hdeeprm.cmd module¶
Command line scripts for managing HDeepRM experiments.
-
hdeeprm.cmd.clean() → None[source]¶ Utility for cleaning current directory outcomes.
WARNING: it also erases the “platform.xml”, “workload.json”, “res_hierarchy.pkl” and “job_limits.pkl” files, use
-sor--softto avoid this behaviour. Its purpose is to facilitate quick experimenting and debugging.- Command line arguments:
soft- (Optional) Cleans only logs, does not remove platform or workload definitions.
-
hdeeprm.cmd.jobstat() → None[source]¶ Utility for obtaining statistics about jobs in the workload trace.
It can calculate several statistics by parsing the workload.json file.
- Command line arguments:
workload- workload.json with the trace.datafield- Job data field from which to obtain information. One of req_time, size, mem or mem_bw.statistic- Statistic to be calculated. One of min, max, mean, median, p95 or p99.
-
hdeeprm.cmd.launch() → None[source]¶ Utility for launching HDeepRM experiments.
It takes care of creating the Platform XML file, the Workload JSON file and the Resource Hierarcy. It also runs both Batsim and PyBatsim.
- Command line arguments:
options_file- Options file in JSON.customworkload- (Optional) Path to the custom workload in case one is used.agent(Optional) File with the learning agent definition.inmodel- (Optional) Path for previous model loading.outmodel- (Optional) Path for saving new model.nbruns- (Optional) Number of train runs for the learning agent.
-
hdeeprm.cmd.metrics() → None[source]¶ Utility for comparing metrics between simulation runs.
Plots a grid with different metrics resulting from out_schedule.csv.
- Command line arguments:
res1- out_schedule.csv file from run 1.res2- out_schedule.csv file from run 2.
-
hdeeprm.cmd.visual() → None[source]¶ Utility for analysing stats from HDeepRM outcomes.
It utilizes Evalys for plotting useful information from output files. Supports queue_size, utilization, lifecycle, gantt, gantt_no_label, core_bubbles, mem_bubbles, mem_bw_overutilization, losses, rewards, action_preferences.
- Command line arguments:
visualization- Type of visualization to be printed.save- (Optional) Save the plot in the provided file path.