#!/usr/bin/env python3
"""Base script to run the C3 code from a main config file."""
import logging
import os
import hjson
import argparse
import c3.utils.tf_utils as tf_utils
import tensorflow as tf
from c3.c3objs import hjson_decode
from c3.parametermap import ParameterMap
from c3.experiment import Experiment
from c3.model import Model
from c3.generator.generator import Generator
from c3.optimizers.optimalcontrol import OptimalControl
from c3.optimizers.calibration import Calibration
from c3.optimizers.modellearning import ModelLearning
from c3.optimizers.sensitivity import Sensitivity
logging.getLogger("tensorflow").disabled = True
# flake8: noqa: C901
[docs]def run_cfg(cfg, opt_config_filename, debug=False):
"""Execute an optimization problem described in the cfg file.
Parameters
----------
cfg : Dict[str, Union[str, int, float]]
Configuration file containing optimization options and information needed to completely
setup the system and optimization problem.
debug : bool, optional
Skip running the actual optimization, by default False
"""
optim_type = cfg.pop("optim_type")
optim_lib = {
"C1": OptimalControl,
"C2": Calibration,
"C3": ModelLearning,
"C3_confirm": ModelLearning,
"confirm": ModelLearning,
"SET": Sensitivity,
}
if not optim_type in optim_lib:
raise Exception("C3:ERROR:Unknown optimization type specified.")
tf_utils.tf_setup()
with tf.device("/CPU:0"):
model = None
gen = None
exp = None
prop_meth = cfg.pop("propagation_method", None)
if "model" in cfg:
model = Model()
model.read_config(cfg.pop("model"))
if "generator" in cfg:
gen = Generator()
gen.read_config(cfg.pop("generator"))
if "instructions" in cfg:
pmap = ParameterMap(model=model, generator=gen)
pmap.read_config(cfg.pop("instructions"))
exp = Experiment(pmap, prop_method=prop_meth)
if "exp_cfg" in cfg:
exp = Experiment(prop_method=prop_meth)
exp.read_config(cfg.pop("exp_cfg"))
if exp is None:
print("C3:STATUS: No instructions specified. Performing quick setup.")
exp = Experiment(prop_method=prop_meth)
exp.quick_setup(cfg)
exp.set_opt_gates(cfg.pop("opt_gates", None))
if "gateset_opt_map" in cfg:
exp.pmap.set_opt_map(
[[tuple(par) for par in pset] for pset in cfg.pop("gateset_opt_map")]
)
if "exp_opt_map" in cfg:
exp.pmap.set_opt_map(
[[tuple(par) for par in pset] for pset in cfg.pop("exp_opt_map")]
)
opt = optim_lib[optim_type](**cfg, pmap=exp.pmap)
opt.set_exp(exp)
opt.set_created_by(opt_config_filename)
if "initial_point" in cfg:
initial_points = cfg["initial_point"]
if isinstance(initial_points, str):
initial_points = [initial_points]
elif isinstance(initial_points, list):
pass
else:
raise Warning("initial_point has to be a path or a list of paths.")
for init_point in initial_points:
try:
opt.load_best(init_point)
print(
"C3:STATUS:Loading initial point from : "
f"{os.path.abspath(init_point)}"
)
except FileNotFoundError as fnfe:
raise Exception(
f"C3:ERROR:No initial point found at "
f"{os.path.abspath(init_point)}. "
) from fnfe
if optim_type == "C1":
if "adjust_exp" in cfg:
try:
adjust_exp = cfg["adjust_exp"]
opt.load_model_parameters(adjust_exp)
print(
"C3:STATUS:Loading experimental values from : "
f"{os.path.abspath(adjust_exp)}"
)
except FileNotFoundError as fnfe:
raise Exception(
f"C3:ERROR:No experimental values found at "
f"{os.path.abspath(adjust_exp)} "
"Continuing with default."
) from fnfe
if not debug:
opt.run()
if __name__ == "__main__":
os.nice(5) # keep responsiveness when we overcommit memory
tf.compat.v1.logging.set_verbosity(tf.compat.v1.logging.ERROR)
parser = argparse.ArgumentParser()
parser.add_argument("master_config")
args = parser.parse_args()
opt_config = args.master_config
with open(opt_config, "r") as cfg_file:
try:
cfg = hjson.load(cfg_file, object_pairs_hook=hjson_decode)
except hjson.decoder.HjsonDecodeError:
raise Exception(f"Config {opt_config} is invalid.")
run_cfg(cfg, opt_config)