Master's thesis in Computer Science at the University of Brasília (PPGI).
The objective is to perform drift detection using meta learning by predicting the performance of a base model (regression or classification) with unknown target.
To Do:
- Offline stage MetaLearner implementation
- Online stage MetaLearner implementation
- Meta model incremental learning
- Statistical meta features implementation
- Clustering meta features implementation
- PSI implementation
- Domain classifier implementation
- Meta/base model hyperparameter tuning
- TimeSeries cross validation usage
- Different base model metrics (meta labels) experiments
- Target delay experiments
- Different base model experiments
- Meta base dimensionality reduction experiments
- Other drift metrics implementation
- Other databases experiments
- Measure elapsed time
- Meta feature selection experiments
- Baseline
- Evaluation metrics for meta model vs baseline
- Drift alert definition
- Drift alert configuration
- Code documentation
- Readme
- Experiments notebooks adjustments to fit new MetaLearner structure