def _check_error_dir(output_dir, save, standalone):
""" Input consistency check
"""
if save and output_dir is None:
print("Specify output_dir to save parameter files")
raise ValueError
if standalone and save:
fu.assert_dir(output_dir, should_be_empty=True)
def generate_paths(self):
param_paths=[HUMAN_WEBOT_NMM_PATH+"config/2D_ind"+str(i)+
"/gaits/current" for i in range(1,self.nb_ind+1)]
log_paths=[HUMAN_WEBOT_NMM_PATH+"log/log_ind"+str(i)
for i in range(1,self.nb_ind+1)]
worlds_paths=[HUMAN_WEBOT_NMM_PATH+"webots/worlds/tmp_2D_noObstacle_GA_2Dind"+str(i)+
".wbt" for i in range(1,self.nb_ind+1)]
for p_path,l_path,w_path in zip(param_paths, log_paths,worlds_paths):
fu.assert_dir(l_path)
s1=fu.assert_file_exists(p_path, should_exist=True)
s2=fu.assert_file_exists(w_path, should_exist=True)
if not s1 or not s2:
print(s1,p_path)
print(s2,w_path)
raise RuntimeError("Missing worlds or parameter path, see Readme webots files generation")
self.param_path=param_paths
self.log_path=log_paths
self.world_path=worlds_paths
def gen_all_file(param_file,
output_dir=None,
standalone=False,
nested=False,
save=False):
""" Generates all pairs {param_name:value} required by param_file
(see data/sensi_template,yaml for reference)
"""
_check_error_dir(output_dir, save, standalone)
all_params = yaml.load(open(param_file, 'r'), Loader=yaml.FullLoader)
file_counter = 1
files = []
for name, parameters in all_params.items():
if LOG_LEVEL <= LOG_DEBUG:
print("[DEBUG] Reading:\t Name:", name, "\nParameters:",
parameters)
if nested and save:
rdir = os.path.join(output_dir, name)
fu.assert_dir(rdir, should_be_empty=True)
elif save:
rdir = output_dir
else:
rdir = None
if parameters[0] == "single":
files.append(
gen_single(parameters[1:], name, output_dir=rdir, save=save))
elif parameters[0] == "range":
files.append(
gen_range(parameters[1:], name, output_dir=rdir, save=save))
elif parameters[0] == "modrange":
files.append(
gen_modrange(parameters[1:], name, output_dir=rdir, save=save))
elif parameters[0] == "dual_modrange":
files.append(
gen_dual_modrange(parameters[1:],
name,
output_dir=rdir,
save=save))
return files, list(all_params.keys())
def __init__(self,args):
if args is not None:
for arg_name,arg_value in args.items():
if hasattr(self, arg_name):
setattr(self, arg_name, arg_value)
if LOG_LEVEL<=LOG_INFO:
print("\n[INFO]Launcher ",self.__class__.__name__," initialized with\n",self.__dict__)
if self.trial_dir is None:
self.trial_dir = os.path.join(ROOT_RESULT_DIR, time.strftime("%j_%H:%M"))
fu.assert_dir(self.trial_dir,should_be_empty=True)
try:
self.worlds = fu.file_list(self.worlds_dir, file_format=".wbt")
except FileNotFoundError:
if LOG_LEVEL<=LOG_WARNING:
print("\n[WARNING]No world file in\n",worlds_dir,"\nignore if optimization")
except AttributeError:
if LOG_LEVEL<=LOG_WARNING:
print("\n[WARNING]No world directory,ignore if optimization")
self.generate_paths()
def generate_paths(self):
"""
Python implementation not suitable for this method at the time of
this project, see generate_paths in CppLauncher for reference
"""
param_paths=[H2D_PATH+"modeling/configFiles/Controllers/current"+str(i)
+".yaml" for i in range(self.nb_ind)]
log_paths=[H2D_PATH+"webots/controllers/GeyerReflex/Raw_files"+str(i)
for i in range(self.nb_ind)]
worlds_paths=[H2D_PATH+"webots/worlds/current"+str(i)+
".wbt" for i in range(self.nb_ind)]
for p_path,l_path,w_path in zip(param_paths, log_paths,worlds_paths):
fu.assert_dir(l_path)
s1=fu.assert_file_exists(p_path, should_exist=True)
s2=fu.assert_file_exists(w_path, should_exist=True)
if not s1 or not s2:
raise RuntimeError("Missing worlds or parameter path, see Readme")
self.param_path=param_paths
self.log_path=log_paths
self.world_path=worlds_paths
raise NotImplementedError
def __init__(self, ev_config):
""" Evolution controller initialisation
Initialization of the evolution controller arguments, creation of
the optimizer object and the trial directory, where the relevant
optimisation parameters are saved
Input:
- ev_config: dict containing all configuration for the evolution,
read from the .yaml file specified at launch
(see ev_config_template.yaml)
Dependencies:
optimizers
"""
self.__dict__.update(ev_config)
self.trial_dir = os.path.join(ROOT_RESULT_DIR,
time.strftime("%j_%H:%M"))
assert_dir(self.trial_dir, should_be_empty=True)
with open(os.path.join(self.trial_dir, "ev_params.yaml"),
"w+") as outparams:
yaml.dump(ev_config, outparams, default_flow_style=False)
if self.optimizer_name == "NSGAII":
self.union_pop = True
self.opti = NSGAIIOptimizer(ev_config)
elif self.optimizer_name == "GA":
self.union_pop = False
self.opti = GAOptimizer(ev_config)
elif self.optimizer_name == "PSO":
self.union_pop = False
self.opti = PSOptimizer(ev_config)
else:
if LOG_LEVEL <= LOG_ERROR:
print("\n[ERROR]Unknown optimizer name\n", self.optimizer_name)
raise ValueError
self.nb_gen = 0
self.flatten_params()
self.current_pop = None
if __name__ == '__main__': mode=sys.argv[1] param=sys.argv[2:] if mode=="plot_with_success": run_dir = param[0] #ref_dir=param[1] metrics=["maxtime", "distance", "mean_speed", "energy","energy_to_dist","corankle","corhip","corknee"] disc_params=[0.0, 19.0, 19.0, 19.0, 19.0, 19.0,19.0,19.0] do_save=True gen_dirs=fu.dir_list(run_dir,"param") for metric,disc_param in zip(metrics,disc_params): if do_save: save_path=os.path.join(run_dir,"figures",metric) fu.assert_dir(save_path) else: save_path=None if len(gen_dirs)>0: for gen_dir in gen_dirs: ind_dirs=fu.dir_list(gen_dir,pattern="ind") plot_with_success( ind_dirs,metric,disc_name='maxtime', disc_params=disc_param,what="max_value",save_path=save_path) else: ind_dirs=fu.dir_list(run_dir,pattern="ind") plot_with_success( ind_dirs,metric,disc_name='maxtime', disc_params=disc_param,what="max_value",save_path=save_path) elif mode=="export_joined_df": """ python process_run.py export_joined_df /data/prevel/runs/dual_modrange_SOL_TA/param1/ 4000 /data/prevel/runs/dual_modrange_SOL_TA/met.csv
def run_batch(self,mode,*args):
""" Implementation of different run launching modes
- parallel_sensitivity_analysis :
Generates and runs parameters set to perform a sensitivity analysis
(see reference file). Raw data generted by each run is saved for
further processing
Input :
parameter file, see sensi_template.yaml
Output :
Saves raw data, with one directory per line on the parameter
file for further processing
Dependencies:
gen_all_file in generate_paramfile_range.py
- check:
Parses the arguments to create a single individual, and runs
webots in full screen, displaying the processed information
Input :
list formated as
[ [ param1, value1, param2, value2] ], see
main for reference
- worlds:
Runs all available worlds (in self.world_path), and returns
the logs directory.
Input:
None, but we must take care to update individuals parameters
(with create_pop) before calling this method
Output:
List containing logs directories for further processing
(see get_fitness_from_dir in data_analysis/process_run.py)
"""
if mode == "parallel_sensitivity_analysis":
param_values_file = args[0][0]
if LOG_LEVEL<=LOG_INFO:
print("\n[INFO]Parameters taken from\n",param_values_file)
all_params,gen_ids=gen_all_file(param_values_file, standalone=True)
gen_count=1
for gen_vals, gen_id in zip(all_params,gen_ids):
save_dir = os.path.join(self.trial_dir, gen_id)
fu.assert_dir(save_dir, should_be_empty=True)
population={}
for ind, cnt in zip(gen_vals,range(len(gen_vals))):
uid="param"+str(gen_count)+"val"+str(cnt)
population[uid]=ind
self.create_pop(population)
self.run_worlds(self.world_path,self.log_path)
runs=self.proc_run.import_runs_from_dir(self.log_path,save=False)
for run, cnt in zip(runs,range(len(runs))):
run.to_csv(os.path.join(save_dir,"raw"+str(cnt)+".csv"))
gen_count+=1
elif mode=="check":
ind={}
for par_idx in range(0,len(args[0]),2):
ind[args[0][par_idx]]=float(args[0][par_idx+1])
if LOG_LEVEL<=LOG_INFO:
print("\n[INFO]Checking individual\n",ind)
self.check_run(ind)
elif mode=="worlds":
self.run_worlds(self.world_path,self.log_path)
return self.log_path
else:
if LOG_LEVEL<=LOG_ERROR:
print("[ERROR]Incorrect input",mode)
raise KeyError