H-OS-ELM migration to learned ELM-AE stacks¶
U5 replaces the v1 fixed random projection stack with a learned ElmAutoEncoderLayer stack and an online RlsSolver head.
Behavior change¶
HierarchicalOsElm::initializefits the ELM-AE feature stack on the initialization chunk.HierarchicalOsElm::updateupdates only the online RLS output head.- The old fixed
hiddenWeights_/hiddenBiases_internals are removed. - The final feature representation is exposed through
featureStack()for inspection and tests.
API notes¶
Existing four-argument construction still works:
feature_elm::HierarchicalOsElm<double> model(
numInputs,
hiddenNodesPerLayer,
feature_elm::ActivationFunction::kSigmoid,
feature_elm::Backend::kCpu);
Additional controls are available through the extended constructor:
feature_elm::HierarchicalOsElm<double> model(
numInputs,
hiddenNodesPerLayer,
feature_elm::ActivationFunction::kSigmoid,
feature_elm::Backend::kCpu,
feature_elm::RlsOptions<double>{},
1e-6,
42u);
Regression notes¶
- Chunked initialization should agree with full initialization where the online update assumptions hold.
- The feature stack weights must change after
initialize; fixed random stacks are no longer valid H-OS-ELM behavior. - See testing.md for the H-OS-ELM test group.