Current Work :

• Created a Streamlit App to automate populating Excel files using raw data from the sensor.
• Visualized and performed preliminary analysis
• Performed Analysis on Sensor data which aided in finding bugs.
• The bug meant that the sensor was sometimes incorrectly detecting posture or unable to detect any posture at all in infants
• Changed the position of the sensor based on the analysis of data using the streamlit app for better detection of posture.

Updates

• Modified the code to assign the Overall Class correctly.
• Researched on vectorizing operations and identifying bottlenecks to optimize code for large files
• Optimized code.
• Encountered incorrect encoding,mixed datatypes and kernel dying issues.
• Handled the bugs. Started preprocessing the videos for posture detection
• Added a jupter notebook file that can be run locally on lab computers where running the web app is not possible
• Added input functions for easy navigation in the jupyter notebook
• Refactored functions. Tested code for files with 10M+ rows.
• Updated script to handle large files on streamlit app, added config file.
• Added Docker image file
• Added visualizations for contiguous rows to debug sensor data and Blocks page for better viewing experience.
• Added Option to filter Block plots.
• Added the option to download the processed file.
• Added file structure for documentation and comments to understand the code.
• Added new plots and a new page for visualizing the change in position.
• Tested new plots for validating the sensor and toy
• Added data validation script
• Added some additional data validation checks. Performed statistical tests for difference in means.
• Added script for new visualizations
• Added code for some new visualizations based on the needs.
• Conducted Repeated measures ANOVA for casual inference.
• Added code to debug non-wear time

Future Work:

• Working on a deep learning model to correctly detect the posture of infants, thereby reducing manual work and labeling the dataset using Datavyu.
• Deployed initial model for posture detection. Received feedback on detection.
• Prototyping on the new fine-tuned version; the previous model incorrectly classifies some classes more than others.
• Working on Q&A Chatbot for interacting with the dataset for Ad-hoc analysis
• Working on denoising data using Low Pass Filter and experimenting with threshold for non-wear time.