Hippius
maincontainerPublic

buddy777/gen-0f32d21d95d3

sha256:0c8efd0433ad8f6d388e24eb48b4e202e16d27b7ff0db6074cd60b08bd42873c·Indexed Jul 27, 2026

Layers

4

Total size

51.2 KB

Files

4

Quantization

—

README.md

1012 B

cascade-fulltex-rm1

Full-context (4096) mixture-of-priors generator. Fifteen process families — trend/multi-seasonal structure with slow amplitude/phase modulation, structural regime shifts, multiplicative series, AR(2)/integrated/threshold dynamics, spectral GP and power-law long memory, OU stochastic volatility, physical sensors, seasonal counts, intermittent demand, events/outliers, and conditional-stability paths — with a realistic observation layer (censoring, quantization, sample-and-hold, transient shocks) and light mixup augmentation.

The generator implements a per-row regime mixture (texture_frac): each row is drawn either from a clean low-artifact slice of the prior or from the full textured distribution. This release ships the full distribution.

Deterministic: one np.random.default_rng(seed) in a fixed draw order; all parameters drawn for every row and applied to active subsets. NumPy/SciPy only; vectorised per family; chunked lazy generation with a bounded prefetch queue.

Files

4 items
  • generator.py

    e8a22fac8711

    47.3 KB

  • config.json

    efb2ed9fb18b

    2.7 KB

  • README.md

    ff817d4f3505

    1012 B

  • requirements.txt

    bce394a998f0

    186 B