def return_avg_median(fig, df, args):
if not _keys_in_columns(df,
'avg_return',
'std_return',
'median_return',
verbose=args.verbose):
return False
data = DataFrame(dict(mean=df.avg_return, std=df.std_return))
draw_curve_from_data('mean_std',
fig.gca(),
data,
np.arange(len(df.avg_return)),
ax_calc=1,
curve_label='average',
x_label='iteration',
y_label='return',
legend_kwargs=dict(loc='upper left'))
plt.plot(np.arange(len(df.median_return)),
df.median_return,
label='median')
if _keys_in_columns(df, 'curr_policy_return', verbose=args.verbose):
# If the algorithm is a subclass of ParameterExploring
plt.plot(np.arange(len(df.curr_policy_return)),
df.curr_policy_return,
label='current')
def return_min_max_avg(fig, df, args):
if not _keys_in_columns(df, "avg_return", "min_return", "max_return", verbose=args.verbose):
return False
data = DataFrame(dict(mean=df.avg_return, min=df.min_return, max=df.max_return))
draw_curve_from_data(
"min_mean_max",
fig.gca(),
data,
np.arange(len(df.avg_return)),
ax_calc=1,
curve_label="average",
x_label="iteration",
y_label="return",
legend_kwargs=dict(loc="upper left"),
)
def return_min_max_avg(fig, df, args):
if not _keys_in_columns(
df, 'avg_return', 'min_return', 'max_return',
verbose=args.verbose):
return False
data = DataFrame(
dict(mean=df.avg_return, min=df.min_return, max=df.max_return))
draw_curve_from_data('min_mean_max',
fig.gca(),
data,
np.arange(len(df.avg_return)),
ax_calc=1,
curve_label='average',
x_label='iteration',
y_label='return',
legend_kwargs=dict(loc='upper left'))
def return_avg_median(fig, df, args):
if not _keys_in_columns(df, "avg_return", "std_return", "median_return", verbose=args.verbose):
return False
data = DataFrame(dict(mean=df.avg_return, std=df.std_return))
draw_curve_from_data(
"mean_std",
fig.gca(),
data,
np.arange(len(df.avg_return)),
ax_calc=1,
curve_label="average",
x_label="iteration",
y_label="return",
legend_kwargs=dict(loc="upper left"),
)
plt.plot(np.arange(len(df.median_return)), df.median_return, label="median")
if _keys_in_columns(df, "curr_policy_return", verbose=args.verbose):
# If the algorithm is a subclass of ParameterExploring
plt.plot(np.arange(len(df.curr_policy_return)), df.curr_policy_return, label="current")
def test_render_curve(data, x_grid):
fix, axs = plt.subplots(nrows=2, ncols=2)
draw_curve_from_data(
'mean_std', axs[0, 0], data, x_grid, ax_calc=1,
x_label='A', y_label='y', vline_level=None, title=None, show_legend=True, area_label='a',
plot_kwargs=dict(alpha=0.1, color='r', ls='--')
)
draw_curve_from_data(
'min_mean_max', axs[0, 1], data, x_grid, ax_calc=1,
x_label=None, y_label=None, vline_level=None, title=None, show_legend=True, curve_label='c',
)
draw_curve_from_data(
'mean_std', axs[1, 0], data, x_grid, ax_calc=1,
x_label='d', y_label=r'$y$', vline_level=None, title='Title', show_legend=True, curve_label='c',
plot_kwargs=dict(alpha=0.1)
)
draw_curve_from_data(
'min_mean_max', axs[1, 1], data, x_grid, ax_calc=1,
x_label=r'$\mu$', y_label='', vline_level=None, title='Title', show_legend=True, area_label='a',
plot_kwargs=dict(alpha=0.1, color='r')
)
def test_render_curve(data, x_grid):
fix, axs = plt.subplots(nrows=3, ncols=2)
draw_curve_from_data(
"mean_std",
axs[0, 0],
data,
x_grid,
ax_calc=1,
x_label="A",
y_label="y",
vline_level=None,
title=None,
show_legend=True,
area_label="a",
plot_kwargs=dict(alpha=0.1, color="r", ls="--"),
)
draw_curve_from_data(
"min_mean_max",
axs[0, 1],
data,
x_grid,
ax_calc=1,
x_label=None,
y_label=None,
vline_level=None,
title=None,
show_legend=True,
curve_label="c",
)
draw_curve_from_data(
"mean_std",
axs[1, 0],
data,
x_grid,
ax_calc=1,
x_label="d",
y_label=r"$y$",
vline_level=None,
title="Title",
show_legend=True,
curve_label="c",
plot_kwargs=dict(alpha=0.1),
)
draw_curve_from_data(
"min_mean_max",
axs[1, 1],
data,
x_grid,
ax_calc=1,
x_label=r"$\mu$",
y_label="",
vline_level=None,
title="Title",
show_legend=True,
area_label="a",
plot_kwargs=dict(alpha=0.1, color="r"),
)
draw_curve_from_data(
"ci_on_mean",
axs[2, 0],
data,
x_grid,
ax_calc=1,
title=None,
show_legend=True,
curve_label="100 reps",
area_label="ci",
cmp_kwargs=dict(num_reps=100, confidence_level=0.5),
plot_kwargs=dict(color="c"),
)
draw_curve_from_data(
"ci_on_mean",
axs[2, 1],
data,
x_grid,
ax_calc=1,
title=None,
show_legend=True,
curve_label="10000 reps",
area_label="ci",
cmp_kwargs=dict(num_reps=10000),
plot_kwargs=dict(alpha=0.4, color="c"),
)
if args.verbose:
print(f"b_Jhat_{n}:\t\t{b_Jhat_n}\n")
print(f"At the last iteration (n={num_iter})")
print(f"mean G_n: {np.mean(G_n_hist, axis=0)[-1]}")
print(f"mean G_true: {np.mean(G_true_hist, axis=0)[-1]}")
print(f"mean b_Jhat_n: {np.mean(b_Jhat_n_hist, axis=0)[-1]}\n")
# Plot
os.makedirs(ex_dir, exist_ok=True)
fig_n, ax = plt.subplots(1, figsize=fig_size, constrained_layout=True)
draw_curve_from_data(
"ci_on_mean",
ax,
n_M_hist,
np.arange(1, num_iter + 1),
ax_calc=0,
x_label="number of domains $n$",
y_label="samples per domain",
curve_label="$n_M$",
)
draw_curve_from_data(
"ci_on_mean",
ax,
n_V_hist,
np.arange(1, num_iter + 1),
ax_calc=0,
x_label="number of domains $n$",
y_label="samples per domain",
curve_label="$n_V$",
)
ax.plot(np.arange(1, num_iter + 1),