thesis

Behavior and electrophysiology analysis code for Churchland Lab thesis work.

V1 GLM is the encoding-model workflow.

Requirements

• uv

Setup

git clone https://github.com/rojasgabriel/thesis.git
cd thesis
uv sync

Prepare a new ephys session

After LabData ingestion and spike sorting, check the session, then populate its clock mappings, unit metrics, quality counts, stability results, and stimulus responses:

uv run populate-session GRB064 20260916_132939
uv run populate-session GRB064 20260916_132939 --apply

The first command is a dry run. The second inserts missing results for project quality criteria 0-2 and parameter set 0, then checks their unit counts. Dry-run imports do not create schemas, tables, or lookup rows. The apply command creates missing analysis tables and seeds their defaults explicitly. Rerunning it keeps existing results. Use uv run insert_audio_events separately when measured OneBox audio must be recovered from raw files.

Ephys analysis figures

THESIS_CACHE_DIR=figures/cache nice -n 10 uv run python -m thesis.ephys.analyses.glm.flash_movement
uv run python -m thesis.ephys.analyses.stim_responses_by_locomotion
uv run double-peak-candidates GRB064 20260916_132939
uv run python -m thesis.ephys.analyses.double_peaks
uv run python -m thesis.ephys.analyses.category_decoding -a GRB006 -s 20240821_121447
uv run python -m thesis.ephys.analyses.rate_decoding -a GRB006 -s 20240821_121447

The first command builds the combined flash PSTHs and observed peak scatters, then the observed/full-model/without-flash-by-exit comparison on identical held-out trial windows and preceding-gap classes. It uses cached trials, a saved response classification, and completed GLM fits. The main outputs are figures/glm/flash_observation.svg and figures/glm/flash_comparison.svg; figures/glm/flash_model.svg is the supporting kernel-only view. glm.md gives the scientific argument and links to its checks. The second command makes the separate working peak comparison at figures/locomotion/locomotion_peaks.svg. The third shows confirmed 15 ms double-peak candidates for one session and writes figures/double_peak/grb064_candidates.svg. The double_peaks command makes the separate cross-session double-peak supplement, including its 30 ms control. The decoding commands decode the stimulus category, the choice, and the stimulus rate from the population firing rates of one session, and write figures/category_decoding/category_decoding.svg and figures/rate_decoding/rate_decoding.svg.

Tested on

• MacBook Pro M2
• macOS Golden Gate 27.0 (beta 6)
• Python 3.11.14