PyTau

API Documentation

https://abuzarmahmood.github.io/pytau/

Outline

• Perform inter-region transition correlations on models fit to each taste individually

Installation

# Create and activate conda environment
conda create -n "pytau_env" python=3.6.13 ipython notebook -y
conda activate pytau_env

# Clone repository
cd ~/Desktop
git clone https://github.com/abuzarmahmood/pytau.git

# Install requirements from specified file
cd pytau
pip install -r requirements.txt

# Test everything is working by running notebook
cd pytau/how_to
bash scripts/download_test_data.sh
cd notebooks
jupyter notebook
# Run a notebook

Data Organization

• Models stored in CENTRAL LOCATION and accessed by indexing an info file, which also contains model parameters and metadata
• Metadata also stored in model file to be able to recreate info file in case something happens

Database for models fit:

• Columns:

  • Model save path

  • Animal Name

  • Session date

  • Taste Name

  • Region Name

  • Experiment name (user-given name to separate batches of fits)

  • Fit Date

  • Model parameters:

    • Model Type (See below)
    • Data Type (Shuffled, simulated, actual)
    • States Num
    • Fit steps
    • Time lims
    • Bin Width

Data stored in models: - Model - Approx - Lambda - Tau - Data used to fit model - Raw data (before preprocessing to feed to model) - Model and data details/parameters (as above)

• Considering we might want to compare different model types:
  • Unconstrained sequantial
  • Baised transition priors, sequential
  • Hard padding between transitions, sequential
  • Joint region model We need to have a standardized pipeline for fitting and retrieving these models

Pipeline

Filelist

1. Model file

   • Invoked to generate model and perform inference

   • Currently <<poisson_all_tastes_changepoint_model.py>>

   • Should be agnostic to, and ignorant of what type of data is being fed. Should only fit model and return output

   • These should be functions

   • Input:

     1. Processed spike trains
     2. Model parameters (model type, states, fit steps)

   • Output:

     1. Model
     2. Approx
     3. Lambda
     4. Tau
     5. Data used to fit model

2. Data pre-processing code:

   • Currently <<changepoint_fit.py>>

   • Should handle different data "types" (shuffle, simulated, actual)

   • These should be functions

   • Input:

     1. Raw spike trains
     2. Spike train parameters (time lims, binning)
     3. Desired data type

   • Output:

     1. Processed spike trains

3. I/O helper code

   • Takes data path and model parameters
     • If model has not been fit, runs fit, else pulls model from memory
   • This should be a "data_handler" class that can load, process, and return the data, write data to file, and write an entry in the database
   • The model creation and model fit functions can be imported as methods to streamline fitting Operations:
     • Loads data from HDF5, preprocesses, and feeds to modeling code
     • Collects outputs from modeling code, and combines with appropriate metadata
     • Writes model file to appropriate directory
     • Row with fit data appended to central dataframe

4. Run script

   • Script to iterate over data
   • Input:
     1. List of data to iterate over

Notes

• Parallelization
  • Parallelization is currently performed using GNU Parallel by setting separate a theano compiledir for each job. This prevents compilation clashes. Refer to this file and this commit