Quickstart

The fastest path is the CLI with a config file (see Configuration & CLI):

femmi run --config configs/default.yaml
python examples/plot_npz.py runs/run.npz     # truth / kappa / std panels

The Python API

examples/quickstart.py is the minimal end-to-end reconstruction from a galaxy shear catalog:

import numpy as np
from femmi.operators import build_operators_catalog
from femmi.catalog   import analytic_gaussian_catalog, reconstruct_catalog

# a synthetic catalog (x, y, g1, g2) with a known truth
cat = analytic_gaussian_catalog(n_gal=2000, sigma=0.5, shape_noise=0.05, seed=0)

# build the FEM-BEM operators with nodes AT the galaxy positions
ops, cm = build_operators_catalog(cat["x"], cat["y"], radius=None, n_boundary=48)

# MAP reconstruction with automatic lambda (Morozov noise matching)
kappa = reconstruct_catalog(ops, cm, cat["g1"], cat["g2"], noise_std=0.05)
print("reconstructed kappa on", len(kappa), "nodes")

What you get back

A femmi run writes runs/<name>.npz with node-aligned fields:

key	meaning
kappa	the reconstruction (MAP, or posterior point estimate)
std	per-node posterior uncertainty (sampling runs only)
truth	the input \(\kappa\) when a synthetic/known truth exists
nodes	the (N, 2) mesh-node coordinates

Plot them with examples/plot_npz.py, which draws each field as a tripcolor over the mesh triangulation.

More teaching examples — priors, uncertainty, KS head-to-head — are indexed in examples/README.md.