def pick_parameters(args, trial, counter):
margs = copy.deepcopy(args)
margs.modelname = "{}_{}".format(args.paramscanstudyname, counter)
margs.outerlayers = trial.suggest_categorical("outerlayers",
[4, 8, 12, 16, 20, 24])
margs.innerlayers = trial.suggest_int("innerlayers", 3, 10)
margs.linlayers = trial.suggest_int("linlayers", 1, 3)
margs.linchannelfactor = trial.suggest_int("linchannelfactor", 1, 2)
margs.lineartransform = trial.suggest_categorical(
"lineartransform", ["permutation", "lu", "svd"])
margs.dropout = trial.suggest_categorical("dropout", [0.0, 0.20])
margs.splinerange = trial.suggest_categorical("splinerange",
[6.0, 8.0, 10.0])
margs.splinebins = trial.suggest_int("splinebins", 3, 20)
margs.batchnorm = trial.suggest_categorical("batchnorm", [False, True])
margs.actnorm = trial.suggest_categorical("actnorm", [False, True])
margs.batchsize = trial.suggest_categorical("batchsize", [
50,
])
margs.msefactor = trial.suggest_loguniform("msefactor", 1.0e-3, 10.0)
margs.uvl2reg = trial.suggest_loguniform("uvl2reg", 1.0e-9, 0.1)
margs.weightdecay = trial.suggest_loguniform("weightdecay", 1.0e-9, 0.1)
margs.clip = trial.suggest_loguniform("clip", 1.0, 100.0)
create_modelname(margs)
return margs
def pick_parameters(args, trial, counter):
margs = copy.deepcopy(args)
margs.modelname = "paramscan_{}".format(counter)
margs.outerlayers = trial.suggest_categorical("outerlayers", [3, 5, 10])
margs.innerlayers = trial.suggest_categorical("innerlayers", [3, 5, 10])
margs.outercouplingmlp = trial.suggest_categorical("outercouplingmlp",
[False, True])
margs.outercouplinglayers = trial.suggest_categorical(
"outercouplinglayers", [1, 2, 3])
margs.dropout = trial.suggest_categorical("dropout", [0.0, 0.20])
margs.splinerange = trial.suggest_categorical("splinerange",
[5.0, 6.0, 8.0])
margs.splinebins = trial.suggest_categorical("splinebins", [5, 10, 20])
margs.batchsize = trial.suggest_categorical("batchsize",
[50, 100, 200, 500])
margs.lr = trial.suggest_loguniform("lr", 1.0e-5, 1.0e-2)
margs.msefactor = trial.suggest_loguniform("msefactor", 1.0e2, 1.0e4)
margs.weightdecay = trial.suggest_loguniform("weightdecay", 1.0e-8, 1.0e-4)
margs.clip = trial.suggest_loguniform("clip", 1.0, 100.0)
create_modelname(margs)
return margs
if __name__ == "__main__":
# Parse args
args = parse_args()
logging.basicConfig(format="%(asctime)-5.5s %(name)-20.20s %(levelname)-7.7s %(message)s", datefmt="%H:%M", level=logging.DEBUG if args.debug else logging.INFO)
# Silence PIL
for key in logging.Logger.manager.loggerDict:
if "PIL" in key:
logging.getLogger(key).setLevel(logging.WARNING)
logger.info("Hi!")
logger.debug("Starting evaluate.py with arguments %s", args)
# Model name
if args.truth:
create_modelname(args)
logger.info("Evaluating simulator truth")
else:
create_modelname(args)
logger.info("Evaluating model %s", args.modelname)
# Bug fix related to some num_workers > 1 and CUDA. Bad things happen otherwise!
torch.multiprocessing.set_start_method("spawn", force=True)
# Data set
simulator = load_simulator(args)
# Load model
if not args.truth:
model = create_model(args, simulator=simulator)
model.load_state_dict(torch.load(create_filename("model", None, args), map_location=torch.device("cpu")))
