Optimization and Testing Summary (SCM)¶
ZeroProofML v0.4 removed the transreal-specific profilers and fusers. The optimisation surface now centres on the SCM primitives that keep ⊥ visible without branching.
Core Pieces¶
zeroproof.scm.ops– Vectorised arithmetic for NumPy/Torch/JAX that carries(payload, ⊥ mask)pairs so kernels stay fuseable.zeroproof.autodiff.policies– Gradient policies (PROJECT,CLAMP,REJECT,PASSTHROUGH) to control how singular paths influence learning.zeroproof.layers.SCMRationalLayer– Projective rational head exposing a bottom mask for coverage tracking and post-hoc decoding.zeroproof.scm.sign– Weak sign projection with hysteresis to stabilise orientation near the singular band.zeroproof.utils.ieee_bridge– Deterministic ingress/egress that collapses NaN/Inf to ⊥ so instrumentation stays consistent.
Testing Approach¶
- Unit tests in
tests/utils/test_ieee_bridge.pyverify IEEE round-trips and ⊥ construction. - Examples in
examples/exercise vectorised ops, gradient policies, and projective tuples as living documentation. - Benchmarks under
benchmarks/measure SCM throughput without any transreal tag handling.
Practical Tips¶
- Log ⊥ coverage per batch; use it as a gating signal in evaluation pipelines.
- Prefer vectorised helpers over Python loops to keep masks aligned and JIT-friendly.
- Clamp or project gradients when poles are expected; leave PASSTHROUGH for already-regularised layers.
- Treat IEEE conversion as an explicit boundary to avoid silent loss of ⊥ information.