def _itr_load_testing_samples(self, itr):
fname = os.path.join(self._itr_dir(itr),
'testing_samples_itr_{0}.npz'.format(itr))
if os.path.exists(fname):
return Sample.load(fname)
elif not os.path.exists(self._itr_dir(itr)):
raise Exception()
else:
return None
def _load_samples(self):
samples_itrs = []
itr = 0
while True:
try:
samples_itrs.append(self._itr_load_samples(itr))
itr += 1
except:
break
self._logger.info('Loaded {0} iterations of samples'.format(
len(samples_itrs)))
### load initial dataset
init_data_folder = params['probcoll'].get('init_data', None)
if init_data_folder is not None:
if itr == 0:
samples_itrs.append([])
num_init_samples = 0
fnames = [
os.path.join(init_data_folder, fname)
for fname in os.listdir(init_data_folder)
]
for fname in fnames:
try:
samples = Sample.load(fname)
except:
continue
self._logger.debug('Loaded samples from {0}'.format(fname))
samples_itrs[0] += samples
num_init_samples += len(samples)
self._logger.info('Loaded initial dataset of {0} samples'.format(
num_init_samples))
return samples_itrs
def plot_final_policy_combined(exp_groups,
des_mps,
des_speed,
min_speed,
ax=None,
fs=20.,
save_path=None):
dt = 0.5
if ax is None:
f, ax = plt.subplots(1, 1, figsize=(5, 3.5))
else:
f = ax.get_figure()
exp_groups.insert(0, [{
'folder': os.path.join(EXP_FOLDER, 'random'),
'label': 'Random policy',
'color': cm.Greys(1.0)
}])
for exp_group_num, exp_group in enumerate(exp_groups):
distance_travelled = []
for exp in exp_group:
samples_folder = os.path.join(exp['folder'],
'prediction/final_policy/')
samples_fnames = [
os.path.join(samples_folder, fname)
for fname in os.listdir(samples_folder) if fname[-4:] == '.npz'
]
assert (len(samples_fnames) == 1)
samples_fname = samples_fnames[0]
samples = Sample.load(samples_fname)
for sample in samples:
distance = dt * (des_mps / des_speed) * (sample.get_U()[:, 1] -
min_speed).sum()
distance_travelled.append(distance)
width = 0.8 / len(exp_groups)
ax.bar([exp_group_num * width], [np.mean(distance_travelled)],
yerr=[np.std(distance_travelled)],
width=width * 0.8,
error_kw=dict(lw=4, capsize=5, capthick=2),
color=exp_group[0]['color'],
ecolor=cm.Greys(0.5),
label=exp_group[0]['label'])
# for exp_group in exp_groups:
# ax.plot([], [], lw=5., label=exp_group[0]['label'], color=exp_group[0]['color'])
lgd = ax.legend(loc='upper center',
fontsize=fs * 0.8,
bbox_to_anchor=(1.35, 0.7))
ax.set_xlim((-0.05, width * len(exp_groups)))
ax.set_ylim((-0.3, ax.get_ylim()[1]))
ax.set_ylabel(r'Distance travelled (m)', fontsize=fs)
for tick in ax.xaxis.get_major_ticks() + ax.xaxis.get_minor_ticks():
tick.label1.set_visible(False)
for tick in ax.yaxis.get_major_ticks():
tick.label.set_fontsize(fs)
if save_path is not None:
f.savefig(save_path, bbox_extra_artists=(lgd, ), bbox_inches='tight')
def plot_final_policy(exp_groups,
des_mps,
des_speed,
min_speed,
ax=None,
fs=20.,
save_path=None):
dt = 0.5
if ax is None:
f, ax = plt.subplots(1, 1, figsize=(10, 3.5))
else:
f = ax.get_figure()
exp_groups.insert(0, [{
'folder': os.path.join(EXP_FOLDER, 'random'),
'label': 'Random policy',
'color': cm.Greys(0.5)
}])
for exp_num, exp in enumerate(flatten_list(exp_groups)):
samples_folder = os.path.join(exp['folder'],
'prediction/final_policy/')
samples_fnames = [
os.path.join(samples_folder, fname)
for fname in os.listdir(samples_folder) if fname[-4:] == '.npz'
]
assert (len(samples_fnames) == 1)
samples_fname = samples_fnames[0]
samples = Sample.load(samples_fname)
distance_travelled = []
for sample in samples:
distance = dt * (des_mps / des_speed) * (sample.get_U()[:, 1] -
min_speed).sum()
distance_travelled.append(distance)
# ax.plot(range(len(distance_travelled)), distance_travelled,
# linestyle='',
# marker=exp['marker'],
# color=exp['color'],
# mew=exp['mew'])
ax.bar(np.arange(len(distance_travelled)) + 0.15 * exp_num + 0.1,
distance_travelled,
width=0.15,
color=exp['color'])
for exp_group in exp_groups:
ax.plot([], [],
lw=5.,
label=exp_group[0]['label'],
color=exp_group[0]['color'])
lgd = ax.legend(loc='upper right', fontsize=fs * 0.8)
ax.set_xlabel(r'Starting position', fontsize=fs)
ax.set_ylabel(r'Distance travelled (m)', fontsize=fs)
N = len(distance_travelled)
ax.set_xticks(np.arange(N) + 0.5)
ax.set_xticklabels([str(x) for x in range(N)])
# Hide major tick labels and customize minor tick labels
ax.xaxis.set_minor_formatter(ticker.NullFormatter())
ax.xaxis.set_major_locator(ticker.FixedLocator(np.arange(N) + 0.5))
ax.xaxis.set_major_formatter(
ticker.FixedFormatter([r'{0}'.format(str(x)) for x in range(N)]))
for tick in ax.xaxis.get_major_ticks() + ax.yaxis.get_major_ticks():
tick.label.set_fontsize(fs)
if save_path is not None:
f.savefig(save_path, bbox_extra_artists=(lgd, ), bbox_inches='tight')
def _itr_load_samples(self, itr):
fname = os.path.join(self._itr_dir(itr),
'samples_itr_{0}.npz'.format(itr))
return Sample.load(fname)