def plot_evolution_happiness(evo_training, method_id, info_params, final=True):
avg_happiness = get_avg_n_points(evo_training['evo_avg_happiness'],
n_points=100)
min_happiness = get_avg_n_points(evo_training['evo_min_happiness'],
n_points=100)
max_happiness = get_avg_n_points(evo_training['evo_max_happiness'],
n_points=100)
x = range(len(avg_happiness))
fig = plt.figure()
plt.plot(x, avg_happiness)
plt.plot(x, min_happiness, linestyle=':', color='lightcoral')
plt.plot(x, max_happiness, linestyle=':', color='lightgreen')
plt.title('Min/Avg/Max Happiness over time (smoothed)')
plt.ylim([-0.05, 1.05])
plt.xlabel('Episode (percentile) \n ' + info_params)
plt.ylabel('Min/Avg/Max Happiness (smoothed)')
plt.grid(True)
plt.tight_layout()
plt.savefig(get_directory_figures(final) +
'{}__happiness.png'.format(method_id),
format='png',
dpi=500)
# plt.show()
plt.close(fig)
def plot_evolution_steps(evo_training, method_id, nmax_steps, info_params, final=True):
fig = plt.figure()
avg_steps = get_avg_n_points(evo_training['evo_n_steps'], n_points=100)
x = range(len(avg_steps))
plt.plot(x, avg_steps)
min_steps = get_min_n_points(evo_training['evo_n_steps'], n_points=min(100, len(evo_training['evo_n_steps'])))
x = range(len(min_steps))
plt.plot(x, min_steps, linestyle=':', color='lightcoral')
max_steps = get_max_n_points(evo_training['evo_n_steps'], n_points=min(100, len(evo_training['evo_n_steps'])))
x = range(len(max_steps))
plt.plot(x, max_steps, linestyle=':', color='lightgreen')
plt.title('Episode Length over time (smoothed)')
plt.axhline(nmax_steps, color='r')
plt.axhline(0, color='b')
plt.ylim([-10, nmax_steps * 1.05])
plt.xlabel('Episode (percentile) \n ' + info_params)
plt.ylabel('Episode Length (smoothed)')
plt.grid(True)
plt.tight_layout()
plt.savefig(get_directory_figures(final) + '{}__steps.png'.format(method_id), format='png', dpi=500)
# plt.show()
plt.close(fig)
def plot_evolution_reward(evo_training, method_id, info_params, final=True):
avg_reward = get_avg_n_points(evo_training['evo_avg_reward_per_step'], n_points=100)
min_reward = get_avg_n_points(evo_training['evo_min_reward_per_step'], n_points=100)
max_reward = get_avg_n_points(evo_training['evo_max_reward_per_step'], n_points=100)
x = range(len(avg_reward))
fig = plt.figure()
plt.plot(x, avg_reward)
plt.plot(x, min_reward, linestyle=':', color='lightcoral')
plt.plot(x, max_reward, linestyle=':', color='lightgreen')
plt.title('Evolution of Min/Avg/Max Reward \n per step per episode over time (smoothed)')
plt.xlabel('Episode (percentile) \n ' + info_params)
plt.ylabel('Min/Avg/Max Reward \n per step per episode (smoothed)')
plt.grid(True)
plt.tight_layout()
plt.savefig(get_directory_figures(final) + '{}__reward.png'.format(method_id), format='png', dpi=500)
# plt.show()
plt.close(fig)
def plot_alpha(evo_alpha, method_id, info_params, final=True):
y = get_avg_n_points(evo_alpha, n_points=100)
x = range(len(y))
fig = plt.figure()
plt.plot(x, y)
plt.yscale('log')
plt.title('Alpha')
plt.xlabel('Episode (percentile) \n ' + info_params)
plt.ylabel('Alpha (log scale)')
plt.grid(True)
plt.tight_layout()
plt.savefig(get_directory_figures(final) +
'{}__alpha.png'.format(method_id),
format='png',
dpi=500)
# plt.show()
plt.close(fig)
def plot_comparison_evolution_steps(evo_training__evo_n_steps, nmax_steps,
info_params):
fig = plt.figure()
cnt = 0
for method in list(evo_training__evo_n_steps.keys()):
y = get_avg_n_points(evo_training__evo_n_steps[method], n_points=100)
x = range(len(y))
plt.plot(x,
y,
label=method,
marker='',
color=COLORS[cnt],
linewidth=1,
alpha=0.75)
cnt += 1
plt.title('Episode Length over time (smoothed)')
plt.axhline(nmax_steps, color='r')
plt.axhline(0, color='b')
plt.ylim([-10, nmax_steps * 1.05])
plt.xlabel('Episode (percentile) \n ' + info_params)
plt.ylabel('Episode Length (Smoothed)')
plt.legend(bbox_to_anchor=(0.5, -0.10),
loc="lower center",
bbox_transform=fig.transFigure,
ncol=4,
fancybox=True,
shadow=True,
borderpad=1)
plt.grid(True)
plt.tight_layout()
plt.savefig('data/figures/Comparison__steps.png',
format='png',
dpi=1000,
bbox_inches='tight')
# plt.show()
plt.close(fig)