auto_tol: Significance tolerance#
- hypercoil.loss.auto_tol(batch_size: int, significance: float = 0.1, tails: int = 2) float[source]#
Automatically set the tolerance for batch-dimension correlations based on a significance level.
From the t-value associated with the specified significance level, the tolerance is computed as
\(r_{tol} = \sqrt{\frac{t^2}{N - 2 - t^2}}\)
Warning
The tolerance computed corresponds to an uncorrected p-value. If multiple tests are performed, it might be necessary to use a more sophisticated correction.
- Parameters:
- batch_sizeint
Number of observations in the batch.
- significancefloat in (0, 1) (default 0.1)
Significance level at which the tolerance should be computed.
- tails1 or 2 (default 2)
Number of tails for the t-test.
- Returns:
- float
Tolerance for batch-dimension correlations.