Releases: amosproj/amos2025ws03-rtdip-timeseries-forecasting
sprint14-release
final-project-release
sprint14-release-candidate
sprint13-release
Sprint Summary
Merge Process to Core Repository Initiated
- Merge process to core repo started this week
- Mkdocs updated and issues fixed for MAD anomaly detection
- Documentation links added for anomaly detection approaches
- Merge issue created in core repo
- Merge branch created in fork (feature/00949), which resolves all merge conflicts with core and only includes components relevant for the RTDIP SDK
- PR is under review
Presentation Preparation
- Demo slides uploaded
- Demo video uploaded
Environment Updates
- Fixed keras v3 incompatibility with transformers
- Changed prophet installation to pip
- Awaiting Shell feedback before proceeding
Important Note
- No new functionality implemented due to focus on merge and presentation preparation
sprint13-release-candidate
Sprint Summary
Merge Process to Core Repository Initiated
- Merge process to core repo started this week
- Mkdocs updated and issues fixed for MAD anomaly detection
- Documentation links added for anomaly detection approaches
- Merge issue created in core repo
- Merge branch created in fork (feature/00949), which resolves all merge conflicts with core and only includes components relevant for the RTDIP SDK
- PR is under review
Presentation Preparation
- Demo slides uploaded
- Demo video uploaded
Environment Updates
- Fixed keras v3 incompatibility with transformers
- Changed prophet installation to pip
- Awaiting Shell feedback before proceeding
Important Note
- No new functionality implemented due to focus on merge and presentation preparation
sprint12-release
This sprint focused on running, evaluating, and documenting models across datasets, with a strong emphasis on the 3W dataset and anomaly detection, while preparing the project for merge into the RTDIP core repository.
Completed Work
1. Make All Datasets Running with All Models and Evaluate All (#114)
Training and evaluation pipelines were executed across the Shell, SCADA, and 3W datasets using consistent data splits and model configurations.
Note:
At the time of this release, not all models have been successfully applied to all datasets:
- Some model–dataset combinations ( LSTM on specific datasets) are pending or failed due to dataset size,(Prophet was not suitable).
- These gaps are explicitly documented in the results tables and are not marked as completed runs.
2. Training, Forecasting, and Evaluation on 3W Dataset (#107)
Documented in the wiki.
3. Anomaly Detection on 3W Dataset (#117)
IQR-based anomaly detection was applied to the 3W dataset, evaluated against ground-truth fault labels using standard metrics, and analyzed through dedicated notebooks and visualizations.
4. Merge Preparation (#115)
Documentation, merge scope, excluded artifacts, and MkDocs pages were reviewed and prepared in alignment with RTDIP release guidelines and prior AMOS merge practices.
sprint12-release-candidate
This sprint focused on running, evaluating, and documenting models across datasets, with a strong emphasis on the 3W dataset and anomaly detection, while preparing the project for merge into the RTDIP core repository.
Completed Work
1. Make All Datasets Running with All Models and Evaluate All (#114)
Training and evaluation pipelines were executed across the Shell, SCADA, and 3W datasets using consistent data splits and model configurations.
Note:
At the time of this release, not all models have been successfully applied to all datasets:
- Some model–dataset combinations ( LSTM on specific datasets) are pending or failed due to dataset size,(Prophet was not suitable).
- These gaps are explicitly documented in the results tables and are not marked as completed runs.
2. Training, Forecasting, and Evaluation on 3W Dataset (#107)
Documented in the wiki.
3. Anomaly Detection on 3W Dataset (#117)
IQR-based anomaly detection was applied to the 3W dataset, evaluated against ground-truth fault labels using standard metrics, and analyzed through dedicated notebooks and visualizations.
4. Merge Preparation (#115)
Documentation, merge scope, excluded artifacts, and MkDocs pages were reviewed and prepared in alignment with RTDIP release guidelines and prior AMOS merge practices.
sprint11-release
Sprint Summary
Visualization Component for Timeseries Decomposition
- SDK visualization component added for decomposition
- All existing result plots in the project follow the standardized visualization rules
- Integrated into RTDIP
Assessment of Robustness and long time predictions
- Introduction of a training, testing and validation split
- Crossvalidation added into existing models
Preprocessing of 3W_Dataset
- Opensource 3W_Dataset has been preprocessed
- Missing or invalid values are handled appropriately
- Timestamps are standardized and sorted chronologically
- Dataset is ready for model usage
SCADA Decomposition
- Timeseries Decomposition has been applied to the SCADA Dataset
- The component can decompose a given time series into trend, seasonal, and residual components
- A visualization of the decomposition results is provided according to the implemented visualization pipeline
User and Build Documentation Refactor
- The wiki page has been updated to include information
- The contents of the page have been adjusted and trimmed
- The wiki now correctly documents how to build and run the AMOS additions to the SDK
sprint11-release-candidate
Sprint Summary
Visualization Component for Timeseries Decomposition
- SDK visualization component added for decomposition
- All existing result plots in the project follow the standardized visualization rules
- Integrated into RTDIP
Assessment of Robustness and long time predictions
- Introduction of a training, testing and validation split
- Crossvalidation added into existing models
Preprocessing of 3W_Dataset
- Opensource 3W_Dataset has been preprocessed
- Missing or invalid values are handled appropriately
- Timestamps are standardized and sorted chronologically
- Dataset is ready for model usage
SCADA Decomposition
- Timeseries Decomposition has been applied to the SCADA Dataset
- The component can decompose a given time series into trend, seasonal, and residual components
- A visualization of the decomposition results is provided according to the implemented visualization pipeline
User and Build Documentation Refactor
- The wiki page has been updated to include information
- The contents of the page have been adjusted and trimmed
- The wiki now correctly documents how to build and run the AMOS additions to the SDK
sprint10-release
Sprint Summary
Refactor of Pandas-only Components into a Spark Compatible version
- Pandas-only components refactored to use Spark (top-level must be Spark)
- Environment compatibility verified through testing
- Documentation updated to reflect infrastructure changes
New Minimal CI Pipeline
- Minimal CI pipeline established to run tests on devcontainer in single Python version
- Core repos CI workflows ignored for now
Visualization Component for Anomaly Detection added
- Standard visualization questions for timeseries anomaly detection defined
- Appropriate visualization types mapped to each analytical question
- Reusable visualization guideline document created
- All existing result plots updated to follow standardized visualization rules
- Visualization components integrated into RTDIP SDK
- Wiki documentation updated with guidance on selecting and justifying visualizations
Fixed broken testcases (decomposition)
- Spark tests now running on session instead module level
Integration of Timeseries Forecasting Results Visualization into the SDK
- Timeseries forecasting results visualization integrated into SDK
- Pipeline now displays visualizations when executed
- Model evaluation implemented using agreed error metrics with visualization
- New components incorporated into SDK
Optimization and Improvement of Anomaly Detection Approach
- Anomaly detection improved
- Component validated on Shell dataset
Software Architecture update
- Software architecture diagram updated to include all recent implementations