Lab Ledger is a micro project meant to help you be a better computational biologist. Given a base directory, labLedger finds git repos and stores their commits for the past week. It then uses a summarization model to summarize your work.
Lab Ledger, a project with no users other than its creator, is back to "pre-release" mode as I change this over to a RAG application, mostly for better understanding of what a RAG application is. Locally, you can use ingest.py on a repo of your choice to make a vector database of your commits.
In a less stable state, summarizer.py currently can perform a "stand up audit" on your commits using the embeddings you created.
Unlike wet-bench scientists, computationalists don't often have lab notebooks. Sometimes they have their own system, and this is mine.
Commit early, commit often
This is an aphorism commonly used in software development, and I try and use it in my experimental work. To make better commits and to adequately log my progress, I try and write down my experiments and review them each week to tell if I understood what progress I made for the last week. This helps me better understand my code and better present my project.
Install using pixi with pixi install. The only requirements are pytorch and the transformers library, but these are only necessary if you're looking for an automated summary.
Mostly this is meant as a personal lab notebook, but I intend to make it more general over time. To use, run reportCommits.sh after configuring the file labrat.conf:
AUTHOR # Your email used for your git commits
BASE_DIR # Where to recursively search for git repos
OUTPUT # Where to store your output
SUMMARIZE # Boolean to run the summarization script
While this project should not require much maintenance, there are several features I plan on implementing:
-
Commit flags. Commits could be prepended with tags such as RESULT: (commit message here) or FIGURE: (commit message here).
-
A TUI that allows you to "browse" your notebook.
While researching names for this project, I came across the project gitSum. It is an incredible name but I think the outputs/scope of each project are sufficiently different. I also had a previous coworker who had a much more extreme computational setup which was the inspiration for this project.