TinyMyo: Update hand gesture classification experiment details - #8
Open
MatteoFasulo wants to merge 24 commits into
Open
TinyMyo: Update hand gesture classification experiment details#8MatteoFasulo wants to merge 24 commits into
MatteoFasulo wants to merge 24 commits into
Conversation
There was a problem hiding this comment.
Pull request overview
This pull request updates the experimental details in the TinyMyo model documentation to reflect revised window sizes and sampling rates used for hand gesture classification and kinematic regression experiments.
Changes:
- Updated classification window sizes from 200 ms to 1 sec (DB5) and 1000 ms to 5 sec (EPN-612, UCI EMG)
- Added sampling rate information (200 Hz) for classification datasets
- Updated regression window sizes from 200 ms/1000 ms to 100 ms/500 ms for Ninapro DB8
- Corrected dataset reference from "NinaPro EPN612 pipeline" to "EPN612 dataset"
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
MatteoFasulo
force-pushed
the
patch-1
branch
3 times, most recently
from
April 17, 2026 16:03
1a0d63e to
636835b
Compare
Updated dataset information and performance metrics for hand kinematic regression.
…s avoids namespace issues with third party libraries (e.g., datasets from HuggingFace)
…lity with old pytorch versions (<2.5)
…ory optimization for DDP - finetuning dataset preload in RAM option and documentation - regression task support for EMG finetuning - additional dataset configurations (log dir, individual h5 file)
…ers, EMG finetuning dataset handling, and improved training script logging
…iency and logging
…, and sequence length
…ssues with huggingface datasets)
- Added label smoothing and classification type options to the fine-tuning YAML. - Updated input normalization settings for pretraining and fine-tuning. - Revised documentation to clarify preprocessing and fine-tuning details. - Improved error handling for classification type in the finetune task.
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
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.
Significant improvements to the EMG pretraining and finetuning data pipeline, focusing on efficiency, configuration flexibility, and dataset handling. The most important changes include a complete rewrite of the
EMGPretrainDatasetto use shared memory for high-performance data loading, simplification of the finetuning dataset, and updates to experiment configuration files for improved training and logging. Below are the key changes grouped by theme:Data Loading and Dataset Refactoring
datasets/emg_pretrain_dataset.py: Completely refactoredEMGPretrainDatasetto use POSIX shared memory for loading HDF5 data, enabling efficient parallel data access and minimizing memory duplication across workers. Added support for min-max scaling and channel padding. The dataset now expects a single HDF5 file with group structure and loads all groups into a shared RAM block on initialization.datasets/emg_finetune_dataset.py: SimplifiedEMGDatasetby removing the cache and lazy file handle logic. The dataset now loads the entire HDF5 file into memory on initialization for faster access and supports both regression and classification modes.Configuration and Experiment Management
config/data_module/emg_pretrain_data_module.yaml: Updated dataset definitions to use explicit train/val/test splits with new HDF5 file paths, and added support for channel padding where required.config/experiment/TinyMyo_pretrain.yaml: Increased pretraining batch size to 512, reduced epochs to 30, enabled bf16 precision, and added WandB logging configuration. Updated optimizer and scheduler settings for improved training dynamics.config/experiment/TinyMyo_finetune.yaml: Reduced number of workers, switched output logging to${env:LOG_DIR}, updated model and trainer parameters, and added WandB logging configuration.Changed details about the experiments performed.