cursus.steps.scripts.piper_metric_generation

check_call(cmd, *a, **k)
install_packages_from_public_pypi(packages)[source]

Install packages from standard public PyPI.

Parameters:

packages (list) – List of package specifications (e.g., [“pandas==1.5.0”, “numpy”])

install_packages_from_secure_pypi(packages)[source]

Install packages from secure CodeArtifact PyPI.

Parameters:

packages (list) – List of package specifications (e.g., [“pandas==1.5.0”, “numpy”])

install_packages(packages, use_secure=False)[source]

Install packages from PyPI source based on configuration.

This is the main installation function that delegates to either public or secure PyPI based on the USE_SECURE_PYPI environment variable.

Parameters:

  • packages (list) – List of package specifications (e.g., [“pandas==1.5.0”, “numpy”])

  • use_secure (bool) – If True, use secure CodeArtifact PyPI; if False, use public PyPI. Defaults to USE_SECURE_PYPI environment variable.

Environment Variables:

USE_SECURE_PYPI: Set to “true” to use secure PyPI, “false” for public PyPI

Example

# Install from public PyPI (default) install_packages([“pandas==1.5.0”, “numpy”])

# Install from secure PyPI os.environ[“USE_SECURE_PYPI”] = “true” install_packages([“pandas==1.5.0”, “numpy”])

detect_and_load_predictions(input_dir, preferred_format=None)[source]

Load the upstream prediction file, robust to the naming/format divergence across cursus inference & eval steps.

Filenames differ by producer (predictions.* / inference_predictions.* / eval_predictions[_with_comparison].*), all single-file and non-sharded, in one of csv/tsv/parquet/json. This globs the known base names (priority order) across the format list (preferred_format first) and loads the first match.

resolve_score_column(df, score_field, id_field=None, label_field=None)[source]

Resolve which column holds the positive-class model score, robust to the naming divergence across cursus inference/eval producers.

Resolution order (first match wins):

  1. explicit score_field (config SCORE_FIELD) if present;

  2. prob_class_1 — the positive-class prob from XGBoostModelInference / PyTorchModelInference / XGBoostModelEval (standard mode);

  3. new_model_prob_class_1 — XGBoostModelEval COMPARISON mode (where prob_class_1 is renamed away);

  4. the sole *_prob column — LightGBMMTModelInference single-task output (<task_name>_prob), excluding id/label columns;

  5. the sole non-id / non-label numeric column in [0, 1].

For MULTI-task LightGBMMT (several *_prob columns) the intended positive class is ambiguous — score_field MUST be set explicitly; this raises.

compute_metadata(df, y_true, dataset_type, amount_field)[source]

Build the PIPER metadata block shared by every Graph-Line / Tabular file.

Keys use the hyphenated PIPER contract names (record-count, fraud-count, fraud-rate, date-range-start, date-range-end, dataset-type). These are JSON string keys, so they are emitted as literal hyphenated strings.

write_curve_csv(output_dir, filename, header, x, y)[source]

Write a 2-column data CSV (with header) FLAT to output_dir.

Parameters:

  • header (Tuple[str, str]) – (x_header, y_header) e.g. (‘FPR’, ‘TPR’) or (‘Recall’, ‘Precision’)

  • x (ndarray) – paired arrays of equal length

  • y (ndarray) – paired arrays of equal length

write_metric_json(output_dir, filename, payload)[source]

Write a .metric JSON file FLAT to output_dir.

emit_roc(output_dir, y_true, variant_score, control_score, variant_model_id, control_model_id, metadata)[source]

Compute ROC curve(s), write per-series CSVs, and emit roc_curve.metric (Graph-Line). Control series is only added when control_score is present.

emit_pr(output_dir, y_true, variant_score, control_score, variant_model_id, control_model_id, metadata)[source]

Compute PR curve(s), write per-series CSVs, and emit pr_curve.metric (Graph-Line). Control series is only added when control_score is present.

precision_recall_curve returns (precision, recall, thresholds). We plot Recall (x) vs Precision (y) so the CSV header is ‘Recall,Precision’.

emit_data_statistics(output_dir, metadata)[source]

Emit data_preprocessing_statistic.metric (Tabular). Reuses the same metadata block computed for the Graph-Line files.

main(input_paths, output_paths, environ_vars, job_args)[source]

Main entry point for PIPER Metric Generation.

Reads eval predictions, recomputes ROC/PR curves, and emits the PIPER contract (.metric JSON + paired 2-column CSVs) FLAT to the output root so PIPER’s output-root scan finds every artifact.

Parameters:

  • input_paths (Dict[str, str]) – Dictionary of input paths (logical -> path)

  • output_paths (Dict[str, str]) – Dictionary of output paths (logical -> path)

  • environ_vars (Dict[str, str]) – Dictionary of environment variables

  • job_args (argparse.Namespace) – Command line arguments

create_health_check_file(output_path)[source]

Create a health check file to signal script completion.