Model Playground

A flexible and effective framework for model fitting and model comparison. This structure might come in handy when you are experimenting with multiple models.

Introduction

This project provides examples of how to perform model recovery. To see an example, run model_recovery.m. Here, we compare psychometric functions of a ternary temporal-order-judgment task from a paper to demonstrate model recovery. In this task, participants were presented with an audiovisual stimulus pair with varying stimulus-onset-asynchrony, and reported the perceived order (“visual first,” “auditory first,” or “simultaneous”). In this example, we compare two models, assuming the measurement distribution is either Gaussian or double-exponential. Will add some basic plots and parameter recovery later.

Usage

1. Define your model: In each folder, define your model, including its likelihood and simulation. Ensure that the nll_[folder name] is consistent with the folder name.
2. Adapt the recovery template: Modify model_recovery_template to include the models you want to compare and the ground-truth parameters.
3. Update parameter samples: Modify generate_param_samples.m to incorporate all the model parameters you are using.
4. Run!: The saved .mat file will contain everything needed to plot the recovery results.

Prerequisite

• BADS: model fitting tool by Maximum Likelihood Estimation. Feel free to use the tool you prefer.

Contact

For any questions or issues, please contact luhe.li@nyu.edu.