def set_height(self, height):
"""(Deprecated) Set the height of the rectangle."""
warning_message('set_width(...) is deprecated. Use property ' \
'myrect.width=... instead.', 'emopt.grid')
libGrid.Rectangle_set_height(self._object, height)
self._yspan = height
def set_width(self, width):
"""(Deprecated) Set the width of the rectangle.
"""
warning_message('set_width(...) is deprecated. Use property ' \
'myrect.width=... instead.', 'emopt.grid')
libGrid.Rectangle_set_width(self._object, width)
self._xspan = width
def set_layer(self, layer):
"""Set the layer of the primitive.
Parameters
----------
layer : int
The new layer.
"""
warning_message('set_layer(...) is deprecated. Use property ' \
'myprim.layer=... instead.', 'emopt.grid')
libGrid.MaterialPrimitive_set_layer(self._object, c_int(layer))
def get_layer(self):
"""Get the layer of the primitive.
Returns
-------
int
The layer.
"""
warning_message('get_layer() is deprecated. Use property ' \
'myprim.layer instead.', 'emopt.grid')
return libGrid.MaterialPrimitive_get_layer(self._object)
def set_for_version(self, value):
self.for_version = self.for_version_prop.get_string_value()
if self.for_version.startswith('3.'):
## disable lisp for wx > 2.8
if self.codewriters_prop.get_string_value() == 'lisp':
misc.warning_message( _('Generating Lisp code for wxWidgets version %s is not supported.\n'
'Set version to "2.8" instead.') % self.for_version )
self.for_version_prop.set_str_value('2.8')
self.set_for_version('2.8')
return
self.codewriters_prop.enable_item('lisp', False)
else:
# enable lisp again
self.codewriters_prop.enable_item('lisp', True)
def _set_language(self):
"Set code generator language and adapt corresponding settings like file dialog wild cards (value: str or int)"
language = self.language
# update wildcards and default extension in the dialog
self._update_output_path(language)
# check that the new language supports all the widgets in the tree
self.check_codegen()
# disable lisp for wx > 2.8
if language == 'lisp':
for_version = self.for_version
if for_version == '3.0':
misc.warning_message(
_('Generating Lisp code for wxWidgets version %s is not supported.\n'
'Set version to "2.8" instead.') % self.for_version)
self.properties["for_version"].set('2.8')
self.properties["for_version"].set_blocked(True)
else:
self.properties["for_version"].set_blocked(False)
# don't change the extension in multiple files mode
if self.multiple_files == 1:
return
# update file extensions
current_name = self.output_path
if not current_name:
return
base, ext = os.path.splitext(current_name)
# is already a valid extension? ext has a leading . but default_extensions hasn't
if ext and ext[1:] in common.code_writers[language].default_extensions:
return
new_name = "%s.%s" % (
base, common.code_writers[language].default_extensions[0])
self.properties["output_path"].set(new_name)
blocked = self.language != "C++"
self.properties["source_extension"].set_blocked(blocked)
self.properties["header_extension"].set_blocked(blocked)
def check_gradient(self, params, indices=[], plot=True):
"""Verify that the gradient is accurate.
It is highly recommended that the accuracy of the gradients be checked
prior to being used. If the accuracy is above ~1%, it is likely that
there is an inconsitency between how the figure of merit and adjoint
sources (dFdx) are being computed.
The adjoint method gradient error is evaluated by comparing the
gradient computed using the adjoint method to a gradient computed by
direct finite differences. In other words, the "correct" derivatives
to which the adjoint method gradient is compared are given by
.. math::
\\frac{\partial F}{\partial p_i} \\approx \\frac{F(p_i + \Delta p) - F(p_i)}{\Delta p}
Note that this method for calculating the gradient is not used in a
practical setting because it requires performing N+1 simulations in
order to compute the gradient with respect to N design variables
(compared to only 2 simulations in the case of the adjoint method).
Parameters
----------
params : numpy.ndarray
design parameters
indices : list or numpy.ndarray
list of gradient indices to check. An empty list indicates that the
whole gradient should be verified. A subset of indices may be
desirable for large problems. (default = [])
Returns
-------
float
Relative error in gradient.
"""
if (indices == []):
indices = np.arange(0, len(params), 1)
# make sure everything is up to date
self.update_system(params)
self.sim.update()
grad_am = self.gradient(params)
grad_fd = np.zeros(len(indices))
fom0 = self.fom(params)
# calculate the "true" derivatives using finite differences
if (NOT_PARALLEL):
info_message('Checking gradient...')
for i in range(len(indices)):
if (NOT_PARALLEL):
print('\tDerivative %d of %d' % (i + 1, len(indices)))
j = indices[i]
p0 = params[j]
params[j] += self._step
fom1 = self.fom(params)
if (NOT_PARALLEL):
grad_fd[i] = (fom1 - fom0) / self._step
params[j] = p0
if (NOT_PARALLEL):
errors = np.abs(grad_fd - grad_am[indices]) / np.abs(grad_fd)
error_tot = np.linalg.norm(
grad_fd - grad_am[indices]) / np.linalg.norm(grad_fd)
if (error_tot < 0.01):
info_message('The total error in the gradient is %0.4E' % \
(error_tot))
else:
warning_message('The total error in the gradient is %0.4E '
'which is over 1%%' % (error_tot), \
'emopt.adjoint_method')
if (plot):
import matplotlib.pyplot as plt
f = plt.figure()
ax1 = f.add_subplot(311)
ax2 = f.add_subplot(312)
ax3 = f.add_subplot(313)
ax1.bar(indices, grad_fd)
ax1.set_title('Finite Differences')
ax2.bar(indices, grad_am[indices])
ax2.set_title('Adjoint Method')
ax3.bar(indices, errors)
ax3.set_title('Error in Adjoint Method')
ax3.set_yscale('log', nonposy='clip')
plt.show()
return error_tot
else:
return None
def step(self, val):
if (np.abs(val) > self.sim.dx / 1e3):
if (NOT_PARALLEL):
warning_message('Step size used for adjoint method may be too '
'large. Consider reducing it to ~1e-3*dx')
self._step = val
def plot_iteration(field,
structure,
W,
H,
foms,
fname='',
layout='auto',
show_now=False,
dark=True,
Nmats=2):
"""Plot the current iteration of an optimization.
Plots the following:
1) A 2D field
2) A 2D representation of the geometry
3) A line plot of the figure of merit history
When plotting 3D structures, multiple 2D slices of the geometry can be
passed in by passing a 3D array to structure. The different "pages" of this
array will be flattened into a 2D representation of the sliced geometry.
Furthermore, multiple figures of merit may be plotted. This is done by
putting more than one key:value pair in the foms dictionary. The key names
will be used as the legend label.
Parameters
----------
field : numpy.ndarray
2D array containing current iteration's field slice
structure : numpy.ndarray
Either a 2D array containing a representation of the current structure
or a 3D array containing a FEW 2D slices.
W : float
The width of the field/structure
H : float
The height of the field/structure
foms : dict
A dictionary containing the fom history. The key strings should be
names which describe each supplied figure of merit
fname : str
Filename for saving
layout : str
Layout method. 'auto' = automatically choose layout based on aspect
ratio. 'horizontal' = single row layour. 'vertical' = single column
layout. 'balanced' = field+structure on left, foms on right.
show_now : bool
If True, show plot now (warning: this is a blocking operation)
dark : bool
If True, use a dark color scheme for plot. (default = True)
Returns
-------
matplotlib.pyplot.figure
The current matplotlib figure
"""
import matplotlib.pyplot as plt
import matplotlib.gridspec as gridspec
from matplotlib.colors import LinearSegmentedColormap
dpi = 300
aspect = W / H
# determine and setup the plot layout
unknown_layout = layout not in [
'auto', 'horizontal', 'vertical', 'balanced'
]
if (unknown_layout):
warning_message("Unknown layout '%s'. Using 'auto'" % (layout),
'emopt.io')
if (layout == 'auto' or unknown_layout):
if (aspect < 1.0):
layout = 'horizontal'
elif (aspect >= 1.0 and aspect <= 2.5):
layout = 'balanced'
else:
layout = 'vertical'
gs = None
Wplot = 10.0
if (layout == 'horizontal'):
f = plt.figure(figsize=(Wplot * aspect * 3, Wplot))
gs = gridspec.GridSpec(1, 3, height_ratios=[1])
ax_field = f.add_subplot(gs[:, 0])
ax_struct = f.add_subplot(gs[:, 1])
ax_foms = f.add_subplot(gs[:, 2])
elif (layout == 'vertical'):
f = plt.figure(figsize=(Wplot, Wplot / aspect * 3))
gs = gridspec.GridSpec(3, 1, height_ratios=[1, 1, 1])
ax_field = f.add_subplot(gs[0, :])
ax_struct = f.add_subplot(gs[1, :])
ax_foms = f.add_subplot(gs[2, :])
elif (layout == 'balanced'):
f = plt.figure(figsize=(Wplot, Wplot / aspect * 2 / 1.5))
gs = gridspec.GridSpec(2, 3, height_ratios=[1, 1])
ax_field = f.add_subplot(gs[0, 0:2])
ax_struct = f.add_subplot(gs[1, 0:2])
ax_foms = f.add_subplot(gs[:, 2])
# define dark colormaps
if (dark):
field_cols = ['#3d9aff', '#111111', '#ff3d63']
field_cmap = LinearSegmentedColormap.from_list('field_cmap',
field_cols)
struct_cols = ['#212730', '#bcccdb']
struct_cmap = LinearSegmentedColormap.from_list(
'struct_cmap', struct_cols)
else:
field_cmap = 'seismic'
struct_cmap = 'Blues'
extent = [0, W, 0, H]
fmin = -1 * np.max(np.abs(field))
fmax = -1 * fmin
ax_field.imshow(field,
extent=extent,
vmin=fmin,
vmax=fmax,
cmap=field_cmap)
# "flatten" multi-layer structure
if (len(structure.shape) == 3):
Nlevels = structure.shape[0] + 1
structure = np.sum(structure, axis=0) / structure.shape[0]
else:
Nlevels = Nmats
smin = np.min(structure)
smax = np.max(structure)
ax_struct.imshow(structure,
extent=extent,
vmin=smin,
vmax=smax,
cmap=struct_cmap)
# outline structure in field plot
ax_field.contour(np.flipud(structure),
extent=extent,
levels=Nlevels,
colors='#666666',
linewidths=0.1)
# set dark colors
# plot title with important iteration number, etc
# sum along z-axis for structure
# define fom plot colors
Nplot = len(foms.keys())
red = np.linspace(0.2, 1.0, Nplot)
blue = np.linspace(1.0, 0.2, Nplot)
green = np.zeros(Nplot)
red_base = 0.0
blue_base = 0.0
green_base = 0.55
i = 0
Niter = 0
current_foms = []
for desc in foms.keys():
fom = foms[desc]
Niter = len(fom)
iters = np.arange(Niter)
current_foms.append(fom[-1])
pcolor = (red_base + red[i], green_base + green[i],
blue_base + blue[i])
i += 1
pline = ax_foms.plot(iters, fom, '.-', color=pcolor, markersize=10)
ax_foms.set_xlabel('Iteration', fontsize=12)
ax_foms.set_ylabel('Figure of Merit', fontsize=12)
ax_foms.legend(foms.keys(), loc=4)
ax_foms.grid(True, linewidth=0.5)
# general tick properties
for ax in [ax_field, ax_struct, ax_foms]:
ax.get_yaxis().set_tick_params(which='both',
direction='in',
top=True,
right=True)
ax.get_xaxis().set_tick_params(which='both',
direction='in',
top=True,
right=True)
# Dark theme easier on eyes
if (dark):
c_text_main = '#BBBBBB'
c_bg_main = '#101010'
c_plot_bg = '#353535'
c_lines = '#666666'
c_plot_tick = '#CCCCCC'
c_plot_grid = '#555555'
f.patch.set_facecolor(c_bg_main)
for ax in [ax_field, ax_struct, ax_foms]:
for tl in ax.get_xticklabels():
tl.set_color(c_text_main)
for tl in ax.get_yticklabels():
tl.set_color(c_text_main)
ax.xaxis.get_label().set_color(c_text_main)
ax.yaxis.get_label().set_color(c_text_main)
for spine in ax.spines:
ax.spines[spine].set_color(c_lines)
ax.get_yaxis().set_tick_params(color=c_plot_tick)
ax.get_xaxis().set_tick_params(color=c_plot_tick)
ax_foms.set_facecolor(c_plot_bg)
ax_foms.grid(True, color=c_plot_grid, linewidth=0.5)
else:
c_text_main = '#000000'
# title contains info of current iteration
f.suptitle(''.join(['Iteration %d, ' % (Niter), 'FOMs = '] +
['%0.4f ' % (fom) for fom in current_foms]),
fontsize=12,
color=c_text_main)
if (fname != ''):
if (dark):
plt.savefig(fname,
dpi=300,
facecolor=c_bg_main,
bbox_inches='tight')
else:
plt.savefig(fname, dpi=300, bbox_inches='tight')
if (show_now):
plt.tight_layout()
plt.show()
def zmaxs(self):
warning_message(
'The list of maximum z coordinates cannot be changed in this way.',
'emopt.grid')
def primitive(self):
warning_message('The primitive list cannot be modified in this way.',
'emopt.grid')
def set_material(self, mat):
warning_message('set_material(...) is deprecated. Use property ' \
'mypoly.material_value=... instead.', 'emopt.grid')
libGrid.Polygon_set_material(self._object, mat.real, mat.imag)
self._value = mat
def Np(self, N):
warning_message('Polygon.Np cannot be modified in this way.', \
'emopt.grid.Polygon')
def ys(self, vals):
warning_message('Polygon.ys cannot be set on its own. Use ' \
'set_points(x,y) instead', 'emopt.grid.Polygon')
def set_material(self, mat):
"""(Deprecated) Set the material value of the Rectangle's interior.
"""
warning_message('set_mateiral(...) is deprecated. Use property ' \
'myrect.material_value=... instead.', 'emopt.grid')
libGrid.Rectangle_set_material(self._object, mat.real, mat.imag)
def fillet(x, y, R, make_round=None, points_per_90=10, equal_thresh=1e-8,
ignore_roc_lim=False):
"""Round corners of a polygon.
This function replaces sharp corners with circular arcs. The radius of
these arcs will be equal to the specified radius as long as the line
segments which make up the corner are sufficiently long. If they are too
short, the arc will be made with a smaller radius.
If a non-closed polygon is supplied, the end points will be ignored. A
polygon is considered closed if the first and last point in the provided
x,y coordinate lists are the same.
Parameters
----------
x : list
The x coordinates of the chain of line segments to round.
y : list
The y coordinates of the chain of line segments to round.
R : float
The desired fillet radius
make_round : list or None (optional)
A list of boolean values which specifies which points should be
rounded. If None, then all roundable points are rounded. If supplying a
list, the list must have the same length as x and y. (default = None)
points_per_90 : int (optional)
The number of points to generate per 90 degrees of the arc. (default =
10)
equal_thresh : float
The threshold used for comparing values. If the |difference| between
two values is less than this value, they are considered equal.
Returns
-------
list, list
The x and y coordinates of the new set of lines segments or polygon.
"""
xn = []
yn = []
i = 0
N = len(x)
N0 = N
# we always ignore the last point
N -= 1
closed = True
if(x[0] != x[-1] or y[0] != y[-1]):
closed = False
i += 1
xn.append(x[0])
yn.append(y[0])
i = 0
while(i < N):
if(make_round is None or make_round[i]):
# get current and adjacent points
x1 = x[i]
y1 = y[i]
if(i == 0):
x0 = x[-2]
y0 = y[-2]
else:
x0 = x[i-1]
y0 = y[i-1]
if(i == N0-1):
x2 = x[1]
y2 = y[1]
else:
x2 = x[i+1]
y2 = y[i+1]
# calc angle
dx10 = x0-x1
dy10 = y0-y1
dx12 = x2-x1
dy12 = y2-y1
d10 = np.sqrt(dx10**2 + dy10**2)
d12 = np.sqrt(dx12**2 + dy12**2)
theta = np.arccos((dx10*dx12 + dy10*dy12)/(d10*d12))
if(theta != 0 and theta != pi):
dxc = (dx10/d10 + dx12/d12)
dyc = (dy10/d10 + dy12/d12)
dc = np.sqrt(dxc**2 + dyc**2)
nxc = dxc/dc
nyc = dyc/dc
nx10 = dx10/d10
ny10 = dy10/d10
nx12 = dx12/d12
ny12 = dy12/d12
# reduce fillet radius if necessary
Ri = R
iprev = i-1
inext = i+1
if(iprev < 0): iprev = N0-1
if(inext > N0-1): inext = 0
if( (d10 < 2*R or d12 < 2*R) and not ignore_roc_lim):
Ri = np.min([d12, d10])/2.0
if(NOT_PARALLEL):
warning_message('Warning: Desired radius of curvature too large at ' \
'point %d. Reducing to maximum allowed.' % \
(i), 'emopt.geometry')
# figure out where the circle "fits" in the corner
S10 = Ri / np.tan(theta/2.0)
S12 = S10
Sc = np.sqrt(S10**2 + Ri**2)
# generate the fillet
theta1 = np.arctan2((S10*ny10 - Sc*nyc), (S10*nx10 - Sc*nxc))
theta2 = np.arctan2((S12*ny12 - Sc*nyc), (S12*nx12 - Sc*nxc))
if(theta1 < 0): theta1 += 2*pi
if(theta2 < 0): theta2 += 2*pi
theta1 = np.mod(theta1, 2*pi)
theta2 = np.mod(theta2, 2*pi)
if(theta2 - theta1 > pi): theta2 -= 2*pi
elif(theta1 - theta2 > pi): theta2 += 2*pi
Np = int(np.abs(theta2-theta1) / (pi/2) * points_per_90)
if(Np < 1):
Np = 1
thetas = np.linspace(theta1, theta2, Np)
for t in thetas:
xfil = x[i] + Sc*nxc + Ri*np.cos(t)
yfil = y[i] + Sc*nyc + Ri*np.sin(t)
# only add point if not duplicate (this can happen if
# the desired radis of curvature equals or exceeds the
# maximum allowed)
if(np.abs(xfil - xn[-1]) > equal_thresh or
np.abs(yfil - yn[-1]) > equal_thresh):
xn.append(xfil)
yn.append(yfil)
else:
xn.append(x[i])
yn.append(y[i])
else:
xn.append(x[i])
yn.append(y[i])
i += 1
if(not closed):
xn.append(x[-1])
yn.append(y[-1])
else:
xn.append(xn[0])
yn.append(yn[0])
return np.array(xn), np.array(yn)
