This program runs the simulations and plots the corresponding panels described in the manuscript 'Unsigned temporal difference errors in cortical L5 dendrites during learning'. This means simulating models of mice with L5 sensory pyramidal neurons learning a go/no-go sensory discrimination task from a sensory representation plagued by distractor signals.
These are instructions to setup a functioning conda environment. If conda is not installed yet, do so following these instructions (linux recommended). Then make sure conda is up-to-date.
conda update conda --all
Create a conda environment with your favorite name, here myenv
conda create -n myenv python=3.10.8
Activate the conda environment
conda activate myenv
Install the required libraries
pip install -r requirements.txt
In order to rerun the full pipeline of simulations yourself and replot all the panels, you will also need a functioning version of Matlab (script was tested with R2022b). The Matlab script will be necessary to extract the performance traces from the simulations. Without Matlab, you will still be able to plot the pre-simulated results used for the paper (and any panel not requiring the performance traces using the results that you simulated yourself).
Before rerunning the simulations, you can test whether the installation was successful. Just enter
python main.py plot
This will plot all the panels (to the folder panels) with the existing simulation results. This is possible as the main branch contains pre-simulated results in the results folder, which are ~1.5GB. If you want to avoid downloading these files and simulate them yourself, you can clone the clean branch instead of the main branch. To rerun all the simulations (except the extraction of the performance trace), enter
python main.py simulate
This will overwrite any simulation outcomes present in the results folder. The results folder will then contain all the simulation results except for the performance traces, which are pre-computed in the folder Smith. In order to extract the performance traces necessary for some panels yourself, you will have to then run Smith/getperfs.m with Matlab. Once the Matlab script is finished you can run
python main.py perf
This should integrate the performances extracted by the code from Smith et al. into the pickle files in the results folder and will allow you to replot all of the panels. If any step is interrupted or skipped, the results files can become corrupted and can only be recovered by rerunning the simulation from the start or downloading the uncorrupted results from the github repository. In case you do not have Matlab installed and nothing was changed in the simulation code, it is also possible to skip the last step. Both the main and clean branch already contain the extracted performance traces for the default simulations. The code will therefore be able to reuse those files.