chemicalchecker.tool.targetmate.nonconformist.nc.SignErrorErrFunc

class SignErrorErrFunc

Bases: RegressionErrFunc

Calculates signed error nonconformity for regression problems.

For each correct output in y, nonconformity is defined as

y_i - \hat{y}_i

References

[1]

Linusson, Henrik, Ulf Johansson, and Tuve Lofstrom. Signed-error conformal regression. Pacific-Asia Conference on Knowledge Discovery and Data Mining. Springer International Publishing, 2014.

Methods

apply	Apply the nonconformity function.
apply_inverse	Apply the inverse of the nonconformity function (i.e., calculate prediction interval).

apply(prediction, y)

Apply the nonconformity function.

Parameters

predictionnumpy array of shape [n_samples, n_classes]

Class probability estimates for each sample.

ynumpy array of shape [n_samples]

True output labels of each sample.

Returns

ncnumpy array of shape [n_samples]

Nonconformity scores of the samples.

apply_inverse(nc, significance)

Apply the inverse of the nonconformity function (i.e., calculate prediction interval).

Parameters

ncnumpy array of shape [n_calibration_samples]

Nonconformity scores obtained for conformal predictor.

significancefloat

Significance level (0, 1).

Returns

intervalnumpy array of shape [n_samples, 2]

Minimum and maximum interval boundaries for each prediction.