abrahamnunes/fitr

Tools for computational psychiatry research.

Fitr is currently being rewritten from the ground up to be faster and more flexible. Stay tuned for updates.

Getting Started

Prerequisites

conda install --yes --file requirements.txt

or

pip install -r requirements.txt

Installation

pip install git+https://github.com/abrahamnunes/fitr.git

or

git clone https://github.com/abrahamnunes/fitr.git
cd fitr
python setup.py develop --user

Running Tests

conda install --yes --file requirements_test.txt

or

pip install -r requirements_test.txt

Documentation

• Full API documentation in the docs/api folder.
• Detailed documentation in docs/tex/fitr_documentation.pdf

Examples

• See examples/ folder
• Tutorials with deeper explanations in docs/tex/fitr_documentation.pdf

Contributing

• Being Updated

License

• This project uses the GNU General Public License v3.0 - see the LICENSE.md file for details

How to Cite

If you use Fitr in your work, we would very much appreciate the citation, which can be done as follows:

• Abraham Nunes, Alexander Rudiuk, & Thomas Trappenberg. (2017). Fitr: A Toolbox for Computational Psychiatry Research. Zenodo. http://doi.org/10.5281/zenodo.439989

fitr's Issues

Update PyPI package with new EmpiricalPriors addition

With the addition of the EmpiricalPriors model fitting method, the PyPI package can be updated.

Hoping you can do this, Alexander. Thanks.

Implement tasks

Fix pypi documentation

Documentation on pypi is not formatted.

Package for PyPI

We need to be able to install this using pip as a proper package. In it's current form, it is more like a Matlab toolbox...

Removing Old Branches

I think the following branches can be safely removed
abrahamnunes-path-1
docs
param_criticism
readmechange

And i'm not sure what the following two are for:
pyup-initial-update
gh-pages

Thoughts @abrahamnunes ?

Get unit tests architecture in

Make .travis test on oxs as well

Remove old requirements

I believe the file requirements_docs is for the old documentation, which should be removed.

Iteration Plan for 0.0.2

This plan captures the work that needs to be done before 0.0.2 of the package gets released.
The planned release date for this is late July.

This iteration will involve a complete re-work of the overall structure of fitr, along with a change in what tools will be provided, and where future work will be focused. This change is being made to reflect a more modular nature that is going to be taken with fitr. This means less off a focus on providing pre-made models, and rather providing more tools to build, visualize, fit, and test models.

Once this issue is closed then issue #66 will be used for final release.

Plan Items

I am following a similar structure to vscode. Below is a summary of the top level plan items.

Legend of annotations

Mark	Description
🏃	work in progress
✋	blocked task
💪	stretch goal for this iteration
🔴	missing issue reference
🔵	more investigation required to remove uncertainty
⚫️	under discussion within the team

fitr structure

• Replace fitr code #69
  • Remove unused fitr code #69
  • Add new fitr code @abrahamnunes #69
• Update README #68
  • Remove Unused Tags #68
  • Change to standardized open source template #68
  • Update to markdown from rst #68 #70
• Remove what's_new.rst #70
• Update CONTRIBUTING #76
  • Change to standardized open source template #76
• Check LICENSE #77
• Clear old issues or re-word #10 #11 #32 #34 #60
• Potentially remove _config.yml
• Update requirements.txt
• Potentially remove requirements_docs #85
• Investigate purpose of requirements_test
• Remove old branches #72

pypi Release

• Do a pass through setup.py
• ✋ Potentially remove setup.cfg if not needed anymore

Documentation

• Remove old documentation #69
• 🔴Find new documentation method

Example

• Remove old examples #69

Testing

• Remove old tests #71
• Update Travis to pass #71

Deferred Items

• Tests for new code #67
• 🔴New examples

Changing fitr code

I think that what we should do with the master branch is to remove all the old code in one commit, and then add the new code in another. I think this will look cleaner in git.

If that sounds good I will delete everything in the fitr subfolder and then you can fill it up later?
@abrahamnunes

EmpiricalPriors BIC and AIC negative sign

Just noticed that BIC and AIC are negative when empirical priors. This is just because the sign of the loglikelihood is reversed when they are calculated.

I will fix this.

Include Stan code in separate folder

Currently, I've put the stan model code in strings within the generative_models.py file, in respective classes. When I tried to include these files in a separate folder with links in the respective class to the .stan file, this was throwing errors.

I think it would be better to enable the package to include .stan files on their own, rather than restricting their inclusion to strings.

Tests for New Code

Once the new code is in this issue will be used to discuss the development of tests.

This issue closes #10

Moving to Python 3

I'm going to drop support for Python 2 in the next merge, and move to exclusive development in Python 3. Pretty much all libraries that fitr uses have robust Python3 implementations, and Python2 is nearing end of life. Consequently, I think spending further time supporting python2 in fitr is suboptimal.

If there are significant concerns about this, please let me know.

MDP Agents on Bandit Tasks

It is difficult to use an MDP agent on a Bandit task, mainly because of the eligibility trace update.

On a contextual 2 armed bandit task, the final action is $\mathbf u' = (0.5, 0.5)^\top$. The 0.5's are necessary in order to facilitate computation of the target $y_t = r_t - \mathbf u'^\top \mathbf Q \mathbf x'$ such that

However, the eligibility trace is updated as

which in a 4 state (2 context, 2 outcome) task with $\lambda = \gamma = 1$, and where $\mathbf x = (1, 0, 0, 0)^\top$, $\mathbf u = (1, 0)^\top$ and $\mathbf x' = (0, 0, 1, 0)^\top$, should result in a trace that looks like

The current setup will allow either the correct trace or the correct target calculation.

I think the solution may be to separate the trace updating function from the value function updating.

Updating tests

Since we are going to work with a new code base how do you feel about simply deleting all the current tests @abrahamnunes ?

Update Readme

The readme needs to be cleared up. I'm thinking a couple things are important for this.

1. Change the format from rst to md
2. Remove clutter from headers
3. Adopt a more standardized format

Any thoughts @abrahamnunes

This should also cover the old issue #11 which should be fixed.

Adding installation instructions for when installing from git

I was wondering if it would be worth mentioning in the git description that to install from the git repo you can run
python setup.py develop --user
in the fitr directory

Need to add Cross Validation methods to compare model predictions with actual data

I started a crossval branch in order to tackle this

Remove rst and switch to md

Some files are rst, and we all prefer md. This affects the readme and whats_new files.

Should not be a hard task to switch them over to md.

Coordinate transformations and identity transform.

Create utility function in fitr.utils to apply coordinate transformations. Currently this is in the data module, but not best put there.

Also we need an identity transformation for special cases where no transformation is desired.

_config.yml isn't needed anymore

Think this was for the old docs

Add GLM into MCMC and EM

The EMFit code by Quentin Huys includes ability to have regressors incorporated in the EM procedure. I will look further at his code and incorporate this into fitr.EM().

I will also look into building this into MCMC models.

Docs not rendering

The documentation written and compiled with sphinx is not rendering correctly.

@ARudiuk could you have a look to see what might be happening?

Many of our functions are decently documented, but for whatever reason I can't get them to render in sphinx with autodoc.

The contributing file is due for an update

The current contributing file is a bit dated, and I think it can be clearer.

I'm going to look into using a template and then filling it in.

Need to add Bayesian Model Selection

The Bayesian Model Selection script found in the fitr-matlab package needs to be translated in order to be used in this python version.

0.0.2 release

Steps for releasing next version

• Make sure issues in #65 are either complete or deferred
• Update setup.py info
• Build distributions
• Upload distributions

Create function that outputs summary of model fit statistics

"Duplicate citations" in documentation

I've been building documentation further on the "docs" branch.

A problem I've been running into is the generation of warnings because of duplicate citations. I've tried moving the references to the top of each module (which does end up shortening the function/class-level docstrings). However, this is somewhat ugly in the docs, from my perspective. Furthermore, there are still duplicates across modules.

Is there a way to have a global citation/reference list for the documentation, such that these warnings do not occur? I also think it would be cleaner this way.

Update packaging for pypi

Add everything from this link:
http://sherifsoliman.com/2016/09/30/Python-package-with-GitHub-PyPI/

Checking license

The current license seems fairly good, but I'm not sure how comfortable you are with the permissions.

There's a nice checklist of permissions at : https://github.com/abrahamnunes/fitr/blob/master/LICENSE.txt

Let me know what you think @abrahamnunes