Debug HPC resources Features
Assuming ranked dataset compute metrics: Header: JobId TimeToFirstFailure S10-NAPFD... Inclusiveness-S10 Inclusiveness-S20 ... SelectionExecutionTime-S10 ...
Metrics obtained per ranklib dataset, per model trained:
- Time to first failure Per selection (10%, 20%, ... 90%, 100%)
- Inclusiveness
- Selection execution time
- nAPFD
What visualizations to report?
- Candlestick plot with nAPFD comparing methods (from best results? from several ranklib datasets)
- Candlestick plot with time to first failure comparing methods (from best results? from several ranklib datasets)
- safe methods w/ selection execution time
- Collect mean time for each task on the test dependencies chain
- Consider a score related to build time for comparison functions on pairwise algs
- Preselection for ensuring running e.g. new test cases
- q-learning
- Comparison with existing methods
Argue, why is it needed on the project to have automatic selection, and not relying on dev knowledge