Symbolic Learning in Python and Julia (github.com)

67 pointsby alex_hirner5 days ago15 comments

brumar 11 hours ago

How do we measure and compare prediction quality in a symbolic/deterministic context that also punish for complexity? I struggled on this in my phd thesis and had to brew my own Minimum Description Length implementation because I found nothing applicable. The statisticians I interrogated were like "don't try to do model selection on symbolic models, make them probabilistic and work from there"

nsajko 8 hours ago

The approach here assigns a cost to each operator, variable and constant. See docs.

verdverm 7 hours ago

Are you familiar with NSGA-2? That was the one I used in my PhD work.

A lot of research in GP/SR went toward methods for balancing density and selection from the pareto front.

One of the bigger problems is growth of meaningless equation material to counteract other pressures in the algorithm, which itself leads to premature convergence on suboptimal solutions. I had some thoughts at the time about ways to penalize this during the growth phase in my own algo, but never got around to them. Mainly needs bookkeeping for contribution to the final equation output.

JanisErdmanis 11 hours ago

Is the Julia part here compiled via JuliaC and if not what would it take to do so?

bikvfeee 8 hours ago

It's 1 command line to compile with juliac.

owls-on-wires 10 hours ago

A deterministic SR engine in Rust, written for SRBench. Happy to see the field getting some front-page attention!

verdverm 8 hours ago

Is there a paper for your method? I'm always keen to see other deterministic algos for SR, it sounds interesting from what I can glean, but I don't read/write rust so may need an agent to help me decipher

I'll see about getting mine added to SRbench as well

verdverm 5 hours ago

Is your project limited in scope to physics equations? The coefficient snapping is interesting, but seems like it would not work well for problems like predator-prey (https://en.wikipedia.org/wiki/Lotka%E2%80%93Volterra_equatio...). I mainly worked with biology and life-sciences data. (real-world, separate from the synthetic benchmark data)

I introduced a new type of "constant" for well known values like this that sit between measured data and free parameters. I think I had some idea similar to your snapping, i.e. to look for places where free parameters could be replaced with const, but multiples of them was not part of that.

saithound 8 hours ago

PySR itself is still alright, mostly because of previous development. But not for long. Quality control has collapsed, and e.g. the examples in the repo are now embarrassing vibeslop.

Pangram reports 100% AI on most comment blurbs in the examples directory. That in itself is not the problem, AI-generated code could still be valid. But in this case, the AI-generated examples have the characteristic shape of slop: plausible sounding technical jargon, combined with zero actual understanding of either the target domain, or even evolutionary search (loss function design). They were vibed by people who had no idea what they were doing.

I suggest taking a look at the "evolved PRNG" example [1].

First, its loss function is basically designed to reward discovering linear feedback shift registers (LFSRs) only. It deliberately gives a massively lower loss to LFSRs than most other functions.

The AI slop phrased this in terms of "certifying the period", but the period certificate used only identifies LFSRs of max period, is unsound on the rest of the search space, and gives a meaningless very high loss to most non-LFSRs. So the regression will almost always just return an LFSR.

Whoever designed the example (Codex I suppose) also knew that most LFSRs are defective, and make poor random number generators. Thus, they included a "shear" penalty to detect such defective LFSRs, but of course they slopped it together.

The 360 correlations they measure should have been added together (having just one strong correlation is bad, and having two strong correlations is twice as bad as havin one). Instead, they take the MEAN of the correlations, so a perfect correlation contributes only 1/360=0.00278 to the loss.

And indeed, the result of their example run, reported in the docs [2] has such a perfect correlation: bit 29 is copied to bit 31 every two iterations.

[1] https://github.com/astroautomata/PySR/blob/master/examples/p...

[2] https://pysr.ai/v2.7.0/examples/objectives#inventing-a-pseud...

verdverm 7 hours ago

> by people who had no idea what they were doing.

I would personally hold this opinion of anyone still using genetic/evolutionary inspired algorithms for symbolic regression. There are multiple deterministic algorithms that do not have the same inherent drawbacks and are generally quite good.

Before I wrote my algo, and was trying to scale out GP, I hit the limits of RNG and had to go find some with seriously huge cycle properties. (howany calls before they reached the end and started over producing the same sequence)

verdverm 5 days ago

A deterministic algorithm for Symbolic Regression, which I recently ported to JAX

verdverm 14 hours ago

weird that my comment time was updated, this was not today

(original post refreshed by HN mods if I understand correctly, maybe dang or tomhow will chime in on this)

Jtsummers 14 hours ago

Second chance pool. All comments get a new timestamp with the submission for a while. Eventually it will have its original timestamp.

jgalt212 10 hours ago

Has anyone thrown the n-body problem at this?

thatsabadlook 14 hours ago

Yay now when you load your python project you have 2 interpreters, 2 gcs, 200mb of ram overhead, and nondeterministic performance due to joat.