def draw_plot(plot_data, ax, func_id, decorations=[],
num_blendcurves=plotparams.GP_DEFAULT_NUM_BLENDCURVES,
blendcurve_alpha=plotparams.GP_DEFAULT_BLENDCURVE_ALPHA):
xp = np.linspace(-20, 20, 100)
xp_fine = np.linspace(-20, 20, 2000)
pmean, pcov = posterior_mean_and_cov(plot_data, xp)
# This is a waste of memory, but we have enough memory and I'd rather
# not rewrite the program yet again
yps = []
for i in range(num_blendcurves):
yp = np.random.multivariate_normal(pmean, pcov)
yps.append(yp)
ax.plot(xp, yp, c="red", alpha=blendcurve_alpha, linewidth=2)
f_true = gpexample_plugin.get_ftrue(func_id)
ax.plot(xp_fine, [f_true(x) for x in xp_fine], **plotparams.GP_FTRUE)
ax.scatter(plot_data.Xseen, plot_data.Yseen, **plotparams.GP_PAST_PROBE)
ax.axes.xaxis.set_ticklabels([])
ax.axes.yaxis.set_ticklabels([])
for artist in decorations:
ax.add_artist(artist)
ax.set_xlim(-20, 20)
ax.set_ylim(-1.5, 1.5)
# ax.legend()
return xp, yps, pmean, pcov
def draw_optplot(plot_data_lists, ax, func_id):
(xmin, xmax) = (1, len(plot_data_lists[0]) - 1)
ax.set_xlim(xmin, xmax)
ax.set_ylim(-1.0, 1.5)
f_true = gpexample_plugin.get_ftrue(func_id)
for plot_data_list in plot_data_lists:
optima = [plotdata_optimum(plot_data) for plot_data in plot_data_list[1:]]
optvals = [f_true(x) for (x,y) in optima]
ax.plot(range(1, len(optvals) + 1), optvals, **plotparams.GP_OPTPLOT_SERIES)
local_optval_lines = [
Line2D(
[0, xmax], [local_optval, local_optval],
**plotparams.GP_OPTPLOT_TRUE_LOCAL_OPT)
for (_, local_optval) in gpexample_plugin.get_true_local_optima(func_id)
]
for l in local_optval_lines:
ax.add_artist(l)
ax.set_title('Estimated Optimum Over Time', **plotparams.GP_OPTPLOT_TITLE)
ax.set_xlabel('Iteration', **plotparams.GP_OPTPLOT_XLABEL)
ax.set_ylabel('$f_{\\rm true}(\\widehat{x})$', **plotparams.GP_OPTPLOT_YLABEL)
ax.legend(handles=local_optval_lines[:1])
(func_id, plot_data) = pickle.load(f)
sns.set(font_scale=3)
fig = plt.figure()
fig.set_dpi(30)
fig.set_figwidth(1.5 * plotparams.GP_PLOT_WIDTH)
fig.set_figheight(plotparams.GP_PLOT_HEIGHT)
gs = gridspec.GridSpec(1, 2, width_ratios=[2,1])
plot_ax = fig.add_subplot(gs[0])
legend_ax = fig.add_subplot(gs[1])
decorations = []
decorations.append(plotutils.centered_rectangle(
create_bayesopt_gp_plots.plotdata_optimum(plot_data),
**plotparams.GP_OPTIMUM))
f_true = gpexample_plugin.get_ftrue(func_id)
def generate_candidate():
return np.random.uniform(-20, 20)
if not ns.omit_candidates:
proposed_xs = [generate_candidate() for _ in range(8)]
plot_ax.scatter(proposed_xs, map(f_true, proposed_xs), **plotparams.GP_UNIFORM_XCANDIDATE)
create_bayesopt_gp_plots.draw_plot(plot_data, plot_ax, func_id, decorations)
make_legend(legend_ax)
outfname = ns.outfile
print "Saving figure %s" % (outfname,)
fig.savefig(outfname, dpi=fig.dpi,bbox_inches='tight')
