def arg_dpc_problem(prefix=''):
"""
Command line parser for DPC problem definition arguments
:param prefix: (str) Optional prefix for command line arguments to resolve naming conflicts when multiple parsers
are bundled as parents.
:return: (arg.ArgParse) A command line parser
"""
parser = arg.ArgParser(prefix=prefix, add_help=False)
gp = parser.group("DPC")
gp.add("-nsteps", type=int, default=10,
help="prediction horizon.") # tuned values: 1, 2
gp.add("-nx_hidden", type=int, default=20,
help="Number of hidden states")
gp.add("-n_layers", type=int, default=4,
help="Number of hidden layers")
gp.add("-bias", action="store_true",
help="Whether to use bias in the neural network block component models.")
gp.add("-epochs", type=int, default=2000,
help='Number of training epochs')
gp.add("-lr", type=float, default=0.001,
help="Step size for gradient descent.")
gp.add("-patience", type=int, default=100,
help="How many epochs to allow for no improvement in eval metric before early stopping.")
gp.add("-warmup", type=int, default=10,
help="Number of epochs to wait before enacting early stopping policy.")
return parser
def arg_dpc_problem(prefix=''):
"""
Command line parser for DPC problem definition arguments
:param prefix: (str) Optional prefix for command line arguments to resolve naming conflicts when multiple parsers
are bundled as parents.
:return: (arg.ArgParse) A command line parser
"""
parser = arg.ArgParser(prefix=prefix, add_help=False)
gp = parser.group("DPC")
gp.add("-nsteps", type=int, default=1, help="prediction horizon.")
gp.add("-Qx", type=float, default=1.0,
help="state weight.") # tuned value: 1.0
gp.add("-Qu", type=float, default=1.0, help="control action weight."
) # tuned value: 1.0, unstable for paper: 0.5
gp.add("-Qn", type=float, default=1.0, help="terminal penalty weight."
) # tuned value: 1.0, unstable for paper: 0.0
gp.add("-Q_sub", type=float, default=0.0, help="regularization weight.")
gp.add("-Q_con_x",
type=float,
default=10.0,
help="state constraints penalty weight.") # tuned value: 10.0
gp.add("-Q_con_u",
type=float,
default=20.0,
help="Input constraints penalty weight.") # tuned value: 20.0
gp.add("-nx_hidden", type=int, default=20, help="Number of hidden states")
gp.add("-n_layers", type=int, default=4, help="Number of hidden layers")
gp.add(
"-bias",
action="store_true",
help="Whether to use bias in the neural network block component models."
)
gp.add("-norm",
nargs="+",
default=[],
choices=["U", "D", "Y", "X"],
help="List of sequences to max-min normalize")
gp.add("-data_seed",
type=int,
default=408,
help="Random seed used for simulated data")
gp.add("-epochs", type=int, default=1000, help='Number of training epochs')
gp.add("-lr",
type=float,
default=0.001,
help="Step size for gradient descent.")
gp.add(
"-patience",
type=int,
default=100,
help=
"How many epochs to allow for no improvement in eval metric before early stopping."
)
gp.add(
"-warmup",
type=int,
default=10,
help="Number of epochs to wait before enacting early stopping policy.")
return parser
import os
from neuromancer import arg
import argparse
from neuromancer import datasets
parser = argparse.ArgumentParser()
parser.add_argument('-gpu', type=int, default=None)
args = parser.parse_args()
if args.gpu is not None:
gpu = f'-gpu {args.gpu}'
p = arg.ArgParser(parents=[
arg.log(),
arg.opt(),
arg.data(),
arg.loss(),
arg.lin(),
arg.ssm()
])
options = {
k: v.choices
for k, v in p._option_string_actions.items()
if v.choices is not None and k != '-norm'
}
systems = list(datasets.systems.keys())
for i, (k, v) in enumerate(options.items()):
for j, opt in enumerate(v):
print(k, opt)
code = os.system(
f'python ../train_scripts/system_id.py -norm Y '
f'{k} {opt} -epochs 1 -nsteps 8 -verbosity 1 -nsim 128 -system {systems[(i*j) % len(systems)]}'
"Y_min": 0.7 * np.ones([nsim, ny]),
"U_max": np.ones([nsim, nu]),
"U_min": np.zeros([nsim, nu]),
"R": psl.Periodic(nx=ny, nsim=nsim, numPeriods=20, xmax=0.9, xmin=0.8)[:nsim, :]
# 'Y_ctrl_': psl.WhiteNoise(nx=ny, nsim=nsim, xmax=[1.0] * ny, xmin=[0.0] * ny)
})
# indices of controlled states, e.g. [0, 1, 3] out of 5 outputs
dataset.ctrl_outputs = args.controlled_outputs
return dataset
if __name__ == "__main__":
# for available systems in PSL library check: psl.systems.keys()
system = 'CSTR' # keyword of selected system
parser = arg.ArgParser(parents=[arg.log(), arg.log(prefix='sysid_'),
arg.opt(), arg.data(system=system),
arg.loss(), arg.lin(), arg.policy(),
arg.ctrl_loss(), arg.freeze()])
path = './test/CSTR_best_model.pth'
parser.add('-model_file', type=str, default=path,
help='Path to pytorch pickled model.')
args = parser.parse_args()
args.savedir = 'test_control'
log_constructor = MLFlowLogger if args.logger == 'mlflow' else BasicLogger
metrics = ["nstep_dev_loss", "loop_dev_loss", "best_loop_dev_loss",
"nstep_dev_ref_loss", "loop_dev_ref_loss"]
logger = log_constructor(args=args, savedir=args.savedir, verbosity=args.verbosity, stdout=metrics)
print({k: str(getattr(args, k)) for k in vars(args) if getattr(args, k)})
device = f"cuda:{args.gpu}" if args.gpu is not None else "cpu"
if umax is not None:
u_max = umax * np.ones([nstep + 1, umax.shape[0]])
ax[1].plot(u_max, 'k--', linewidth=2)
ax[1].set(xlabel='time')
ax[1].grid()
ax[1].set_xlim(0, nstep)
plt.tight_layout()
if save_path is not None:
plt.savefig(save_path + '/closed_loop_quadcopter_dpc.pdf')
if __name__ == "__main__":
"""
# # # Arguments
"""
parser = arg.ArgParser(parents=[arg.log(), arg_dpc_problem()])
args, grps = parser.parse_arg_groups()
args.bias = True
"""
# # # 3D quadcopter model
"""
# Discrete time model of a quadcopter
A = np.array(
[[1., 0., 0., 0., 0., 0., 0.1, 0., 0., 0., 0., 0.],
[0., 1., 0., 0., 0., 0., 0., 0.1, 0., 0., 0., 0.],
[0., 0., 1., 0., 0., 0., 0., 0., 0.1, 0., 0., 0.],
[0.0488, 0., 0., 1., 0., 0., 0.0016, 0., 0., 0.0992, 0., 0.],
[0., -0.0488, 0., 0., 1., 0., 0., -0.0016, 0., 0., 0.0992, 0.],
[0., 0., 0., 0., 0., 1., 0., 0., 0., 0., 0., 0.0992],
[0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0., 0.],
disturbances_max_influence_lb,
disturbances_max_influence_ub,
]
return objectives, constraints
if __name__ == "__main__":
# for available systems in PSL library check: psl.systems.keys()
# for available datasets in PSL library check: psl.datasets.keys()
system = 'aero' # keyword of selected system
parser = arg.ArgParser(parents=[
arg.log(),
arg.opt(),
arg.data(system=system),
arg.loss(),
arg.lin(),
arg.ssm()
])
grp = parser.group('OPTIMIZATION')
grp.add("-eval_metric",
type=str,
default="loop_dev_ref_loss",
help="Metric for model selection and early stopping.")
args, grps = parser.parse_arg_groups()
print({k: str(getattr(args, k)) for k in vars(args) if getattr(args, k)})
device = f"cuda:{args.gpu}" if args.gpu is not None else "cpu"
log_constructor = MLFlowLogger if args.logger == 'mlflow' else BasicLogger
