Really enjoying winding down to Square Enix tonight

Thoughts and prayers Mikell

And we will very carefully not mention selection bias

For my Bayesians out there, this is P(B|A) when the more interesting estimand is P(A|B).

How it’s going

It’s like the commencement speaker who says “congratulations! Now you’re done with the easy part”

If you discover any breaking cases in the Adder app please let me know. I've been having a really good time putting it together and now feel an unreasonable amount of pride in it.

But instead I nerdsniped myself into writing an arbitrary precision addition function

codeberg.org/brohrer/graf...

Because it was focused on such a tiny slice of functionality with a particular use case, I was able to craft it to be simple and to give exactly the behavior I want.

The mpmath library in Python is capable of arbitrary precision, but still gives things like

-1.2 + 1.21 = 0.009999999999...

mpmath.org

The answer I was looking for was Python's decimal library. It gives

0.1 + 0.2 = 0.3
and
-1.2 + 1.21 = 0.01

docs.python.org/3/library/de...

several proposed solutions.

My own take on a similar class of problem is the Fuzzy Naive Cartographer. It’s focused on reinforcement learning applications and diverges from existing solutions in fun (to me) ways.

PDF:
brandonrohrer.com/cartographer

But once you do that, it’s no longer unobserved context, but signal.

The question of how to behave in a system where you can’t see some important information (partially observable) is described by POMDPs- partially observable Markov decision processes. POMDPs are a problem statement and there are

I don’t have an answer here but think that’s just because you’re asking a very important question.

To oversimplify, “context” here is all the unmeasured stuff, fundamentally unobservable to a machine. To formally treat it you need to measure it, to reduce it to numbers and/or symbols.

Never mind 🙂 I just read the top post. Let me look

I’m not sure i have anything for this. “Formal” brings to mind mathematically formal, in which case linear algebra fits, but I’m guessing that’s not what you’re looking for.

Can you say more about what you’d like?

Concurrency is much improved. Latency seems good now too-less than 10ms from the db. I expect Postgres will be more than enough for this application.

I had several—lack of input validation, floating point artifacts, and a db that couldn’t handle concurrency (sqlite).

Thanks to the suggestions of you and others I lined up some db alternatives. This new version uses Postgres.