def plot2d(self,
plotfreq=1,
show_plot=False,
fname='plots/2dtestplot.pdf'):
xs = np.arange(self.domain[0][0], self.domain[0][1], self.res)
ys = np.arange(self.domain[1][0], self.domain[1][1], self.res)
xs, ys = np.meshgrid(xs, ys)
def pred(i, j):
#if (i*len(xs[0])+j)%100==0:
#print('making ' +str(i*len(xs[0])+j) + 'th prediction')
return self.pred_y([xs[i][j], ys[i][j]])
#print('about to make Z')
Z = [[pred(i, j) for j in range(len(xs[i]))] for i in range(len(xs))]
#Z = [ [ self.objective_func([xs[i][j],ys[i][j]]) for j in range(len(xs[i]) )] for i in range(len(xs))]
"""
xs = self.domain_buffer.transpose()[0]
ys = self.domain_buffer.transpose()[1]
def prediction(i):
if i%100==0:
print('making ' +str(i) + 'th prediction')
print('with x = ('+str(xs[i])+', '+str(ys[i])+')')
return self.pred_y([xs[i],ys[i]])
Z = [ prediction(i) for i in range(len(xs)) ]
"""
print('got the Z')
plt.figure()
CS = plt.contour(xs, ys, Z, 20)
points = plt.plot([x[0] for x in self.X], [x[1] for x in self.X],
'ko',
label="sample points")
print('contoured the thing')
plt.title('branin model test n = ' + str(self.n))
if fname: plt.savefig(fname, bbox_inches='tight')
if show_plot: plt.show()
def plot2d(self, plotfreq=1, show_plot=False, fname='plots/2dtestplot.pdf'):
xs = np.arange(self.domain[0][0], self.domain[0][1], self.res)
ys = np.arange(self.domain[1][0], self.domain[1][1], self.res)
xs, ys = np.meshgrid(xs, ys)
def pred(i,j):
#if (i*len(xs[0])+j)%100==0:
#print('making ' +str(i*len(xs[0])+j) + 'th prediction')
return self.pred_y([xs[i][j],ys[i][j]])
#print('about to make Z')
Z = [[pred(i,j) for j in range(len(xs[i]))] for i in range(len(xs))]
#Z = [ [ self.objective_func([xs[i][j],ys[i][j]]) for j in range(len(xs[i]) )] for i in range(len(xs))]
"""
xs = self.domain_buffer.transpose()[0]
ys = self.domain_buffer.transpose()[1]
def prediction(i):
if i%100==0:
print('making ' +str(i) + 'th prediction')
print('with x = ('+str(xs[i])+', '+str(ys[i])+')')
return self.pred_y([xs[i],ys[i]])
Z = [ prediction(i) for i in range(len(xs)) ]
"""
print('got the Z')
plt.figure()
CS = plt.contour(xs, ys, Z,20)
points = plt.plot([x[0] for x in self.X],[x[1] for x in self.X], 'ko',label="sample points")
print('contoured the thing')
plt.title('branin model test n = ' + str(self.n))
if fname: plt.savefig(fname, bbox_inches='tight')
if show_plot: plt.show()
def domain_buffer(self):
"""
Returns:
np.array: a list of sample points that cover the domain in a grid with cell length of self.res in every dimension
"""
grid_bounds = [np.arange(self.domain[i][0], self.domain[i][1], self.res) for i in range(self.k)]
if self.k>1:
mesh = np.transpose(np.meshgrid(*grid_bounds))
else:
mesh = grid_bounds[0]
points = mesh.reshape(len(mesh.flatten())/self.k,self.k)
return points
def domain_buffer(self):
"""
Returns:
np.array: a list of sample points that cover the domain in a grid with cell length of self.res in every dimension
"""
grid_bounds = [
np.arange(self.domain[i][0], self.domain[i][1], self.res)
for i in range(self.k)
]
if self.k > 1:
mesh = np.transpose(np.meshgrid(*grid_bounds))
else:
mesh = grid_bounds[0]
points = mesh.reshape(len(mesh.flatten()) / self.k, self.k)
return points
def plot1d(self,
dim=0,
plot_objective=False,
plot_err=True,
plot_improvement=True,
plot_next_sp=True,
show_plot=False,
fname=None,
legend=False):
"""
Args:
plot_objective (bool): whether the objective function should be plotted. This is only feasible if that function can be called enough times to generate a smooth plot
plot_improvement (bool): option to overlay expected emprovement at each x
plot_next_sp (bool): whether the selection of the next sample point is plotted
show_plot (bool): whether pyplot.show is called at the end of the function
dim (int): the index of the dimension of interest
x_delta (float): the resolution of the plot
fname: if set, the plot is saved to this
Uses pyplot to make a nice 1d plot of the predictor function and its error. Optionally, overlay a plot of the
"""
x_min = self.domain[dim][0]
x_max = self.domain[dim][1]
#plt.rc('text', usetex=True)
#plt.rc('font', family='serif')
font = {'fontname': 'Palatino'}
fig, ax = plt.subplots()
plt.subplots_adjust(left=0.25, bottom=0.25)
plt.xlabel('x')
ax.set_ylabel('predicted y(x)')
plt.title('Predictor Surface with ' + str(self.n) + ' samples')
pred_range = np.arange(x_min, x_max, self.res)
if plot_objective:
obj = [self.objective_func([x]) for x in pred_range]
exp_imp_line, = ax.plot(pred_range,
obj,
'k--',
linewidth=2,
label='objective func')
preds = self.prediction_buffer
# errors are illustrated with 2 times sigma (two standard devs)
# plot the predictor and +/- errors
pred_line, = ax.plot(pred_range, preds, label='predictor surface')
if plot_err:
# to illustrate a 95% confidence interval
errors = self.error_buffer
# elem-wise sum/difference of above two arrays
pl_errors = map(add, preds, errors)
mi_errors = map(sub, preds, errors)
ax.fill_between(pred_range,
pl_errors,
mi_errors,
facecolor="lightgreen")
p_err_line, = ax.plot(pred_range,
pl_errors,
color="green",
label='plus/minus error')
m_err_line, = ax.plot(pred_range, mi_errors, color="green")
y_min = np.amin(preds) - np.amax(errors)
y_max = np.amax(preds) + np.amax(errors)
plt.axis([x_min, x_max, -4, 19])
# just for making some plots right now
#plt.axis([self.domain[0][0], self.domain[0][1], -3, 12])
# legend handler, labels
leg_hand, leg_lab = ax.get_legend_handles_labels()
# make another axis (exp improv. is at a smaller scale than predictor)
# plot the expected improvement
if plot_improvement:
ax2 = ax.twinx()
imps = [self.exp_improvement([x]) for x in pred_range]
# if you're plotting, might as well use that info for maximization
exp_imp_line, = ax2.plot(pred_range,
imps,
color='r',
label='expected improvement')
ax2.set_ylabel('expected improvement', color='r')
for tl in ax2.get_yticklabels():
tl.set_color('r')
ax2.axis([x_min, x_max, 0, 1.])
# combining labels from both axes
lh2, ll2 = ax2.get_legend_handles_labels()
leg_hand += lh2
leg_lab += ll2
# plot a vertical dotted line at the next sample point
if plot_next_sp:
ax3 = ax.twinx()
point = ax.axvline(self.next_sample,
color='k',
linestyle='dotted',
label='next sample')
lh3, ll3 = ax3.get_legend_handles_labels()
leg_hand += lh3
leg_lab += ll3
# plot the actual sample points
points = ax.plot([x[dim] for x in self.X],
self.Y,
'ko',
label='sample points')
if legend:
plt.legend(leg_hand, leg_lab, loc=9, numpoints=1, fontsize=10)
if fname:
# bbox is tight bc I'll add margins in latex if I want to, thankyouverymuch
plt.savefig(fname, bbox_inches='tight')
plt.tight_layout()
if show_plot: plt.show()
plt.close()
def plot1d(self, dim=0, plot_objective=False, plot_err=True, plot_improvement=True, plot_next_sp=True, show_plot=False, fname=None, legend=False):
"""
Args:
plot_objective (bool): whether the objective function should be plotted. This is only feasible if that function can be called enough times to generate a smooth plot
plot_improvement (bool): option to overlay expected emprovement at each x
plot_next_sp (bool): whether the selection of the next sample point is plotted
show_plot (bool): whether pyplot.show is called at the end of the function
dim (int): the index of the dimension of interest
x_delta (float): the resolution of the plot
fname: if set, the plot is saved to this
Uses pyplot to make a nice 1d plot of the predictor function and its error. Optionally, overlay a plot of the
"""
x_min=self.domain[dim][0]
x_max=self.domain[dim][1]
#plt.rc('text', usetex=True)
#plt.rc('font', family='serif')
font = {'fontname':'Palatino'}
fig, ax = plt.subplots()
#plt.subplots_adjust(left=0.25, bottom=0.25)
plt.xlabel('x')
ax.set_ylabel('predicted y(x)')
plt.title('Predictor Surface with '+str(self.n)+' samples')
pred_range = np.arange(x_min, x_max, self.res)
if plot_objective:
obj = [ self.objective_func([x]) for x in pred_range ]
exp_imp_line, = ax.plot( pred_range, obj, 'k--', linewidth=2, label= 'objective func' )
# plot the actual sample points
points = ax.plot([x[dim] for x in self.X],self.Y, 'ko',label='sample points')
preds = self.prediction_buffer
# errors are illustrated with 2 times sigma (two standard devs)
# plot the predictor and +/- errors
pred_line, = ax.plot(pred_range, preds, label='predictor surface')
if plot_err:
# to illustrate a 95% confidence interval
errors = self.error_buffer
# elem-wise sum/difference of above two arrays
pl_errors = map(add, preds, errors)
mi_errors = map(sub, preds, errors)
ax.fill_between(pred_range, pl_errors, mi_errors, facecolor="lightgreen")
p_err_line, = ax.plot(pred_range, pl_errors, color="green", label= 'plus/minus error')
m_err_line, = ax.plot(pred_range, mi_errors, color="green")
y_min=np.amin(preds)-np.amax(errors)
y_max=np.amax(preds)+np.amax(errors)
plt.axis([x_min, x_max, -4, 19])
# just for making some plots right now
#plt.axis([self.domain[0][0], self.domain[0][1], -3, 12])
# legend handler, labels
leg_hand, leg_lab = ax.get_legend_handles_labels()
# make another axis (exp improv. is at a smaller scale than predictor)
# plot the expected improvement
if plot_improvement:
ax2 = ax.twinx()
ax2.axis([x_min, x_max, 0, 1])
imps = [ self.exp_improvement([x]) for x in pred_range ]
# if you're plotting, might as well use that info for maximization
exp_imp_line, = ax2.plot(pred_range, imps, color='r',label= 'expected improvement')
ax2.set_ylabel( 'expected improvement', color='r')
for tl in ax2.get_yticklabels():
tl.set_color('r')
if plot_next_sp:
# plot a vertical dotted line at the next sample point
point = ax2.axvline(self.next_sample, color='k', linestyle='dotted', label= 'next sample')
# combining labels from both axes
lh2,ll2 = ax2.get_legend_handles_labels()
leg_hand += lh2
leg_lab += ll2
ax2.tick_params(axis="y", labelcolor="r")
if legend:
plt.legend(leg_hand,leg_lab,numpoints=1,fontsize=10)
if fname:
# bbox is tight bc I'll add margins in latex if I want to, thankyouverymuch
plt.savefig(fname, bbox_inches='tight')
plt.tight_layout()
if show_plot: plt.show()
plt.close()
"""
.. module:: tests.plot
:platform: Unix, Windows
:synopsis: tests 2d plotting
.. moduleauthor:: Drew Blount <*****@*****.**>
"""
from smbo import np, plt
import matplotlib.mlab as mlab
from test_funcs import branin, branin_domain
delta = 0.25
dom = branin_domain()
x = np.arange(dom[0][0], dom[0][1], delta)
y = np.arange(dom[1][0], dom[1][1], delta)
X, Y = np.meshgrid(x, y)
Z1 = mlab.bivariate_normal(X, Y, 1.0, 1.0, 0.0, 0.0)
Z2 = mlab.bivariate_normal(X, Y, 1.5, 0.5, 1, 1)
# difference of Gaussians
print(str(Y[0]))
Z = [ [ branin( [X[i][j], Y[i][j]] ) for j in range(len(X[i]) )] for i in range(len(X))]
plt.figure()
CS = plt.contour(X, Y, Z,100)
plt.title('branin test')
plt.show()
"""
.. module:: tests.plot
:platform: Unix, Windows
:synopsis: tests 2d plotting
.. moduleauthor:: Drew Blount <*****@*****.**>
"""
from smbo import np, plt
import matplotlib.mlab as mlab
from test_funcs import branin, branin_domain
delta = 0.25
dom = branin_domain()
x = np.arange(dom[0][0], dom[0][1], delta)
y = np.arange(dom[1][0], dom[1][1], delta)
X, Y = np.meshgrid(x, y)
Z1 = mlab.bivariate_normal(X, Y, 1.0, 1.0, 0.0, 0.0)
Z2 = mlab.bivariate_normal(X, Y, 1.5, 0.5, 1, 1)
# difference of Gaussians
print(str(Y[0]))
Z = [[branin([X[i][j], Y[i][j]]) for j in range(len(X[i]))]
for i in range(len(X))]
plt.figure()
CS = plt.contour(X, Y, Z, 100)
plt.title('branin test')
plt.show()