[GSOC-25] Update and Simplification of Gaze Tracking Prediction Pipeline

[jvJUCA]

This PR includes several improvements and simplifications to the gaze tracking prediction service:

• Flask route updates

  • /api/session/batch_predict now generates a temporary CSV for iris data
  • Uses the simplified prediction function
  • Returns a flattened results list including timestamp, precision, and accuracy

• Docker

  • Adjustments to Dockerfile and .dockerignore
  • Uses JSON array in CMD for better security
  • Excludes unnecessary files from the container

• Dependencies

  • requirements.txt updated for Python 3.11 and scikit-learn 1.6.1 compatibility

Key Changes

app/services/gaze_tracker.py

• Added predict_new_data_simple
• Removed filters that discarded negative values
• Simplified training and prediction pipeline

app/routes/session.py

• /api/session/batch_predict updated to use the simplified function
• Temporary CSV created for batch predictions
• Result formatted as a flat list of predictions

Docker

• .dockerignore updated to exclude unnecessary files
• Dockerfile updated to install dependencies, clean build tools, and expose port 8080

Benefits

• Cleaner and more reliable production code
• Avoids empty arrays caused by negative-value filters
• Keeps compatibility with front-end and RuxaiLab expected formats
• Improved Docker build and runtime

Testing

• Manual tests with batch prediction using calibration CSV and new data
• Verified precision and accuracy metrics
• Tested webhook sending to RuxaiLab successfully