Changelog
- Corrected classical visualizations.
- Record agent initial probabilities into JSON file. Used for visualizations.
- New losses, rewards and action_preferences visualizations.
- Added learning experiment.
- Refactored environment, now main class is Environment.
- User-defined action subsets and observation types.
- Added thesis experiments.
- New hdeeprm-jobstat utility for exploring workload trace statistics.
- Tracking of memory over-utilization periods.
- New visualizations for exploring over-utilizations.
- Fixed agent bugs.
- New hdeeprm-metrics utility for plotting a grid comparing metrics
from two different simulation runs.
- Renamed hdeeprm-stat to hdeeprm-visual.
- New hdeeprm-clean utility for fast testing. Erases all files in current
directory except for options.json.
- Added agent and platform examples.
- New hdeeprm-stat utility for plotting information from simulation outcomes.
- Automatic generation of job limits from the workload file.
- Automatic reference speed calculation based on average FLOPs per core
in the system.
