spatial-frequency-model

Model from the spatial frequency preferences paper. This repo has two components: - model code: if you want to play around the model without the full repo for the paper. - webapp: for exploring how the model's predictions respond to changes in parameter values.

If you use this model in an academic publication, please cite the paper and the Zenodo doi for the Github repo with code associated with the paper.

Model code

To use the model code present here, you need to first set up the python environment. The following will walk you through it:

• Install miniconda on your system for python 3.7.
• Navigate to this directory and run the following:

sh conda create -n sfm python==3.7 conda activate sfm pip install -e .

Now, when you open python, you can access the model by importing this library and making use of it:

python import sfm model = sfm.model.LogGaussianDonut()

See the example notebook in the paper repo for how to use the model: Binder, github repo. Also see the readme for how to use the parameter values presented in the paper.

Webapp

~~This webapp is live on my website, but you can also do the following to run it locally.~~ As of May 2024, I'm no longer hosting this on my website, but you can still do the following to run it locally.

Screenshots

The webapp allows you to change the parameters of the model and view the resulting predictions, both in terms of the responses of a single voxel to the stimuli used in our experiment, and the preferred period as a function of eccentricity and polar angle in response to those stimuli, as shown in the paper's figures.

With default parameters (no effect of orientation on tuning):

After fiddling with the parameters, to pull apart the tuning in response to different stimuli:

Docker container

The recommended way to run this app is using the provided docker image and docker-compose. Download and install docker and docker-compose, make sure the docker daemon is running (if you're not sure, try running sudo dockerd in the terminal), then run:

sudo docker-compose up

to download the image from dockerhub and start it up. Open http://localhost:8050/spatial-frequency-model in your browser to view the app.

Build Dockerfile

If you want to build the Dockerfile yourself (for testing local changes in this repo or because the dockerhub image has gone down), just run:

sudo docker build --tag spatial-frequency-model:latest ./

from this directory, then sudo docker-compose up and open http://localhost:8050/spatial-frequency-model in your browser.

Closing the container

If you opened this image with docker-compose, as recommended, you should be able to press ctrl+c once to kill it gracefully. If that doesn't work (or you started it in background mode), you may have to find the docker container ID and tell docker to kill. In a new terminal window, run:

sudo docker ps

which will list all running docker images. One of these should, under IMAGE, say billbrod/spatial-frequency-model or spatial-frequency-model. Copy the value under CONTAINER ID and then run:

sudo docker kill [CONTAINER ID]

which should kill the open process.

Run locally

If you want to run locally, you can either build the Dockerfile or install the package locally. If you want to do that, I recommend using conda or something similar to manage your python environments (see below for more information if you don't know how to set up an environment).

Once you have set up and activated the environment you want to install the package into, run the following:

pip install dash pip install -e . python webapp/app.py

And open http://localhost:8050/spatial-frequency-model in your browser (if it doesn't do so automatically). The dash version probably doesn't matter for anything we do, but the docker container uses 1.18.1 if you want to match it.

The app uses a variety of .svg images of equations (since MathJax and other LaTeX renderers apparently don't work in Dash right now). They can be found in the equations directory and are generated using the MathJax command line interface from the .tex files found there as well. A small python script, equations/convert.py, is used to script this conversion. The Docker image will automatically re-generate these every time it's built (in case the .tex files change), but you can generate them manually:

1. Install node.js and npm. (On Ubuntu: sudo apt install nodejs npm)
2. Use npm to install the MathJax command line interface: npm install --global mathjax-node-cli
3. Run the included python script: from this directory, run python equations/convert.py (it takes no arguments; it will convert all .tex files found in the same directory).

Using conda for python environments

To use conda to manage your python environments, make sure you have conda installed (if you don't already have it installed, see here for how to install miniconda, which is a minimal install, make sure to pay attention to how to set up your shell after installing it), then from this directory run the following on your command line:

conda create -n spatial-frequency python==3.7 conda activate spatial-frequency

This will create a new python 3.7 environment called spatial-frequency which will only contain the basics, and then "activate" that environment. Once an environment has been activated, all python-related commands (e.g., python, pip) will use those versions. This allows you to keep different versions and different packages in different environments. After you've created and activated this new environment, follow the steps above to install the webapp.