I visited the python conference in Groningen: PyGrunn. On this page I collect some thoughts and learnings.

FastAPI-Restly

CRUD is a problem with fastapi. There are many plugins for fastapi attempting to provide better CRUD operations. With fastapi you create a function for each operation separately. Many developers end up writing their own utils for this. Restly provides a standardized way to generate the CRUD operation endpoints for a given python class

The dark arts of ML benchmarking

“There are two possible outcomes: if the result confirms the hypothesis, then you’ve made a measurement. If the result is contrary to the hypothesis, then you’ve made a discovery”

Enrico Fermi

When benchmarking, log more data not less. Storage is cheap, regret is expensive. Do processing later, log the raw results. You can also store the git hash in your experiment table to see exactly which code produced the result.

How to store and route your (physical) mail

You get envelopes from the tax authority. Sometimes they want information after multiple years. How to keep an archive? paperless-ngx is an open source document management system. Also possible to integrate a document scanner, and output integrations to automatically forward some types of documents.

The limits of python (and rust)

Application is converting data from medical experiments to the standard format called SDTM. Using pola.rs was a lot faster than native python. At one point he was tempted to move also the gui for the program to rust, but guis on rust are not mature yet.

JSON - freedom or chaos

A config json, used by multiple teams, very large, without a schema definition. Using pydantic to validate it. Pydantic has extra types, that can be very useful. A cool story on how to incrementally add validation for this file, while not putting too much pressure on all the teams involved.