Expand description
Free Probability for Rust.
A port of free-probability-c with corrections and enhancements.
§Why free probability?
Xavier initialization. He initialization. Kaiming initialization. They match Marchenko-Pastur. Free probability explains why.
§Modules
moments— Empirical moment computation, moment↔cumulant transformsr_transform— R-transform (free additive convolution)s_transform— S-transform (free multiplicative convolution)marchenko_pastur— Marchenko–Pastur density and momentsprediction— Layer-combination prediction and gradient analysis
Modules§
- marchenko_
pastur - Marchenko–Pastur distribution.
- moments
- Moment computation and free cumulant ↔ moment transforms.
- prediction
- Gradient analysis and eigenvalue-distribution prediction for deep networks.
- r_
transform - R-transform: the free additive convolution analogue.
- s_
transform - S-transform: the free multiplicative convolution analogue.
Structs§
- Empirical
Dist - An empirical distribution sampled at discrete points.
Constants§
- FP_
MAX_ ORDER - Maximum supported moment/cumulant order.