Repository navigation
Compose emgpredictor - #142
Open
ECEEvanCampbell wants to merge 307 commits into
Open
ECEEvanCampbell wants to merge 307 commits into
ECEEvanCampbell wants to merge 307 commits into
Conversation
Online regressor method visualize() had lag when plotting. Modified so it uses the matplotlib animation API instead.
Added ability to grab timestamp when calling get_data.
This method was essentially a copy of the RegressorController functionality. Replaced it with a call to the RegressorController.
Adding cyberglove parsing to Ninapro DB2.
Updates SiFi streamer to accept bridge version. Splits single SiFi streamer into separate streamers for armband and biopoint.
Removed a duplicate method that parent class already had.
.gitignore didn't ignore all .pyc files.
Controller had many methods that the user shouldn't call. Converted these methods to private.
Added class for Fitts task using regression. Needs to be modified for classification.
SocketController used a deque, which isn't thread-safe and meant that the SocketController's data parameter didn't update in all threads when it was updated in another. Replaced deque with multiprocessing queue, which is thread safe.
ClassifierController was not parsing all messages properly. Added parameters to allow for definite parsing and some checks to tell the user if they set something up incorrectly.
OnlineEMGClassifier.visualize hadn't been updated since shared memory changes. Modified to use new ClassifierController.
Regression worked in Fitts environment, but not classification. Modified data parsing to handle classification properly. Also added a parameter that the user can pass in to customize which commands cause which directions.
Sphinx 5.0.0 cannot be imported on Python 3.13 (it imports the stdlib `imghdr` module removed in 3.13), and the pinned myst-parser 0.18.1 / sphinx-rtd-theme 1.0.0 are incompatible with modern Sphinx. Bump the `# For Docs` pins to Sphinx>=8.1.3, myst-parser>=4.0.0, and sphinx-rtd-theme>=3.0.2 so `sphinx-build docs/source` succeeds, including the new Online Adaptation tutorial. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
memory until the end.
memory items now carry a state block (generation, total samples, closed)
and a notifier pool, so a write announces itself and each observer pairs
those published facts with its own criterion, instead of copying whole
buffers just to learn whether a counter moved:
- shared_memory_manager: Snapshot, commit/apply, subscribe/unsubscribe,
read_window/read_since
- reactive: hooks, criteria and executors over stateful shared memory
- event_log: cross-process record of every propagation decision, cheap
enough to leave switched on
- adaptation/hooks: environment_feedback -> memory_flag -> adapt_flag
expressed as reactions; Memory/AdaptationManager wait rather than spin
- EMGClassifier/EMGRegressor gain a reactive run path and fall back to
the polled loop when no notifier pool is available
- streamers hand the default notifier pool to the child process
GUI. pipeline.py exposes the same document, registry and compiler the editor drives, and runs a saved pipeline headless from the command line.
each generated from streamer/registry signatures so a new device or node appears without edits here. gui.py runs a manual render loop on every path so panels get a per-frame slice on the render thread, reuses an already-open panel instead of building a second one over the same tags, and no longer requires an online data handler up front.
spectrum, wavelet decomposition, TDPSD and Hjorth moments) so each is computed once per window batch. An unrecognised feature name now raises ValueError instead of being silently dropped, which previously surfaced much later as a dimension mismatch.
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
6 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This branch composes the online EMGPredictor/adaptation workflow. The most recent work on it (this session) adds three things:
The streamer writes each modality buffer and its sample counter under separate locks, and readers copied them in two separate critical sections. A batch landing between the two reads made the counter run ahead of the copied buffer, so log_to_file spliced dropped/duplicated samples into the saved signal (visible as "noise" that visualize never showed, because it is stateless per frame).
SharedMemoryManager.get_variables([...]) — copies several variables under a single held lock (one acquire when they share a lock; deadlock-safe id-ordered acquisition otherwise).
assign_shared_memory_locks(...) — gives and _count the same lock; used by all streamer factories in streamers.py.
OnlineDataHandler.get_data and MemoryManager.process_data now take an atomic (data, count) snapshot. SharedMemoryOutputWriter shares the buffer's lock with its _count.
Because every writer updates data before count, an atomic read can only ever see the count lagging the data (self-correcting) — never leading it, which is the failure mode.
Online Adaptation tutorial
New Sphinx page (docs/source/documentation/adaptation/) documenting the adaptation suite: a within-subject SGT initialization followed by an online environment that adapts the model in the loop. Grounded in the CoAdaptUnderCurriculum reference workflow and trimmed to the minimal wiring (the four cooperating processes, get_edil_adaptation_objects, the model/memory/feedback interfaces the user supplies, and the save_dir/file_path/load_dir linkages). Wired into the Modules toctree and the API reference (emg_toolbox.rst).
Doc toolchain bump
requirements.txt # For Docs pins bumped to Sphinx>=8.1.3, myst-parser>=4.0.0, sphinx-rtd-theme>=3.0.2. Sphinx 5.0.0 fails to import on Python 3.13 (removed stdlib imghdr) and the old myst/rtd-theme pins are incompatible with modern Sphinx. The full docs (including the new tutorial) now build cleanly.
Reviewer notes
The docs build was verified locally end-to-end (sphinx-build docs/source) — build succeeds; remaining warnings are all pre-existing in other pages.
The lock/snapshot change touches the shared-memory contract used by every streamer; the shared-lock convention is applied uniformly so existing streamers keep working.
This branch also carries the earlier compose_emgpredictor commits (incremental-learning updates, SiFi streamer, numpy>2.0).