No description
825ed6ad33
Stream 2 of the clean-slate rewrite: nisps/ml/ replaces src/memlp/ with a header-only, heap-free MLP that satisfies nisps::core::MLEngine. Files (nisps/ml/): - activations.hpp — ReLU (leaky 0.01 for parity), sigmoid, tanh - loss.hpp — MSE per-sample (fixes meml-ues double-scaling: returns the sample's MSE without an extra 1/N multiplication; the training loop averages explicitly) - init.hpp — uniform/Xavier/spread-aware weight init - training.hpp — gradient clip helper (±10.0 matches legacy) - rl.hpp — move_weights with per-layer Xavier scaling, weight decay (10% * spread), gaussian noise via the deterministic Rng (matches the legacy JS sum-of-three-uniforms shape); draw_weights also spread-aware - stats.hpp — per-layer mean/max/dead/saturating diagnostics - mlp.hpp — 4-layer (3 hidden + sigmoid output) MLP class with std::array-backed weights, biases, gradient accumulators, dataset ring buffer (default 128 examples), loss history (default 4096 iters). Bias is a separate per-layer parameter — no input-vector mutation. Flat get_weights/set_weights layout: weights all layers (row-major, layer order), then biases all layers. Tests (tests/cpp/, all 50 passing under -Wall -Wextra -Werror -Wpedantic): - test_mlp_init.cpp — deterministic seeding, spread regimes, static_assert MLEngine concept satisfied - test_mlp_inference.cpp — golden hand-computed forward pass match, sigmoid output range, set_input bounds - test_mlp_training.cpp — XOR convergence (loss < 0.01 in <2k iters), ring-buffer eviction - test_mlp_loss.cpp — meml-ues regression test: reported loss equals hand-computed average MSE without extra 1/N scaling; sample weights honoured - test_mlp_rl.cpp — move_weights respects output_pin_mask (final-layer rows + biases preserved); spread regimes; grad clear after draw_weights - test_mlp_serialize.cpp — get_weights/set_weights round-trip preserves inference exactly; eval_loss is non-mutating; infer_batch matches individual inference Verification: - Clean build, no warnings - 50 tests pass (22 prior + 28 new) - No std::vector / new / malloc in nisps/ml/ - All float literals .f-suffixed in code (comments excepted) |
||
|---|---|---|
| .beads | ||
| .claude/worktrees/agent-ae87fe47/playground/js/shapeseq | ||
| .github/workflows | ||
| .vscode | ||
| codegen | ||
| data | ||
| docs/redesign | ||
| modes | ||
| nisps | ||
| nisps-core | ||
| playground | ||
| schemas | ||
| scripts | ||
| src | ||
| tests | ||
| vcv | ||
| voicespaces | ||
| .gitignore | ||
| .gitmodules | ||
| AGENTS.md | ||
| ChannelStripAudioApp.hpp | ||
| CLAUDE.md | ||
| IMLInterface.hpp | ||
| LICENSE | ||
| MAP.md | ||
| MEMLNaut-NISPS.ino | ||
| NISPS_CORE_EXTRACTION_PLAN.md | ||
| NISPS_CORE_TASKS.md | ||
| package-lock.json | ||
| package.json | ||
| PAFSynthAudioApp.hpp | ||
| playwright.config.js | ||
| README.md | ||
| ThruAudioApp.hpp | ||
| XiasriAnalysis.cpp | ||
| XiasriAnalysis.hpp | ||
| XIASRIAudioApp.hpp | ||
Neural Interactive Shaping of Parameter Spaces
https://musicallyembodiedml.github.io/memlnaut/approaches/nisps
Firmware
The hardware firmware targets the MEMLNaut RP2350 build and uses repo-local helper scripts for the known-good build configuration:
git submodule update --init --recursive
scripts/build-firmware.sh
scripts/flash-firmware.sh
scripts/build-and-flash-firmware.sh
Notes:
- The scripts build for
rp2040:rp2040:solderparty_rp2350_stamp_xlwithOptimize3. - The build forces C++20 because the firmware uses
std::spanand concepts. build-firmware.shaccepts an optional variant name such asMEMLCeliumorBreakOr. Matching remains case-insensitive, somemlceliumstill works. If you omit it in an interactive shell, the script parsesMEMLNaut-NISPS.ino, prompts for a variant, and rewrites the activeMEMLNAUT_MODE_TYPEbefore building.flash-firmware.shaccepts an optional mountpoint argument, or auto-detects common UF2 bootloader mounts such as/run/media/$USER/RP2350and/run/media/$USER/RPI-RP2.
Web Playground
Try NISPS in your browser — no hardware required:
cd playground
python3 -m http.server
# Open http://localhost:8000
Train a neural network to map joystick positions to generative visuals through interactive machine learning. Two learning modes: direct example mapping and reinforcement learning with thumbs up/down feedback.
The playground UI includes an Expand toggle on the visual surface so you can make the canvas nearly full-screen while compressing parameter/control panels into a minimal strip beneath it.