1 RriOrdinalBinner

rri_metrics.RriOrdinalBinner(
    num_classes=0,
    edges=(lambda: torch.empty((0,), dtype=(torch.float32)))(),
    midpoints=None,
    bin_means=None,
    bin_stds=None,
    bin_counts=None,
    _rri_chunks=list(),
)

RRI → ordinal label mapping (CORAL-compatible).

1.1 Attributes

Name Description
num_classes Number of ordinal classes \(K\).
edges Quantile edges. Shape (K-1,).
midpoints Bin midpoints. Shape (K,).
bin_means Bin means. Shape (K,).
bin_stds Bin standard deviations. Shape (K,).
bin_counts Per-class sample counts. Shape (K,).

1.2 Methods

Name Description
transform Convert oracle RRI values to ordinal labels.
labels_to_levels Convert ordinal labels to CORAL level targets using this binner.
rri_to_levels Convert continuous RRI values directly to CORAL level targets.
class_midpoints Return per-class RRI midpoints derived from quantile edges.
class_priors Return class priors from fitted counts or fall back to uniform.
threshold_priors Return cumulative priors P(y > k) for CORAL thresholds.
expected_from_probs Compute expected RRI proxy from class probabilities.
save Save a fitted binner as JSON.
load Load a fitted binner from JSON.
load_fit_data Load flattened RRI samples from a saved binner fit-data file.
fit_from_iterable Fit a binner from a stream of RRIs, optionally resumable via fit_data_path.

1.2.1 transform

rri_metrics.RriOrdinalBinner.transform(rri)

Convert oracle RRI values to ordinal labels.

1.2.1.1 Parameters

Name Type Description Default
rri Tensor Oracle RRI values. Shape (...,). required

1.2.1.2 Returns

Name Type Description
Tensor Tensor["...", int64] labels in [0, K-1].

1.2.2 labels_to_levels

rri_metrics.RriOrdinalBinner.labels_to_levels(labels)

Convert ordinal labels to CORAL level targets using this binner.

1.2.3 rri_to_levels

rri_metrics.RriOrdinalBinner.rri_to_levels(rri)

Convert continuous RRI values directly to CORAL level targets.

1.2.4 class_midpoints

rri_metrics.RriOrdinalBinner.class_midpoints()

Return per-class RRI midpoints derived from quantile edges.

1.2.4.1 Returns

Name Type Description
Tensor Tensor["K", float32] bin midpoints for ordinal classes.

1.2.5 class_priors

rri_metrics.RriOrdinalBinner.class_priors()

Return class priors from fitted counts or fall back to uniform.

1.2.6 threshold_priors

rri_metrics.RriOrdinalBinner.threshold_priors()

Return cumulative priors P(y > k) for CORAL thresholds.

1.2.7 expected_from_probs

rri_metrics.RriOrdinalBinner.expected_from_probs(probs)

Compute expected RRI proxy from class probabilities.

1.2.7.1 Parameters

Name Type Description Default
probs Tensor Tensor["... K"] class probabilities. required

1.2.7.2 Returns

Name Type Description
Tensor Tensor["..."] expected RRI values using bin midpoints.

1.2.8 save

rri_metrics.RriOrdinalBinner.save(path, *, overwrite=False)

Save a fitted binner as JSON.

1.2.9 load

rri_metrics.RriOrdinalBinner.load(path)

Load a fitted binner from JSON.

1.2.10 load_fit_data

rri_metrics.RriOrdinalBinner.load_fit_data(path)

Load flattened RRI samples from a saved binner fit-data file.

1.2.10.1 Parameters

Name Type Description Default
path str | Path Path to a .pt file that stores rri_chunks. required

1.2.10.2 Returns

Name Type Description
Tensor Tensor["N", float32] of finite RRI samples.

1.2.11 fit_from_iterable

rri_metrics.RriOrdinalBinner.fit_from_iterable(
    iterable,
    *,
    num_classes=15,
    target_items=None,
    max_skips=0,
    fit_data_path=None,
    resume=False,
    save_every=1,
    on_progress=None,
)

Fit a binner from a stream of RRIs, optionally resumable via fit_data_path.

1.2.11.1 Notes

  • All fit data is stored on CPU.
  • Fit data is saved on Ctrl-C / exceptions when fit_data_path is provided.
  • The iterable may yield either rri tensors or (rri, meta) tuples.