LoginSign Up
← Back to all sparks
E

eratosthenes

INFRA · APIS
Velocity2.5

Archaeological Synchronism

github.com/scollinselliott/eratosthenes ↗

eratosthenes spends 0.1.0 hardening inputs rather than adding chronology methods.

archaeologybayesian-inferencemcmcinput-validationr-package
◆Current state
eratosthenes does Bayesian estimation of archaeological chronologies from relative sequences, absolute constraints and artifact assemblages. The 0.0.9 line built out the inference diagnostics — traceplots, histograms, batch-means MCSE reporting, displacement estimation — and then consolidated artifact probability-density estimation into a single gibbs_ad_type(). The 0.1.0 tag turns outward instead, adding validators for every user-supplied structure and replacing seq_check() with a more informative seq_diag().
◆Where it's heading
The package is moving from research code to something a non-author can run. Consolidating estimation behind one function, then wrapping every input class in a validator, are the two steps that make failures legible instead of cryptic, and the diagnostics added earlier serve the same end for the sampler itself. Nothing in the window changes the underlying model; the work is all about making it usable and its output checkable.
◆Prediction
With inputs validated and diagnostics in place, the next release is more likely to extend the constraint or assemblage modelling than to keep reworking the interface, though the feed's three sparse tags give little to read a cadence from.

◆Recent moves

  1. 10d ago

    Input validators added; seq_check() replaced by seq_diag()

    The 0.1.0 tag is an interface pass: validators now police events and sequences, absolute constraints, and finds and assemblage relationships, and the old seq_check() is folded into sequences() with a more informative seq_diag() taking its place. It makes bad input fail clearly rather than deep inside the sampler.

    View source ↗
  2. 1y ago

    Artifact p.d.f. estimation consolidated into gibbs_ad_type()

    Artifact probability-density estimation was consolidated into a single gibbs_ad_type(), alongside graphics fixes. The same simplification instinct that 0.1.0 applies to inputs, applied here to the estimation surface.

    View source ↗
  3. 1y ago

    MCMC diagnostics arrive: traceplots, histograms, batch-means MCSE

    The release that made the sampler inspectable: traceplots and histograms, consistent batch-means and MCSE reporting, and functions for estimating displacement. Convergence reporting is what lets a chronology estimate be defended rather than just produced.

    View source ↗