Bayesian Analysis of Strawlab Tethered Data

Exploring and understanding

• Find discussions in the wiki.

• Slides. I gave a talk early on the process of writing the code (so the slides are fairly outdated). I can send it to you, or you should be able to find it in the talks directory on the strawlab central storage, somewhere like "santi/2014/2014-02-13-labmeeting-bayesian". If you open the HTML and put it in full screen (so elements should not be moved), at some point you will be able to see a graphical representation of one of the models, which should prove useful to understand what they do. Also be careful and do not look at the results, they changed a lot since them and I even remember some plot labels were messed when I gave the talk.

• Andreas FAQ python, with answers

Install

All dependencies are either declared in setup.py or hardcopied in the externals package. This makes everything work on current anaconda (as per 2015/03/23).

• Install conda / anaconda.
• Install thebas dependencies:

conda install h5py pymc numpy scipy pandas matplotlib dill
conda install pydot  # this will unfortunatelly downgrade pyparsing
pip install joblib argh

• Install thebas. Either clone the repository and install in development mode (recommended)...

git clone https://github.com/strawlab/thebas
pip install -e thebas

... or pip install-it from github

pip install 'git+git://github.com/strawlab/thebas.git'

Everything should already be installed in strz (using the default anaconda environment in /opt/anaconda).

Use

This is a small tour on the different scripts provided.

# Activate anaconda or just use anaconda's python executable
alias python='/opt/anaconda/bin/python'

# Go to the cloned repository
cd thebas

# The data is read from 
#   *"/mnt/strawscience/data/forSANTI/closed_loop_perturbations"*.
# This munges these data into a more suitable format for analysis 
# This is an example of data loading that also does some crazy printing
python thebas/sinefitting/perturbation_experiment.py
# Note that the last version of the munged data is in this repo
# (directory data). So every further script could be run without
# any access to strawscience.

# This shows examples of command lines that can be used to fit bayesian models...
python thebas/sinefitting/samplers.py cl

# To run many fittings in parallel, we can use gnu parallel
python thebas/sinefitting/samplers.py cl | parallel -j12

# Let's toy-fit one model, 4 chains, only 800 iterations...
python -u thebas/sinefitting/samplers.py sample --pbproject dcn --freq 4 --genotype-id VT37804_TNTE --model-id gpa_t3 --iters 800 --burn 400 &>~/dcn__gpa_t3__VT37804_TNTE__4.log
# To add new models, just edit thebas/sinefitting/models (warning loads of useful copy and paste there!)
# Copy a model to modify and add it at the end to the dictionary "MODEL_FACTORIES"
# pbproject can be one of "dcn" or "hs"

# The traces and generated plots stay here:
ls /mnt/strawscience/santi/dcn-tethered

# --- There are some results-analysis scripts in thebas/sinefitting/reports

# Plot bias against the wing-beat-amplitude
python -u thebas/sinefitting/reports/bias_vs_wba.py

# Generate the "Bayesian BODE plots"
python -u thebas/sinefitting/reports/bbode.py

# Run NHST wannabes
python -u thebas/sinefitting/reports/nhst.py

# Show information about the stimulus perturbation
python -u thebas/sinefitting/reports/perturbation.py