lir.metrics package

lir.metrics.cllr(llr_data: LLRData, weights: tuple[float, float] = (1, 1)) → float[source]

Calculate a log likelihood ratio cost (C_llr) for a series of log likelihood ratios.

Nico Brümmer and Johan du Preez, Application-independent evaluation of speaker detection, In: Computer Speech and Language 20(2-3), 2006.

Parameters:

llr_data (LLRData) – LLRs and their metadata, wrapped in an LLRData object.
weights (tuple[float, float], optional) – The relative weights of the classes.

Returns:

CLLR, the log likelihood ratio cost.

Return type:

float

lir.metrics.cllr_cal(llr_data: LLRData, weights: tuple[float, float] = (1, 1)) → float[source]

Calculate the difference between the C_llr before and after isotonic calibration.

Parameters:

llr_data (LLRData) – LLRs and their metadata, wrapped in an LLRData object.
weights (tuple[float, float], optional) – The relative weights of the classes.

Returns:

CLLR_cal, the difference after isotonic calibration.

Return type:

float

lir.metrics.cllr_min(llr_data: LLRData, weights: tuple[float, float] = (1, 1)) → float[source]

Estimate the discriminative power from a collection of log likelihood ratios.

Parameters:

llr_data (LLRData) – LLRs and their metadata, wrapped in an LLRData object.
weights (tuple[float, float], optional) – The relative weights of the classes.

Returns:

CLLR_min, a measure of discrimination.

Return type:

float

lir.metrics.llr_lower_bound(llrs: LLRData) → float | None[source]

Provide corresponding lower bound for provided LLR data.

When an LLRData object contains a lower bound, return it. If not, return None.

Parameters:: llrs (LLRData) – LLRs and their metadata, wrapped in an LLRData object.
Returns:: The LLR lower bound, or None.
Return type:: float | None

lir.metrics.llr_upper_bound(llrs: LLRData) → float | None[source]

Provide corresponding upper bound for provided LLR data.

When an LLRData object contains an upper bound, return it. If not, return None.

Parameters:: llrs (LLRData) – LLRs and their metadata, wrapped in an LLRData object.
Returns:: The LLR upper bound, or None.
Return type:: float | None

Submodules

lir.metrics.overestimation.llr_overestimation(llrs: ndarray, y: ndarray, **kwargs: Any) → float[source]

Calculate the mean absolute value of the LLR-overestimation.

Parameters:

llrs (np.ndarray) – Array of log-likelihood ratios.
y (np.ndarray) – Array of labels (1 for H1 and 0 for H2).
**kwargs (Any) – Additional keyword arguments forwarded to calc_llr_overestimation.

Returns:

Mean absolute value of the overestimation grid, or np.nan when unavailable.

Return type:

float