Python Setup

cd wearable_har_v2

# Create virtual environment
uv venv .venv --python 3.12

# Install all dependencies
uv pip install -r requirements.txt

# Verify
.venv/bin/python -c "
from host.protocol import crc32, ACTIVITY_LABELS, PKT_STM32_SIZE
assert crc32(b'123456789') == 0xCBF43926, 'CRC32 check failed'
assert PKT_STM32_SIZE == 214, f'expected 214, got {PKT_STM32_SIZE}'
assert len(ACTIVITY_LABELS) == 11, f'expected 11 labels, got {len(ACTIVITY_LABELS)}'
print('OK — protocol module loaded, CRC verified,', len(ACTIVITY_LABELS), 'activity classes')
"

Dependencies

Package	Purpose
numpy	Numeric array operations
torch	PyTorch: model training + TorchScript inference
pyserial	Serial port access (CDC)
h5py	HDF5 recording format
streamlit	Dashboard UI
scikit-learn	LOSO cross-validation splits
mkdocs-material	Documentation site (dev dependency)
wireviz	Wiring diagram generation (dev dependency)
diagrams	Architecture diagram generation (dev dependency)

Module overview

Module	Import	Purpose
protocol.py	from host.protocol import ...	Packet constants, CRC32, labels
serial_reader.py	from host.serial_reader import SerialReader	Magic sync, CRC validate, parse
feature_extraction.py	from host.feature_extraction import FeatureExtractor	2 s windows → 328-dim vectors
data_recorder.py	from host.data_recorder import DataRecorder	HDF5 recording
train_model.py	python -m host.train_model ...	BiLSTM-CNN training + export
live_inference.py	from host.live_inference import LiveClassifier	TorchScript inference
dashboard.py	streamlit run host/dashboard.py	Dashboard UI

Quick verification (no hardware)

# Feature extractor correctness
.venv/bin/python -c "
from host.feature_extraction import FeatureExtractor
from host.serial_reader import SensorFrame, IMUData

ext = FeatureExtractor()
f = SensorFrame()
for i in range(6):
    f.imus[i] = IMUData(q_w=1.0, az=9.81)
for _ in range(199):
    ext.add_frame(f)
feats = ext.add_frame(f)  # fills 200th frame → 2s window
print(f'Feature dim: {feats.shape[0]} (expected 328)')
assert feats.shape == (328,), f'FAIL: {feats.shape}'
print('OK')
"

# Train on synthetic data (2 epochs, smoke test)
.venv/bin/python -m host.train_model --synthetic --epochs 2 --output /tmp/test_model.pt

Directory layout after setup

wearable_har_v2/
├── .venv/
├── host/
│   ├── protocol.py
│   ├── serial_reader.py
│   ├── feature_extraction.py
│   ├── data_recorder.py
│   ├── train_model.py
│   ├── live_inference.py
│   └── dashboard.py
├── models/           # created after training
├── recordings/       # created after recording
├── tests/
└── docs/