def __init__(self, options):
#Data structures for storing info
self.user_job_last3 = {}
self.user_job_last2 = {}
self.user_job_last1 = {}
self.job_x = {}
#machine learning thing
m = LinearModel(self.n_features)
if options["loss"] == "squaredloss":
from valopt.losses.squared_loss import SquaredLoss
l = SquaredLoss(m)
elif options["loss"] == "absloss":
from valopt.losses.abs_loss import AbsLoss
l = AbsLoss(m)
elif options["loss"] == "weightedsquaredloss":
from valopt.losses.weighted_squared_loss import WeightedSquaredLoss
l = WeightedSquaredLoss(m)
else:
raise ValueError(
"predictor config error: no valid loss specified.")
if "max_runtime" in options.keys():
self.max_runtime = options["max_runtime"]
else:
self.max_runtime = False
self.model = NAG(m, l, options["eta"], verbose=False)
def __init__(self, options):
#Data structures for storing info
self.user_job_last3 = {}
self.user_job_last2 = {}
self.user_job_last1 = {}
self.user_sum_runtimes = {}
self.user_sum_cores = {}
self.user_n_jobs = {}
self.job_x = {}
self.user_last_ending = {}
#Statistics oriented.
self.last_loss = 0
if options["scheduler"]["predictor"]["quadratic"]:
print("Using predictor with quadratic features")
self.quadratic = True
if options["scheduler"]["predictor"]["cubic"]:
self.cubic = True
self.n_features = int(3 * self.n_features +
kPn(self.n_features, 2) +
kPn(self.n_features, 3))
else:
self.n_features = int(3 * self.n_features +
2 * kPn(self.n_features, 2))
self.cubic = False
else:
self.cubic = False
self.quadratic = False
#machine learning thing
m = LinearModel(self.n_features)
if "max_runtime" in options["scheduler"]["predictor"].keys():
self.max_runtime = options["scheduler"]["predictor"]["max_runtime"]
else:
self.max_runtime = False
if options["scheduler"]["predictor"]["loss"] == "squaredloss":
from valopt.losses.squared_loss import SquaredLoss
l = SquaredLoss(m, maxloss=self.max_runtime)
elif options["scheduler"]["predictor"]["loss"] == "composite":
from valopt.losses.composite import CompositeLoss
if options["scheduler"]["predictor"]["leftside"] == "abs":
from valopt.losses.losscurves.abs import Abscurve
leftside = Abscurve(
m, options["scheduler"]["predictor"]["leftparam"])
elif options["scheduler"]["predictor"]["leftside"] == "square":
from valopt.losses.losscurves.square import Squarecurve
leftside = Squarecurve(
m, options["scheduler"]["predictor"]["leftparam"])
elif options["scheduler"]["predictor"]["leftside"] == "exp":
from valopt.losses.losscurves.exp import Expcurve
leftside = Expcurve(
m, options["scheduler"]["predictor"]["leftparam"])
else:
raise ValueError(
"predictor config error: no leftside specified")
if options["scheduler"]["predictor"]["rightside"] == "abs":
from valopt.losses.losscurves.abs import Abscurve
rightside = Abscurve(
m, options["scheduler"]["predictor"]["rightparam"])
elif options["scheduler"]["predictor"]["rightside"] == "square":
from valopt.losses.losscurves.square import Squarecurve
rightside = Squarecurve(
m, options["scheduler"]["predictor"]["rightparam"])
elif options["scheduler"]["predictor"]["rightside"] == "exp":
from valopt.losses.losscurves.exp import Expcurve
rightside = Expcurve(
m, options["scheduler"]["predictor"]["rightparam"])
else:
raise ValueError(
"predictor config error: no rightside specified")
l = CompositeLoss(m, rightside, leftside,
options["scheduler"]["predictor"]["threshold"])
else:
raise ValueError(
"predictor config error: no valid loss specified.")
if "lambda" in options["scheduler"]["predictor"].keys():
if options["scheduler"]["predictor"]["regularization"] == "l1":
from valopt.losses.regularizations.l1 import L1
from valopt.losses.regularized_loss import RegularizedLoss
l = RegularizedLoss(
m, l, L1(), options["scheduler"]["predictor"]["lambda"])
elif options["scheduler"]["predictor"]["regularization"] == "l2":
from valopt.losses.regularizations.l2 import L2
from valopt.losses.regularized_loss import RegularizedLoss
l = RegularizedLoss(
m, l, L2(), options["scheduler"]["predictor"]["lambda"])
else:
raise ValueError(
"predictor config error: lambda present and no valid regularizer specified."
)
if options["scheduler"]["predictor"]["gd"] == "NAG":
from valopt.algos.nag import NAG
self.model = NAG(m,
l,
options["scheduler"]["predictor"]["eta"],
verbose=False)
elif options["scheduler"]["predictor"]["gd"] == "sNAG":
from valopt.algos.snag import sNAG
self.model = sNAG(m,
l,
options["scheduler"]["predictor"]["eta"],
verbose=False)
if not options["scheduler"]["predictor"]["weight"]:
wstr = compile("1", "<string>", "eval")
else:
wstr = compile(options["scheduler"]["predictor"]["weight"],
"<string>", "eval")
def weight(job):
m = float(job.num_required_processors)
r = float(job.actual_run_time)
#log=np.log
log = math.log
return eval(wstr)
self.weight = weight
