def plot_step_per_episode(model):
"""
:param model:
:return:
"""
out_dir = 'out/' + model + '/img/'
check_dir(out_dir)
input_dir = 'out/' + model + '/step-per-episode.txt'
with open(input_dir, 'r') as f:
lines = f.readlines()[1:]
lines = np.array(lines)
lines = np.char.strip(lines, '\n')
lines = np.array([x.split(',') for x in lines])
episodes = lines[:, 1].astype(np.int)
episode_counter = Counter(episodes) # number of steps for episode
mavg_episode_counter = moving_average(np.array(
list(episode_counter.values())),
window=100)
plt.figure()
plt.plot(np.array(list(episode_counter.keys()))[99:], mavg_episode_counter)
plt.xlabel('episode', fontsize=11)
plt.ylabel('step', fontsize=11)
plt.title('Steps per Episode: "' + model + '"', weight='bold', fontsize=12)
plt.savefig(out_dir + 'step-per-episode.pdf')
plt.close()
def plot_return(model):
"""
:param model:
"""
out_dir = 'out/' + model + '/img/'
check_dir(out_dir)
input_dir = 'out/' + model + '/return.txt'
with open(input_dir, 'r') as f:
lines = f.readlines()[1:]
lines = np.array(lines)
lines = np.char.strip(lines, '\n')
lines = np.array([x.split(',') for x in lines])
steps = lines[:, 0].astype(np.int)
scores = lines[:, 2].astype(np.float)
steps = steps.astype(np.float)
scores = scores.astype(np.float)
plt.figure()
plt.plot(steps, scores)
plt.xlabel('step', fontsize=11)
plt.ylabel('average score per play', fontsize=11)
ax = plt.gca()
ax.xaxis.set_major_formatter(EngFormatter())
plt.title('Evaluation Return: "' + model + '"', weight='bold', fontsize=12)
plt.savefig(out_dir + 'evaluation-return.pdf')
plt.close()
def plot_return_moving_average(model):
"""
:param model:
"""
out_dir = 'out/' + model + '/img/'
check_dir(out_dir)
input_dir = 'out/' + model + '/return_moving_average.txt'
with open(input_dir, 'r') as f:
lines = f.readlines()[1:]
lines = np.array(lines)
lines = np.char.strip(lines, '\n')
lines = np.array([x.split(',') for x in lines])
return_moving_averages = lines[:, 0].astype(np.float)
episodes = np.arange(30, len(return_moving_averages) + 30)
plt.figure()
plt.plot(episodes, return_moving_averages)
plt.xlabel('episodes', fontsize=11)
plt.ylabel('return per episode', fontsize=11)
ax = plt.gca()
ax.xaxis.set_major_formatter(EngFormatter())
plt.title('Training Return: "' + model + '"', weight='bold', fontsize=12)
plt.savefig(out_dir + 'training-return.pdf')
plt.close()
def plot_return_comparison(model1, model2, label1, label2):
"""
:param model1:
:param model2:
:param label1:
:parameter label2:
"""
out_dir = 'out/' + model2 + '/img/comparison/'
check_dir(out_dir)
input_dir1 = 'out/' + model1 + '/return.txt'
input_dir2 = 'out/' + model2 + '/return.txt'
with open(input_dir1, 'r') as f:
lines = f.readlines()[1:]
lines = np.array(lines)
lines = np.char.strip(lines, '\n')
lines = np.array([x.split(',') for x in lines])
steps1 = lines[:, 0]
steps1 = steps1.astype(np.float)
score1 = lines[:, 2]
score1 = score1.astype(np.float)
with open(input_dir2, 'r') as f:
lines = f.readlines()[1:]
lines = np.array(lines)
lines = np.char.strip(lines, '\n')
lines = np.array([x.split(',') for x in lines])
steps2 = lines[:, 0].astype(np.float)
score2 = lines[:, 2].astype(np.float)
plt.figure()
plt.plot(steps1[:len(steps2) - 1], score1[:len(score2) - 1], label=label1)
# plt.plot(steps1, score1, label=label1)
plt.plot(steps2, score2, label=label2)
plt.xlabel('episodes', fontsize=11)
plt.ylabel('score', fontsize=11)
plt.xlim([-10000, 310000])
plt.ylim([-1, 30])
ax = plt.gca()
ax.xaxis.set_major_formatter(EngFormatter())
plt.legend()
plt.title('Evaluation returns of models "' + model1 + '" and "' + model2 +
'"',
weight='bold',
fontsize=12)
plt.savefig(out_dir + 'scores-comparison-' + model1 + '-' + model2 +
'.pdf')
plt.close()
def plot_loss_comparison(models):
"""
:param models:
"""
out_dirs = []
steps = []
losses = []
mavg_losses = []
for i, e in enumerate(models):
out_dir = 'out/' + e + '/img/comparison/'
out_dirs.append(out_dir)
check_dir(out_dir)
input_dir = 'out/' + e + '/loss.txt'
with open(input_dir, 'r') as f:
lines = f.readlines()[1:]
lines = np.array(lines)
lines = np.char.strip(lines, '\n')
lines = np.array([x.split(',') for x in lines])
steps.append(lines[:, 0].astype(np.int))
losses.append(lines[:, 2].astype(np.float))
mavg_losses.append(moving_average(losses[i], window=50))
plt.figure()
for i, e in enumerate(models):
plt.plot(steps[i][50::1000], mavg_losses[i][::1000], label='Loss ' + e)
plt.xlabel('episodes', fontsize=11)
plt.ylabel('loss', fontsize=11)
ax = plt.gca()
ax.xaxis.set_major_formatter(EngFormatter())
ax.yaxis.set_major_formatter(EngFormatter())
plt.legend()
plt.title('Temporal-difference errors', weight='bold', fontsize=12)
for i in range(len(models)):
m = str(models).replace('[',
'').replace(']',
'').replace(', ',
'-').replace("'", "")
plt.savefig(out_dirs[i] + m + '-losses-comparison.pdf')
plt.close()
def plot_loss_moving_average(model):
"""
:param model:
:return:
"""
out_dir = 'out/' + model + '/img/'
check_dir(out_dir)
input_dir = 'out/' + model + '/loss.txt'
with open(input_dir, 'r') as f:
lines = f.readlines()[1:]
lines = np.array(lines)
lines = np.char.strip(lines, '\n')
lines = np.array([x.split(',') for x in lines])
steps = lines[:, 0].astype(np.int)
losses = lines[:, 2].astype(np.float)
losses = moving_average(losses, window=50)
plt.figure()
plt.plot(steps[50::1000], losses[::1000])
plt.xlabel('step', fontsize=11)
plt.ylabel('loss', fontsize=11)
ax = plt.gca()
ax.xaxis.set_major_formatter(EngFormatter())
plt.title('Temporal difference error: "' + model + '"',
weight='bold',
fontsize=12)
plt.savefig(out_dir + 'temporal-difference-error.pdf')
plt.close()