install than Anaconda and anyway you will make your own environments.
• vscode. I know it's Microsoft but it's just a really good IDE and will save you a lot of headaches.
• Copilot. In vscode, just install the Copilot extension. Use your Imperial College credentials for free access.

Coding setup

Setup Python environment

Read this first.

Edit the environment.yml file giving it a new name specific to your project and adding any needed Python packages.

Then run:

conda env create -f environment.yml

vscode will automatically pick up on this environment, or you can explicitly use it by doing:

conda activate newproject
jupyter notebook
...

Add new packages by editing the environment.yml file and then running:

conda env update --file environment.yml  --prune

Some interesting packages to note in the template environment.yml file are:

• ipykernel: Lets Jupyter notebook detect this environment. I always stick this in every environment I use.
• pip and pip:: These lines let you install some packages that don't have Conda versions using Pip
• ipywidgets: Interactive notebooks with sliders that recompute and regenerate figures, etc.

Directory structure

Here's how I chose to do it for this repo based on Patrick Mineault's recommendation:

|-- scripts            Jupyter notebooks etc. go here
|-- newproject         reusable source code goes here, rename as appropriate
|-- paper              latex source code if you're using that
|-- data               input datasets, etc. excluded from git in .gitignore
|-- results            exclude from git if it's heavy, can include if it's just some light csv files
|-- docs               this folder is aspirational
|-- tests              so is this one
 -- .gitignore         use this to list files that shouldn't be included in the repo, e.g. binaries
 -- environment.yml
 -- README.md
 -- setup.py

Setup installable project package

• Read this.
• Rename newproject directory to something related to your project, but only using lowercase characters and underscores, so it can be an installable package.
• Update setup.py file with this new name and any other metadata.
• Install it locally in an editable way with pip install -e ..

Code templates

TODO

Managing data files

TODO

Plotting publication ready figures

TODO

Training/testing networks

TODO

Spiking neural networks with PyTorch

TODO

Remote development and high performance computing

Use vscode remotely on lab servers or HPC

TODO

Running jobs on HPC

TODO

Paper writing

Using Overleaf is probably in general the easiest way to work together on a paper. You can still use the latex files in the paper/ directory of this repository to initialise your Overleaf document.

You can also install TeX locally (on Windows I recommend MiKTeX). "LaTeX Workshop" is an OK extension to vscode for working with LaTeX that will do things like rebuild each time you save and show the pdf in a panel, like Overleaf does. An advantage of local development is that you can use Copilot to help with the writing.

You can also do both, see Overleaf Git integration.

Read through the paper.tex file and comments for good practices on writing LaTeX docs.

A good way to organise the code is to have a separate supplementary.tex file for Supplmentary Materials because journals often ask for this. You can set up diffs by having a paper-old.tex and diff.tex file (useful for responding to reviews).

For the bibliography, I recommend using Zotero to track papers. You can get BiBTeX by clicking a paper in Zotero and pressing Ctrl-Shift-C to copy into clipboard, then paste that into the .bib file.