Child Mind Institute - Detect Sleep States Detect sleep onset and wake from wrist-worn accelerometer data
Project based on dataset from competition - https://www.kaggle.com/competitions/child-mind-institute-detect-sleep-states
- 1. Train series were taken from parquet file in chunks and separated into csv files with following pattern "serie_id".csv
- 2. Train events data were changed into segmenatation mask (0 - awake/wakeup, 1 - onset/sleep) which was included with train series.
- 3. If sleep event in train event was not recognized - segmentation mask is assumed as unkonwn (-1) which will be ignored during training
- 4. First idea is to use U-net model with Conv1D. First learnings made
- 5. Train output mask needs to be converted into events
- 6. Adding new features (positional encoding, serie converted to one day - paddding with 0 for smaller time series)
- 7. Update of the U-net model with multi class outputs
- 8. Check changes, error analysis
- 9. Further exploration of possible solutions (Transformer architecture?)