Fitr: A Platform for Computational Psychiatry Research

Overview

In decision-making tasks, it is often of interest to understand the psychological mechanisms by which a subject makes choices. One way to approach this problem is by specifying a reinforcement learning (RL) model that describes those mechanisms, and then fitting it (by tuning its free parameters) to subjects’ actual behavioural data. There are many ways to perform this analysis, which for the most part might be inaccessible to researchers without extensive mathematical training. We are building Fitr as an attempt to make this process simpler to implement.

For researchers interested in the computational aspects of decision-making research, Fitr also aims to offer pre-packaged routines to streamline the study & development of new tasks, models, and model-fitting/selection procedures.

Goals

  1. Offer a free and open platform upon which researchers can easily
    • Build and validate behavioural tasks
    • Build and validate behavioural models
    • Develop new model-fitting procedures
    • Fit models to behavioural data from subjects in vivo
  2. Implement the state of the art methods for behavioural modelling studies in computational psychiatry

  3. Integrate well with tools for collecting behavioural data

4. Integrate well with tools for collecting neurophysiological data <<<<<<< HEAD


>>>>>>> 2918408ba405464122dd27d7e4ca77f8fbb15028 Guiding Principles ——————

  1. Accessibility
    • Fitr should be open-source, free, and not dependent on commercial tools
  2. Parsimony
    • Build tools that can turn data into results with minimal coding
  3. Modularity
    • Build modules, classes, and functions in a way that facilitates the computational modelling workflow as it applies to
      • Developing and validating tasks
      • Developing and validating models
      • Fitting/selecting models using data from human subjects
  4. Flexibility
    • There are many ways to fit a model. Users should be able to easily test multiple models and multiple fitting methods without much additional code.
    • Allow users to easily integrate their own code, where desired
    • Facilitate development of “Pipelines” for Computational Psychiatry research
  5. Don’t re-invent the wheel (unless you have to)
    • If excellent open-source tools exist, don’t rebuild them. Rather, make it possible for users to integrate them easily into their workflow
    • Always give credit wherever credit is due
  6. Build for communication and reproducibility
    • Make it easy for researchers to generate the high-quality tables and plots necessary to communicate the results of their modelling studies.
    • Make it easy to reproduce results generated by Fitr pipelines

What we’re working on

  • Adding new tasks and new models
  • Writing more tutorials
  • End-to-end model-fitting and model-selection
  • Improving existing model-fitting algorithms
  • Adding new model-fitting algorithms (Variational Bayes)
  • Model-based neuroimaging

What’s New

See here

Citing Fitr

Let us know if you publish a paper using Fitr and we will post it here. If you use Fitr in your work, please cite it so that we can (A) know how people have been using it, and (B) support further funding of our work.

Indices and tables