registry.ClockRegistry

registry.ClockRegistry(clocks, version, path)

Loaded catalogue. Immutable except for user-registered local weights.

Methods

Name	Description
coefficients	Feature ids and coefficients, in the file’s order.
compatible_with	Clocks this dataset can actually be scored on.
deconvolution_table	The reference table (CpGs x cell types) a deconvolution entry reads.
filter	Exact-match filter; tuple-valued fields match on membership.
has_coefficient_vector	True when this clock has one weight per feature to hand out.
import_dunedinpace	Register DunedinPACE from the authors’ R package (installed, a source
import_pc_clocks	Register the PC clocks from the authors’ `CalcAllPCClocks.RData`.
intercept	The constant a coefficient file carries as an `(Intercept)` row.
pace_model	The weights, background means and model means DunedinPACE scores with.
quadratic_model	A model with squared-CpG terms (cAge): base CpGs, a frame of their
reference_label	What a warning calls the clock’s reference values, or None.
reference_values	The clock’s per-feature fill values, or None when it has none.
register_local_reference	Fill values for the absent features of a clock whose weights the user
register_local_weights	Supply a coefficient file the package does not distribute.
search	Substring match over id, name, predicts, notes and citation.
unavailable_message	The text a user sees when a clock cannot be scored. Worth care.
untraced	Clocks whose coefficients have no established primary source.
weight_record	What the run manifest records about this clock’s coefficients.

coefficients

registry.ClockRegistry.coefficients(clock_id)

Feature ids and coefficients, in the file’s order.

Raises

Name	Type	Description
	WeightsUnavailableError	For a licensed clock with nothing registered, or an untraced clock whose fetch has not happened. The message names the remedy.

compatible_with

registry.ClockRegistry.compatible_with(data, *, min_coverage=0.8)

Clocks this dataset can actually be scored on.

Compatibility is coverage, not platform. A clock trained on 450K runs perfectly well on EPIC data that happens to carry its probes, and fails on 450K data that has been filtered down to 20,000 probes. Only the feature list can answer the question, so bundled clocks are checked against it and the rest are reported by declared platform, with the difference made explicit in the reason string.

deconvolution_table

registry.ClockRegistry.deconvolution_table(
    clock_id,
    platform=None,
    present=None,
)

The reference table (CpGs x cell types) a deconvolution entry reads.

The table declared for platform; otherwise, when present (the data’s feature ids) is given, the one the data covers best; otherwise the first. A table the user registered for a licensed entry wins.

filter

registry.ClockRegistry.filter(**criteria)

Exact-match filter; tuple-valued fields match on membership.

reg.filter(availability=“bundled”, scale_type=“age_years”)

has_coefficient_vector

registry.ClockRegistry.has_coefficient_vector(clock_id)

True when this clock has one weight per feature to hand out.

False for a network, whose weights are layer matrices. The distinction matters to anything that squares a weight or sums the absolute weights: those operations are defined on a linear model and have no counterpart in a five-layer network. False too for a deconvolution entry, whose parameters are a table of cell-type means.

import_dunedinpace

registry.ClockRegistry.import_dunedinpace(path, out_dir=None)

See :func:falconage.registry.pace_import.import_dunedinpace.

import_pc_clocks

registry.ClockRegistry.import_pc_clocks(path, out_dir=None)

See :func:falconage.registry.pc_import.import_pc_clocks.

intercept

registry.ClockRegistry.intercept(clock_id)

The constant a coefficient file carries as an (Intercept) row.

None of the bundled files has one; their constants are add steps in the registry. A file a user registers often does, and without this the row would be aligned as a CpG the data lacks and filled.

pace_model

registry.ClockRegistry.pace_model(clock_id)

The weights, background means and model means DunedinPACE scores with.

quadratic_model

registry.ClockRegistry.quadratic_model(clock_id)

A model with squared-CpG terms (cAge): base CpGs, a frame of their linear and squared weights in each of its two models, and the two intercepts. See :mod:falconage.models.quadratic.

reference_label

registry.ClockRegistry.reference_label(clock_id)

What a warning calls the clock’s reference values, or None.

reference_values

registry.ClockRegistry.reference_values(clock_id)

The clock’s per-feature fill values, or None when it has none.

register_local_reference

registry.ClockRegistry.register_local_reference(clock_id, values)

Fill values for the absent features of a clock whose weights the user supplied, as the clock’s authors publish them (a mapping or a two-column feature_id,value file). Used exactly as a bundled reference is.

register_local_weights

registry.ClockRegistry.register_local_weights(clock_id, path, sha256=None)

Supply a coefficient file the package does not distribute.

Validates against the scaffold – two columns, no duplicate feature ids, finite values, and the declared feature count when the registry has one. A file that does not match is rejected with the mismatch named, which also makes this the way to check a coefficient set somebody handed you.

Returns the SHA-256 of the accepted file; it goes into the run manifest as weights_source: user_supplied, so a result computed from a licensed copy is distinguishable from one computed from a redistributed set.

search

registry.ClockRegistry.search(term)

Substring match over id, name, predicts, notes and citation.

unavailable_message

registry.ClockRegistry.unavailable_message(clock_id)

The text a user sees when a clock cannot be scored. Worth care.

untraced

registry.ClockRegistry.untraced()

Clocks whose coefficients have no established primary source.

weight_record

registry.ClockRegistry.weight_record(clock_id)

What the run manifest records about this clock’s coefficients.