mobgap.pipeline.evaluation.icc

mobgap.pipeline.evaluation.icc(

df: DataFrame,

reference_col_name: str = 'reference',

detected_col_name: str = 'detected',

*,

icc_type: str = 'icc2',

nan_policy: Literal['raise', 'omit'] = 'raise',

) → tuple[float, tuple[float, float]]

Calculate the intraclass correlation coefficient (ICC) for the detected and reference values.

Parameters:

df

The DataFrame containing the reference and detected values.

reference_col_name

The identifier of the column containing the reference values.

detected_col_name

The identifier of the column containing the detected values.

icc_type

The type of the ICC. Can be one of “icc1”, “icc2”, “icc3”, “icc1k”, “icc2k”, “icc3k”. See the documentation of the pingouin.intraclass_corr function for more information. Default is “icc2”, often also referred to as ICC(2,1).

nan_policy

How to handle NaN values. Can be one of “raise” (error is raised), or “omit” (NaN values are ignored). Default is “raise”.

Returns:

icc, ci95

A tuple containing the intraclass correlation coefficient (ICC) as first item and the lower and upper bound of its 95% confidence interval (CI95%) as second item.

Notes

Note, that in case of ICC2, the confidence interval is reported as [np.nan, np.nan] if the ICC is 1 or 0 (aka perfect agreement or disagreement) as the confidence interval is not defined in this case. Other implementations might return [1, 1] in this case.