def compute(self, inputs, outputs):
"""Compute the objective expression"""
global counter
counter += 1
cpacs_path = mif.get_tooloutput_file_path(Rt.modules[-1])
# Save the CPACS file for this iteration
if counter % Rt.save_iter == 0:
loc = optim_dir_path + '/Geometry/' + 'iter_{}.xml'.format(counter)
shutil.copy(cpacs_path, loc)
# Add new variables to dictionnary
tixi = cpsf.open_tixi(cpacs_path)
dct.update_dict(tixi, optim_var_dict)
# Change local wkdir for the next iteration
tixi.updateTextElement(opf.WKDIR_XPATH,
ceaf.create_new_wkdir(Rt.date, Rt.type))
for obj in Rt.objective:
var_list = splt('[+*/-]', obj)
for v in var_list:
if not v.isdigit() and v != '':
exec('{} = inputs["{}"]'.format(v, v))
result = eval(obj)
if Rt.minmax == 'min':
outputs['Objective function ' + obj] = -result
else:
outputs['Objective function ' + obj] = result
cpsf.close_tixi(tixi, cpacs_path)
def setup(self):
""" Setup inputs and outputs"""
declared = []
for obj in Rt.objective:
var_list = splt('[+*/-]', obj)
for v in var_list:
if v not in declared:
self.add_input(v)
declared.append(v)
self.add_output('Objective function ' + obj)
def var_and_signs( csin ):
# Ex: if csin = 'Qsol+Qr-Qlat-Qsens' (or '+Qsol+Qr-Qlat-Qsens')
# this function will return:
# ['Qsol', 'Qr', 'Qlat', 'Qsens'] , [1.0, 1.0, -1.0, -1.0]
from re import split as splt
#
sgn1 = '+'
cvar = splt('\-|\+',csin)
if cvar[0] == '':
if csin[0] == '-': sgn1 = '-'
cvar = cvar[1:]
ccum = ''
for cc in cvar: ccum=ccum+'|'+cc
ccum = ccum+'|'
ctt=splt("'"+ccum+"'",csin)
csgn = ctt[:-1]
isgn = []
for cs in csgn: isgn.append(float(cs+'1'))
return cvar, isgn
def extract_data_set(Tool):
"""Get training data from file.
Retrieve training dataset from a file or generates it automatically. The
file format must be a CSV with each variable parameters and values are in
a row. The file must also contain a 'Name' and a 'type' column which are
needed to describe the variable.
Args:
Tool (Prediction_tool object):
Returns:
x (n*m numpy array): Set of m points in the n-dimensionnal design space.
y (p*m numpy array): Set of m points in the p-dimensionnal result space.
"""
df = pd.read_csv(Tool.user_file)
df = df.rename(columns={'Unnamed: 0': 'Name'})
# Separate the input and output points
x = df.loc[[i for i, v in enumerate(df['type']) if v == 'des']]
y = df.loc[[i for i, v in enumerate(df['type']) if v == 'obj']]
df = df.loc[[i for i, v in enumerate(df['type']) if v in ['obj', 'des']]]
df_data = df[['Name', 'type', 'getcmd', 'setcmd']]
df = df.set_index('Name')
y = y.set_index('Name')
# Extract the iteration results only
x = x[[i for i in x.columns if i.isdigit()]]
df = df[[i for i in df.columns if i.isdigit()]]
# Add user-specified objectives
for obj in Tool.objectives:
if obj not in df.index:
var_list = splt('[+*/-]', obj)
for v in var_list:
if not v.isdigit() and v != '' and v in df.index:
exec('{} = df.loc["{}"]'.format(v, v))
df_data = Tool.df.append({
'Name': obj,
'type': 'obj'
},
ignore_index=True)
y.loc[obj] = eval(obj)
Tool.objectives = y.index
Tool.df = pd.concat([Tool.df, df_data], ignore_index=True)
y = y[[i for i in y.columns if i.isdigit()]]
log.info('Set extracted')
return x.transpose().to_numpy(), y.transpose().to_numpy()
def extract_data_set(file):
"""Get training data from file.
Retrieve training dataset from a file or generates it automatically. The
file format must be a CSV with each variable parameters and values are in
a row. The file must also contain a 'Name' and a 'type' column which are
needed to describe the variable.
Args:
file (str) : Path to user-specified file (CSV format)
Returns:
x (DataFrame) : Set of points in the design space
y (DataFrame) : Set of points in the result space
TODO:
* Maybe better deal with the dataframe, case of multiple objective functions ?
"""
df = pd.read_csv(file)
df = df.rename(columns={'Unnamed: 0': 'Name'})
# Drop duplicate (first and second columns are the same)
df = df.drop('0', 1)
# Separate the input and output points
x = df.loc[[i for i, v in enumerate(df['type']) if v == 'des']]
y = df.loc[[i for i, v in enumerate(df['type']) if v == 'obj']]
df = df.set_index('Name')
# Extract the iteration results only
x = x[[i for i in x.columns if i.isdigit()]]
df = df[[i for i in df.columns if i.isdigit()]]
# Compute the objectives in the aeromap case, else they are already given
if 'Variable_history' not in file:
y = pd.DataFrame()
for i, obj_expr in enumerate(objectives):
for name in splt('[+*/-]', obj_expr):
obj_expr = obj_expr.replace(name, 'df.loc["{}"]'.format(name))
exec('y["{}"] = {}'.format(objectives[i], obj_expr))
else:
y = y[[i for i in y.columns if i.isdigit()]].transpose()
return x.transpose().to_numpy(), y.to_numpy()
def compute_obj():
"""
Interpret the objective variable command.
Get all the values needed to compute the objective function and
evaluates its expression.
Returns
-------
None.
"""
# Get variable keys from objective function string
var_list = splt('[+*/-]', Rt.objective)
# Create local variable and assign value
for v in var_list:
exec('{} = res_var_dict["{}"][1][-1]'.format(v, v))
result = eval(Rt.objective)
log.info('Objective function {} : {}'.format(Rt.objective, result))
# Evaluate objective function expression
return -result