PyramidFormer heaping-detector

0.8M parameters. Outputs, for every single age 0-100, the probability that the reported count is inflated by digit preference (more than 5% above the true count). Trained and evaluated on synthetic heaping: 80% of pyramids heaped at multiples of 5 within +-2 years, with a random subset of active heaps and random strengths, 20% clean, plus roster noise. Real survey heaping differs by country, sex and age; this has not been validated on real survey data.

Evaluation (held-out countries, 3404 pyramids, synthetic)

Method Precision Recall F1 False-positive rate on clean pyramids
Digit rule (flag every age ending 0/5) 0.473 1.000 0.642 0.188
PyramidFormer (p >= 0.5) 0.975 0.995 0.985 0.000

PyramidFormer AUROC 1.000, average precision 0.999. The digit rule has perfect recall by construction on this simulator, so compare precision and the false-positive rate on clean pyramids.

Training run

quick run: 2000 optimisation steps.

Training data

UN World Population Prospects 2024, single-age population by sex, countries/areas, 1950-2023 (estimates), as processed in SallySims/wpp2024-single-age-pyramids. Split by country: 80% train / 10% validation / 10% test, so every number above is on countries the model never saw. WPP pyramids are modelled estimates, not raw census counts. Source: United Nations, DESA, Population Division (2024). World Population Prospects 2024. https://population.un.org/wpp/ Check the UN terms of use before reuse. Research prototype; not for official statistics. This repo is self-contained: pyramidformer.py (model code), model.safetensors, config.json. There is no hosted inference widget because the architecture is custom; use the snippet below.

Usage

from huggingface_hub import snapshot_download
import importlib.util, os
p = snapshot_download("SallySims/pyramidformer-heaping-detector")
spec = importlib.util.spec_from_file_location("pyramidformer", os.path.join(p, "pyramidformer.py"))
P = importlib.util.module_from_spec(spec); spec.loader.exec_module(P)
model, norm = P.load_pretrained(p)
import numpy as np
# obs: (n, 2, 101) reported population, [male, female] x single ages 0..100; years: (n,) survey years
obs = obs / obs.sum(axis=(1, 2), keepdims=True)
prob = P.predict_heaping(model, norm.encode(obs), years)    # (n, 101): P(age is inflated by digit preference)
flagged = prob >= 0.5
Downloads last month
-
Safetensors
Model size
832k params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Dataset used to train SallySims/pyramidformer-heaping-detector