def projection(config, step=1e-3):
""" info = SU2.run.projection(config,step=1e-3)
Runs an gradient projection with:
SU2.run.decomp()
SU2.run.DOT()
Assumptions:
Writes tecplot file of gradients
Adds objective suffix to gradient plot filename
Inputs:
config - an SU2 config
step - a float or list of floats for geometry sensitivity
finite difference step
Outputs:
info - SU2 State with keys:
GRADIENTS.<config.OBJECTIVE_FUNCTION>
Updates:
config.MATH_PROBLEM
Executes in:
./
"""
# local copy
konfig = copy.deepcopy(config)
# choose dv values
Definition_DV = konfig['DEFINITION_DV']
n_DV = len(Definition_DV['KIND'])
if isinstance(step, list):
assert len(step) == n_DV, 'unexpected step vector length'
else:
step = [step] * n_DV
dv_old = [
0.0
] * n_DV # SU2_DOT input requirement, assumes linear superposition of design variables
dv_new = step
konfig.unpack_dvs(dv_new, dv_old)
# filenames
objective = konfig['OBJECTIVE_FUNCTION']
grad_filename = konfig['GRAD_OBJFUNC_FILENAME']
output_format = konfig['OUTPUT_FORMAT']
plot_extension = su2io.get_extension(output_format)
adj_suffix = su2io.get_adjointSuffix(objective)
grad_plotname = os.path.splitext(
grad_filename)[0] + '_' + adj_suffix + plot_extension
# Run Projection
SU2_DOT(konfig)
# read raw gradients
raw_gradients = su2io.read_gradients(grad_filename)
os.remove(grad_filename)
# Write Gradients
data_plot = su2util.ordered_bunch()
data_plot['VARIABLE'] = range(len(raw_gradients))
data_plot['GRADIENT'] = raw_gradients
data_plot['FINDIFF_STEP'] = step
su2util.write_plot(grad_plotname, output_format, data_plot)
# gradient output dictionary
gradients = {objective: raw_gradients}
# info out
info = su2io.State()
info.GRADIENTS.update(gradients)
return info
def projection(config, state={}, step=1e-3):
""" info = SU2.run.projection(config,state,step=1e-3)
Runs an gradient projection with:
SU2.run.decomp()
SU2.run.DOT()
Assumptions:
Writes tecplot file of gradients
Adds objective suffix to gradient plot filename
Inputs:
config - an SU2 config
state - only required when using external custom DV
step - a float or list of floats for geometry sensitivity
finite difference step
Outputs:
info - SU2 State with keys:
GRADIENTS.<config.OBJECTIVE_FUNCTION>
Updates:
config.MATH_PROBLEM
Executes in:
./
"""
# local copy
konfig = copy.deepcopy(config)
# choose dv values
Definition_DV = konfig['DEFINITION_DV']
n_DV = sum(Definition_DV['SIZE'])
if isinstance(step, list):
assert len(step) == n_DV, 'unexpected step vector length'
else:
step = [step] * n_DV
dv_old = [
0.0
] * n_DV # SU2_DOT input requirement, assumes linear superposition of design variables
dv_new = step
konfig.unpack_dvs(dv_new, dv_old)
# filenames
objective = konfig['OBJECTIVE_FUNCTION']
grad_filename = konfig['GRAD_OBJFUNC_FILENAME']
output_format = konfig['OUTPUT_FORMAT']
plot_extension = su2io.get_extension(output_format)
adj_suffix = su2io.get_adjointSuffix(objective)
grad_plotname = os.path.splitext(
grad_filename)[0] + '_' + adj_suffix + plot_extension
# Run Projection
SU2_DOT(konfig)
# read raw gradients
raw_gradients = su2io.read_gradients(grad_filename)
os.remove(grad_filename)
info = su2io.State()
if ('CUSTOM' in konfig.DV_KIND):
if ('OUTFLOW_GENERALIZED' in objective):
weight = 1.0
if (len(objective.split(',')) > 1):
obj = objective.split(',').index('OUTFLOW_GENERALIZED')
weight = float(konfig['OBJECTIVE_WEIGHT'].split(',')[obj])
import downstream_function # Must be defined in run folder
chaingrad = downstream_function.downstream_gradient(
konfig, state, step)
n_dv = len(raw_gradients)
custom_dv = 1
for idv in range(n_dv):
if (konfig.DV_KIND[idv] == 'CUSTOM'):
raw_gradients[idv] = chaingrad[4 + custom_dv] * weight
custom_dv = custom_dv + 1
else:
n_dv = len(raw_gradients)
custom_dv = 1
for idv in range(n_dv):
if (konfig.DV_KIND[idv] == 'CUSTOM'):
raw_gradients[idv] = 0.0
custom_dv = custom_dv + 1
# Write Gradients
data_plot = su2util.ordered_bunch()
data_plot['VARIABLE'] = range(len(raw_gradients))
data_plot['GRADIENT'] = raw_gradients
data_plot['FINDIFF_STEP'] = step
su2util.write_plot(grad_plotname, output_format, data_plot)
# gradient output dictionary
objective = objective.split(',')
if (len(objective) > 1):
objective = ['COMBO']
gradients = {objective[0]: raw_gradients}
# info out
info.GRADIENTS.update(gradients)
return info