3c1cf171a7
feat(lib): add _assets module for bundled-weight resolution
...
Provide bundled_path() and resolve() helpers that locate ONNX
weights and JSON metadata via importlib.resources, falling back to
a user-supplied directory. Missing assets raise ModelLoadError.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com >
2026-05-12 01:04:37 +08:00
8182bb1a01
feat(lib): add errors module
...
Introduce AiMouseError base class plus ModelLoadError and
GenerationError subclasses so downstream consumers can catch the
umbrella or specific failure modes.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com >
2026-05-12 01:02:57 +08:00
17b406c28b
feat(lib): add private _coord.py with numpy transforms
...
Copy of coord.py (which is already pure numpy) into the private
underscored module to be consumed by upcoming mouse.py rewrite.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com >
2026-05-12 01:02:25 +08:00
6702269494
test(tools): cover export_onnx with tiny synthetic models
2026-05-12 00:57:32 +08:00
fc7b2bd236
refactor(tests): split into tests/unit and tests/tools
2026-05-12 00:47:00 +08:00
3507fdc202
refactor: move eval/ and data_adapters/ to tools/
2026-05-12 00:40:33 +08:00
d1ecb13175
refactor: move server/ to tools/server/
2026-05-12 00:37:47 +08:00
6c96ab68c8
refactor(scroll): move trainer/models/collector to tools/scroll/
2026-05-12 00:34:05 +08:00
ba52c49edf
refactor: move trainer/models/utils/config to tools/
2026-05-12 00:30:55 +08:00
e3d4626031
test: capture scroll generate() golden output (pre-migration)
2026-05-12 00:20:13 +08:00
98daef54ca
test: capture mouse generate() golden output (pre-migration)
2026-05-12 00:19:39 +08:00
05be116cde
feat(eval): Markdown report builder with matplotlib plots
...
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com >
2026-05-10 13:39:13 +08:00
dc38f031b8
feat(eval): kinematics metrics, FFT spectrum, KL divergence
...
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com >
2026-05-10 13:33:19 +08:00
3fb4d3a8c0
feat(generator): add 5-point gaussian smoothing on lateral
2026-05-10 13:27:15 +08:00
50dbf40709
refactor(generator): remove deterministic speed_profile and hard log_dt clip
...
These post-processing hacks were added to compensate for small-data
training. With Balabit pretraining they suppress the multimodal
timing distribution and cause the template-y Δt curves seen in the
verify UI.
2026-05-10 13:22:57 +08:00
1d9dff4f2a
fix(trainer): rename _LOG_1_1 → _LOG_INV_0_9, add tests for aug_ids 2-5
2026-05-10 13:11:27 +08:00
e992823aef
feat(trainer): add resume_from for two-stage training
2026-05-10 13:08:37 +08:00
bba8fc567b
feat(trainer): replace eager _augment with streaming TrajectoryDataset
...
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com >
2026-05-10 12:51:52 +08:00
0b1d18d15c
feat(adapter): implement process_session and CLI
2026-05-10 12:42:43 +08:00
cc855e0ae2
feat(adapter): implement segment quality filters
2026-05-10 12:38:17 +08:00
82f803403c
feat(adapter): implement click-anchored segmentation
2026-05-10 12:35:43 +08:00
1559f7ab11
feat(adapter): implement Balabit CSV parser
2026-05-10 12:32:20 +08:00
97a835769e
feat(adapter): scaffold balabit data adapter package
2026-05-10 12:29:11 +08:00
4d414fd180
chore: initialize git repo, add matplotlib dep, extend config
...
- Add .gitignore for Python/data/models
- Add matplotlib>=3.8.0 for eval plots
- Add PretrainConfig, FinetuneConfig, BalabitAdapterConfig, EvalConfig dataclasses
2026-05-10 12:24:07 +08:00
