ML_train_luigi.cfg

[core]
no_lock = True
log_level = DEBUG

[worker]
keep_alive: True
ping_interval: 20
wait_interval: 20
max_reschedules: 1

[scheduler]
retry_count: 0

[DEFAULT]
name = ML_train
ENV_NAME = Base
; grid storage protocol and path usable from submitting machine and worker nodes of cluster
; job in- and output will be stored in $wlcg_path under subdirectory of analysis $name
wlcg_path = root://cmsdcache-kit-disk.gridka.de//store/user/${USER}/LAW_storage
; default htcondor job submission configuration (modifiable for each task)
htcondor_accounting_group = cms.higgs
htcondor_remote_job = True
htcondor_request_cpus = 1
; for all cores in total
htcondor_universe = docker
;image without GPU libraries
# htcondor_docker_image = mschnepf/slc7-condocker:latest
; create log files in htcondor jobs
transfer_logs = True
; set local scheduler
local_scheduler = True
; set tolerance for workflow success with failed branches
tolerance = 0.00
acceptance = 1.00
; submit only missing htcondor workflow branches (should always be true)
only_missing = True

; bootstrap file to be sourced at beginning of htcondor jobs (relative PATH to framework.py)
bootstrap_file = setup_law_remote.sh


[CreateTrainingDataShard]
htcondor_request_cpus = 1
htcondor_walltime = 36000
htcondor_request_memory = 4000
htcondor_request_disk = 2000000
htcondor_requirements = (TARGET.ProvidesEKPResources==True) && (TARGET.ProvidesIO==True)
additional_files = ["ml_configs", "sm-htt-analysis"]


[RunTraining]
ENV_NAME = ML_LAW
htcondor_request_cpus = 2
htcondor_request_gpus = 1
htcondor_walltime = 3600
htcondor_request_memory = 4000
htcondor_requirements = ( TARGET.CloudSite =?= "topas" ) 
#&& (Machine =?= "f03-001-179-e.gridka.de")
htcondor_request_disk = 1000000
additional_files = ["ml_configs", "sm-htt-analysis"]

[RunTesting]
ENV_NAME = ML_LAW
htcondor_request_cpus = 1
htcondor_request_gpus = 1
htcondor_walltime = 3600
htcondor_request_memory = 10000
htcondor_request_disk = 10000000
htcondor_requirements = ( TARGET.CloudSite =?= "topas" ) 
#&& (Machine =?= "f03-001-179-e.gridka.de")
additional_files = ["ml_configs", "sm-htt-analysis"]
[RunAllAnalysisTrainings]