Kaleidophon/deep-significance

Provides implementations of specialized statistical tests including Almost Stochastic Order (ASO), bootstrap, and permutation-randomization methods designed to handle the stochastic nature of neural network training across multiple runs and hyperparameter settings. Works directly with PyTorch tensors, TensorFlow tensors, NumPy arrays, and JAX arrays without conversion overhead. Includes Bonferroni correction for multiple comparisons across datasets and bootstrap power analysis for determining adequate sample sizes.

339 stars. No commits in the last 6 months.