Skip to content

Per-RFU Outcome Metrics

The dnanet.evaluation.metrics.per_RFU module provides RFU outcome collection and binned F1 computation for precision/recall/F1 analysis by RFU threshold.

PerRFUOutcomeMetric

Collect RFU values for true positives, false positives, and false negatives.

from dnanet.evaluation.metrics.per_RFU import PerRFUOutcomeMetric

The metric intentionally does not bin RFU values during evaluation. It stores only the RFU values needed to compute precision, recall, and F1 later with arbitrary bin edges.

Constructor args: - threshold — Prediction threshold (default: 0.5)

State: - tp_rfus — RFU values for true positives - fp_rfus — RFU values for false positives - fn_rfus — RFU values for false negatives

Methods: - update(preds, targets, rfu_values) — Collect RFU values for a batch - compute() → dict[str, Tensor] — Return accumulated RFU values - compute_outcomes() → dict[str, Tensor] — Stable one-dimensional tensors

RFU I/O Functions

from dnanet.evaluation.metrics.per_RFU import (
    write_rfu_outcome_file,
    write_rfu_outcome_npz,
    write_rfu_outcome_csv,
    load_rfu_outcome_npz,
    load_rfu_outcome_csv,
)
  • write_rfu_outcome_file(path, outcomes) — Write based on file extension (.npz or .csv)
  • write_rfu_outcome_npz(path, outcomes) — Write as compressed NPZ archive
  • write_rfu_outcome_csv(path, outcomes) — Write as compact outcome,rfu CSV rows
  • load_rfu_outcome_npz(path) → dict[str, np.ndarray] — Load NPZ file
  • load_rfu_outcome_csv(path) → dict[str, np.ndarray] — Load CSV file

Binned F1 Computation

from dnanet.evaluation.metrics.per_RFU import (
    compute_binned_f1,
    compute_binned_f1_from_npz,
    compute_binned_f1_from_csv,
)
  • compute_binned_f1(outcomes, bin_edges) → list[dict] — Compute precision, recall, and F1 for arbitrary RFU bins
  • compute_binned_f1_from_npz(path, bin_edges) — Load NPZ and compute binned F1
  • compute_binned_f1_from_csv(path, bin_edges) — Load CSV and compute binned F1

Each returned dict contains: bin_left, bin_right, tp, fp, fn, support, precision, recall, f1.

Constants

  • RFU_OUTCOMES = ('tp', 'fp', 'fn')