- Updated logging statements and format - Filter directory list in recursive mode for more accurate progress bar - Added metrics parser module for parsing JSON metrics output - Updated plot appearance - Replaced deprecated pandas append with concatenate - Used 20-color cycler for plots - Added tests for metrics report argument parser - Added violation checks for sample efficiency metric (positive slope and STE threshold) - Implemented new metrics: mean training and evaluation performance - Ensured deterministic order of processing in recursive mode for consistent task coloring - Save plots in a different subdirectory within results directory - Re-organized example logs and results - Updated README and documentation
3.0.0
- Enforced black formatting - Created git workflows for linting, building, and running examples - Updated README - Updated license year to include 2022 - Removed program-specific evaluation package - Configured logging for tqdm compatibility
2.15.0
- Handled additional edge cases for storing STEs - Filtered STEs by “complete” experiences before storing - Modified normalization function to accept constant performance and normalize to 0 - Replaced majority of print statements with logging library - Updated reference to lifelong learning metrics specification document on arxiv - Added additional documentation on data preprocessing
2.14.0
- Implemented supplemental metrics subpackage for ABC transfer - Designed task coloring in plots to be consistent across runs when computing recursive metrics or evaluation - Optimized code by limiting groupby calls - Refactored episodes to experiences in variable names and plot labels - Updated example evaluation results
2.13.0
- Filtered out wake EBs in plotting methods - Sanitized metrics JSON parser example notebook - Minor STE plot appearance changes - Updated README and copyright statement
2.12.0
- Implemented class for parsing metrics JSON files (WIP) - Added more data to the metrics JSON output including: - List of relative performance values with multiple runs of STE - List of intermediate results used to calculate sample efficiency - List of sample efficiency values with multiple runs of STE - Added additional output file for regime-level metrics containing terminal performance for each regime