def MAB_regret_plot2(fname, xs_dict, colors=None, title=None, fig_kwargs=None, cut_time=None, name_filter=None, name_change=None):
if colors is None:
colors = sns.color_palette('Set1', len(xs_dict))
sns.set(style="white", font_scale=1.4, rc={"lines.linewidth": 20})
plt.figure(**with_default(fig_kwargs, dict()))
for i, (name, value_matrix) in enumerate(xs_dict.items()):
if name_filter is not None:
if not name_filter(name):
continue
mean_x = np.mean(value_matrix, axis=0)
time_points = sparse_index(with_default(cut_time, len(mean_x)))
plt.plot(time_points, mean_x[time_points], lw=1, label=name_change[name], color=colors[i], linestyle='-' if 'once' in name else "--")
plt.xlabel('Trials')
plt.ylabel('Cum. Regrets')
plt.xticks([])
plt.yticks([])
plt.ylim([-5, 100])
if title is not None:
plt.title(title)
for l in plt.gca().lines:
print(l.get_linewidth())
plt.setp(l, linewidth=2)
sns.despine()
plt.savefig(fname, bbox_inches='tight', pad_inches=0.02)
plt.close()
def MAB_optimal_probability_plot2(fname, arm_freqs, colors=None, title=None, fig_kwargs=None, cut_time=None, name_filter=None, name_change=None, no_legend=False):
""" Draw figures showing optimal arm selection probability (based on multiple runs) """
if colors is None:
colors = sns.color_palette('Set1', len(arm_freqs))
sns.set(style="white", font_scale=1.4, rc={"lines.linewidth": 20})
plt.figure(**with_default(fig_kwargs, dict()))
for i, (name, arm_freq) in enumerate(arm_freqs.items()):
if name_filter is not None:
if not name_filter(name):
continue
time_points = sparse_index(with_default(cut_time, len(arm_freq)))
plt.plot(time_points, arm_freq[time_points], lw=1, label=name_change[name], color=colors[i], linestyle='-' if 'once' in name else "--")
if not no_legend:
plt.legend(loc=4)
plt.xlabel('Trials')
plt.ylabel('Opt. Arm Prob.')
plt.xticks([])
plt.ylim(-0.05, 1.02)
if title is not None:
plt.title(title)
for l in plt.gca().lines:
print(l.get_linewidth())
plt.setp(l, linewidth=2)
for line in plt.gca().legend().get_lines():
line.set_linewidth(2)
sns.despine()
plt.savefig(fname, bbox_inches='tight', pad_inches=0.02)
plt.close()
def naked_MAB_optimal_probability_plot(axes: Axes,
arm_freqs,
cut_time,
legend=False,
hide_ylabel=False,
hide_yticklabels=False,
**_kwargs):
for i, (name, arm_freq) in list(enumerate(arm_freqs.items())):
time_points = sparse_index(with_default(cut_time, len(arm_freq)), 200)
axes.plot(time_points,
arm_freq[time_points],
lw=1,
label=name.split(' ')[0] if '(TS)' in name else None,
color=COLORS[i],
linestyle='-' if '(TS)' in name else '--')
if legend:
axes.legend(loc=4, frameon=False)
axes.set_xlabel('Trials')
if not hide_ylabel:
axes.set_ylabel('Probability')
axes.get_yaxis().set_label_coords(-0.15, 0.5)
axes.set_yticks([0, 0.5, 1.0])
if hide_yticklabels:
axes.set_yticklabels([])
axes.set_ylim(-0.05, 1.02)
def sup_KL(mu_ref, divergence, lower=None):
""" Find largest mu that satisfies KL(mu_ref, mu) <= divergence """
if divergence <= 0:
return mu_ref
if KL(mu_ref, 1.0) <= divergence:
return 1.0
return brenth(lambda x: KL(mu_ref, x) - divergence,
with_default(lower, mu_ref), 1)
def __init__(self,
G: CausalDiagram,
F=None,
P_U=None,
D=None,
more_U=None):
self.G = G
self.F = F
self.P_U = P_U
self.D = with_default(D, defaultdict(lambda: (0, 1)))
self.more_U = set() if more_U is None else set(more_U)
self.query00 = functools.lru_cache(1024)(self.query00)
def naked_MAB_regret_plot(axes: Axes,
xs_dict,
cut_time,
band_alpha=0.1,
legend=False,
hide_ylabel=False,
adjust_ymax=1,
hide_yticklabels=False,
**_kwargs):
for i, (name, value_matrix) in list(enumerate(xs_dict.items())):
mean_x = np.mean(value_matrix, axis=0)
sd_x = np.std(value_matrix, axis=0)
lower, upper = mean_x - sd_x, mean_x + sd_x
time_points = sparse_index(with_default(cut_time, len(mean_x)), 200)
axes.plot(time_points,
mean_x[time_points],
lw=1,
label=name.split(' ')[0] if '(TS)' in name else None,
color=COLORS[i],
linestyle='-' if '(TS)' in name else '--')
axes.fill_between(time_points,
lower[time_points],
upper[time_points],
color=COLORS[i],
alpha=band_alpha,
lw=0)
if legend:
axes.legend(loc=2, frameon=False)
if not hide_ylabel:
axes.set_ylabel('Cum. Regrets')
axes.get_yaxis().set_label_coords(-0.15, 0.5)
if adjust_ymax != 1:
ymin, ymax = axes.get_ylim()
axes.set_ylim(ymin, ymax * adjust_ymax)
if hide_yticklabels:
axes.set_yticklabels([])