trulens.benchmark.alignment_report¶

trulens.benchmark.alignment_report ¶

Alignment diagnostics for benchmark and judge-alignment results.

Generate alignment diagnostics for benchmark predictions.

PARAMETER	DESCRIPTION
`predicted_scores`	Predicted judge scores. This may be a one-dimensional sequence, a pandas series, a numpy array, score-confidence pairs, or a `TruBenchmarkExperiment` result of `(scores, meta_scores)`. TYPE: `Any`
`true_labels`	Ground-truth labels or scores in the same order as the predictions. TYPE: `Any`
`examples`	Optional examples used only when displaying worst misses. TYPE: `Any \| None` DEFAULT: `None`
`threshold`	Threshold used for binary alignment metrics. TYPE: `float` DEFAULT: `0.5`
`thresholds`	Optional thresholds for confusion-matrix rows. TYPE: `Sequence[float] \| None` DEFAULT: `None`
`n_bins`	Number of bins for calibration and distribution data. TYPE: `int` DEFAULT: `10`
`top_n`	Number of worst misses to include. TYPE: `int` DEFAULT: `10`

to_dataframe() -> dict[str, DataFrame]

Return report sections as pandas dataframes.

print_summary() -> str

Print and return a concise console-friendly report.

plot() -> dict[str, Any]

Return matplotlib figures for calibration and score distribution.

to_html() -> str

Return a single HTML document containing all report sections.