OpenMDAO MetaModel visualization

This example loads an OpenMDAO MetaModel stored in latom.data.metamodels, predicts additional outputs interpolating existing data and plots the corresponding response surface.

@authors: Alberto FOSSA’ Giuliana Elena MICELI

import numpy as np

from latom.surrogate.om_metamodels import MetaModel

# MetaModel settings
distributed = False  # variables distributed across multiple processes
extrapolate = False  # extrapolation for out-of-bounds inputs

# interpolation method among slinear, lagrange2, lagrange3, cubic, akima, scipy_cubic, scipy_slinear, scipy_quintic
interp_method = 'slinear'

# evaluation points
isp = np.linspace(300, 400, 5)  # Isp values for evaluation [s]
twr = np.linspace(2, 3, 5)  # twr values for evaluation [-]

training_data_gradients = True  # compute gradients wrt output training data
vec_size = np.size(isp)  # number of points to evaluate at once
rec_file = 'asc_const_mm.pkl'  # name of the file in latom.data.metamodels in which the solution is serialized
kind = 'prop'  # quantity to display, 'prop' for propellant fraction or 'final' for final/initial mass ratio

# initialize MetaModel
mm = MetaModel(distributed=distributed, extrapolate=extrapolate, method=interp_method,
               training_data_gradients=training_data_gradients, vec_size=vec_size, rec_file=rec_file)

# predict additional outputs
mm.p['twr'] = twr
mm.p['Isp'] = isp
mm.p.run_model()  # run the model to interpolate stored data
m_prop = mm.p['mm.m_prop']  # predicted propellant fraction [-]

print('Predicted propellant fraction: ', m_prop)

# plot response surface
mm.plot(nb_lines=40, log_scale=False, kind=kind)