cmbemu 0.1.0a1

CMB-spectrum emulator competition: pre-generated training data, Wishart-likelihood scoring, and CPU timing harness built on classy_szfast.

cmbemu

A CMB power-spectrum emulator competition: pre-generated training/test data, Wishart-likelihood scoring, and a CPU timing harness built on classy_sz / classy_szfast.

Install

pip install cmbemu

Quickstart

import cmbemu as cec

# (1) Fetch the pre-generated datasets (cached under ~/.cache/cmbemu)
train = cec.load_train()
test  = cec.load_test()

# (2) Optionally generate more spectra to study data-scaling
extra = cec.generate_data(n=20_000, seed=42, save_to="extra.npz")

# (3) Define your emulator
class MyEmulator:
    def predict(self, params: dict) -> dict:
        # params has keys: omega_b, omega_cdm, H0, tau_reio,
        # ln10^{10}A_s, n_s.
        # Returns a dict with keys tt, te, ee (len lmax_cmb+1)
        # and pp (len lmax_pp+1).
        ...

# (4) Score
result = cec.get_score(MyEmulator())
print(result["combined_S"], result["mae_total"]["mae"], result["timing"]["t_cpu_ms_mean"])

See scoring.md for the full likelihood and score definitions.

Hardware note

Timing is done CPU-only, single-threaded warm sequential calls — this mimics how an emulator is used inside a typical MCMC chain. Before running get_score with strict_cpu=True, export:

export JAX_PLATFORMS=cpu
export CUDA_VISIBLE_DEVICES=

Project layout

• src/cmbemu/ — the installable package
• scripts/ — internal dataset-building and diagnostic scripts
• scoring.md — formal score definition
• STATUS.md — current state of the competition design