EvoForest-WM: Discovered Neuro-Symbolic Foundational World Model for Multivariate Time-Series · GitHub

/" data-turbo-transient="true" />

Skip to content

-->

Search Gists

Search Gists

Sign in

Sign up

You signed in with another tab or window. Reload to refresh your session.<br>You signed out in another tab or window. Reload to refresh your session.<br>You switched accounts on another tab or window. Reload to refresh your session.

Dismiss alert

{{ message }}

Instantly share code, notes, and snippets.

kayuksel/evoforest_wm.py

Last active<br>June 24, 2026 23:29

Show Gist options

Download ZIP

Star

(0)

You must be signed in to star a gist

Fork

(0)

You must be signed in to fork a gist

Embed

Select an option

Embed<br>Embed this gist in your website.

Share<br>Copy sharable link for this gist.

Clone via HTTPS<br>Clone using the web URL.

No results found

Learn more about clone URLs

Clone this repository at &lt;script src=&quot;https://gist.github.com/kayuksel/cbb304038471befead2c1926610e9ff4.js&quot;&gt;&lt;/script&gt;

" readonly="readonly" data-autoselect="true" data-target="primer-text-field.inputElement " aria-describedby="validation-42f8e52b-8141-4dd8-be4b-982762eacce3" class="form-control FormControl-monospace FormControl-input FormControl-small rounded-left-0 rounded-right-0 border-right-0" type="text" name="gist-share-url-sized-down" />

Save kayuksel/cbb304038471befead2c1926610e9ff4 to your computer and use it in GitHub Desktop.

Embed

Select an option

Embed<br>Embed this gist in your website.

Share<br>Copy sharable link for this gist.

Clone via HTTPS<br>Clone using the web URL.

No results found

Learn more about clone URLs

Clone this repository at &lt;script src=&quot;https://gist.github.com/kayuksel/cbb304038471befead2c1926610e9ff4.js&quot;&gt;&lt;/script&gt;

" readonly="readonly" data-autoselect="true" data-target="primer-text-field.inputElement " aria-describedby="validation-4858522a-bf2a-4370-9b77-98f89c449e05" class="form-control FormControl-monospace FormControl-input FormControl-small rounded-left-0 rounded-right-0 border-right-0" type="text" name="gist-share-url-original" />

Save kayuksel/cbb304038471befead2c1926610e9ff4 to your computer and use it in GitHub Desktop.

Download ZIP

EvoForest-WM: Discovered Neuro-Symbolic Foundational World Model for Multivariate Time-Series

Raw

evoforest_wm.py

This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.<br>Learn more about bidirectional Unicode characters

Show hidden characters

import math

import torch

import torch.nn.functional as F

L = 64

def _g(seed):

return torch.Generator().manual_seed(seed)

class WMTransform:

"""The frozen TS-WM patch encoder phi: (B, L) -> (B, 141). Pure closed form over seeded banks."""

def __init__(self, L=L, device="cpu", dtype=torch.float32):

self.L = L

self.device = device

# --- seeded random-projection banks (the only stochastic content; fixed integer seeds) ---

self.trf_mu = torch.rand(12, generator=_g(11)) # Gaussian-window centers in [0,1]

self.trf_sig = torch.exp(torch.rand(12, generator=_g(12)) * 2.59 - 3.51) # window widths

self.srf_W = torch.randn(12, 6, 12, generator=_g(21)) / 3.4641 # random MLP on the 12 stats

self.srf_b = torch.randn(12, 6, generator=_g(22))

self.srf_u = torch.randn(12, 6, generator=_g(23)) / 2.4495

self.crf_w = torch.randn(16, 1, 9, generator=_g(31)) # 16 random conv kernels (size 9)

for k in ("trf_mu", "trf_sig", "srf_W", "srf_b", "srf_u", "crf_w"):

setattr(self, k, getattr(self, k).to(device=device, dtype=dtype))

# the column width and ordered builders of the 18 families

self._families = [

("stats", 12), ("srf_mlp", 12), ("autocorr", 2), ("spectral", 2), ("turning", 2),

("trf_gausswin", 12), ("crf_ppv", 32), ("hilbert_env", 2), ("crf_max", 32),

("morphology_updown", 4), ("fftbands", 6), ("perm_entropy", 3), ("curvature", 3),

("conv_position", 3), ("ar_residual", 4), ("acf_first_min", 3), ("histogram_mode", 3),

("ricker_wavelet", 4),

# ----- family 0: stats (12) — also consumed by srf_mlp -----

def stats(self, w):

m = w.mean(1); s = w.std(1).clamp(min=1e-8); c = w - w.mean(1, keepdim=True)

return torch.stack([

m, s,

(c ** 3).mean(1) / (s ** 3 + 1e-8), # skew

(c ** 4).mean(1) / (s ** 4 + 1e-8) - 3.0, # excess kurtosis

torch.quantile(w, 0.25, dim=1), torch.quantile(w, 0.5, dim=1), torch.quantile(w, 0.75, dim=1),

torch.quantile(w, 0.75, dim=1) - torch.quantile(w, 0.25, dim=1), # IQR

(c[:, :-1] * c[:, 1:]).sum(1) / ((c[:, :-1] ** 2).sum(1).sqrt() * (c[:, 1:] ** 2).sum(1).sqrt() + 1e-8),

(w[:, 1:] - w[:, :-1]).abs().mean(1), w.amin(1), w.amax(1),

], dim=1)

# ----- family 1: random-MLP projection of the stats vector (12) -----

def srf_mlp(self, w, st):

h = torch.relu(torch.einsum("nm,kdm->nkd", st, self.srf_W) + self.srf_b.unsqueeze(0))

return (h * self.srf_u.unsqueeze(0)).sum(2)

# ----- family 2:...