NMAproject

Neuronal Mouse Behavior Detection

This repository is for code used for detecting mouse behavior based on neuronal spiking data from the Steinmetz dataset. This project includes Python scripts for preprocessing data, optimizing hyperparameters, and implementing a recurrent neural network (RNN) model. The repository is structured as follows:

Files

1. optimize.py

   • Script for optimizing hyperparameters using Optuna.
   • Imports custom preprocessing script (wrangler.py) and a custom RNN model (model.py).
   • Utilizes GPU (if available) and sets random seeds for reproducibility.
   • Defines a function to scale input features.
   • Loads or preprocesses data based on the specified version and region.
   • Encodes target labels using LabelEncoder.
   • Implements a function (optim) for hyperparameter optimization using Optuna.
   • Saves optimization results in the 'flatbrain.db' database.

2. wrangler.py

   • Custom preprocessing script.
   • Downloads required data using dl_alldata() and dl_st() functions.
   • Implements preprocessing steps, including binning spike times and filtering based on wheel movement.
   • Defines a function (preprocess) for preprocessing neuronal spiking data.

3. util.py

   • Utility functions used in the preprocessing pipeline.
   • Includes functions for binning spike times and applying transformations to data based on provided coordinates.

4. model.py

   • Defines the RNN model (RNNClassifier) and the RNN model wrapper (RNNmodel).
   • Implements training, prediction, and cross-validation methods.
   • Utilizes PyTorch for model implementation.
   • Custom collate function for DataLoader is defined to handle variable-length sequences.

5. download.py

   • Script for downloading the Steinmetz dataset.

Getting Started

1. Clone the repository: git clone https://github.com/yourusername/mouse-behavior-detection.git
2. Install the required dependencies: pip install -r requirements.txt
3. Run the optimization script: python optimize.py

Notes

• Make sure to have GPU support for faster training; otherwise, the code will default to CPU.
• Adjust the parameters and configurations in the scripts based on your specific use case.

File Tree

download.py - downloads steinmetz data files and concats to alldata.py

download_lfp.py - downloads some slightly preprocessed steinmetz data

wrangler.py - wrangles all the steinmetz data into a usable numpy array

model.py - contains all model code for the RNN

util.py - contains some utility functions for rebinning

plots.py - contains plotting funcitons for rebinning and raster plots

rebin.py - contains rebinning functions, used for data augmentation

movement_onset_detection.py - contains a function to perform movement initiation detection, used to determine start time of relevant data

flatbrain.db - contains optimization history

datacheckpoint.pkl - a saved version of the wrangler output, used to save time but can be deleted to regenerate data

environment.yml - conda environment needed to replicate results

figs - figure output folder

tutorials - some ipynb files used for inspiration