def test_hist_from_returned_objects(self):
plt.figure()
# UniformAPPDSampler
t = time.time()
sampler = UniformAPPDSampler(
ranges=self.ranges,
cal_width=self.width,
cal_height=self.height,
reference=self.reference,
temperature=5.0,
appd_range_bins=20,
init_jobs=4,
appd_range_dicovery_samples=HISTOGRAM_TEST_SAMPLES,
width=int(self.width / 4),
height=int(self.height / 4),
min_cropped_size=(int(self.width / 4 / 1.5),
int(self.height / 4 / 1.5)))
sampler = ParallelBufferedSampler(sampler=sampler,
n_jobs=4,
buffer_size=50)
print(
"ParallelBufferedSampler with UniformAPPDSampler initialization time: %.02f"
% (time.time() - t))
appds = list()
t = time.time()
for i in range(HISTOGRAM_TEST_SAMPLES):
# The sampler will enforce the min correct size
c = sampler.next()
appds.append(
c.appd(reference=self.reference,
width=self.width,
height=self.height,
normalized=True))
_ = c.rectify(image=self.test_image, mode='preserving')
print(
"ParallelBufferedSampler with UniformAPPDSampler, APPD calculation and rectification for %d normalized samples (full res): %.02f, AVG: %.04f"
% (HISTOGRAM_TEST_SAMPLES, time.time() - t,
(time.time() - t) / HISTOGRAM_TEST_SAMPLES))
plt.hist(appds, ec="k")
sampler.stop()
plt.grid()
plt.savefig(
path_here(
'ParallelBufferedSampler_UniformAPPDSampler_test_hist_from_returned_objects.png'
))
plt.close()
# Save a plot that shows the sample sequence (useful for debugging bad random initialization)
plt.figure()
plt.plot(appds, 'x')
plt.grid()
plt.savefig(
path_here(
'ParallelBufferedSampler_UniformAPPDSampler_test_hist_from_returned_objects_sequence.png'
))
plt.close()
def test_ParallelBufferedSampler(self):
plt.figure()
# UniformAPPDSampler
t = time.time()
sampler = UniformAPPDSampler(
ranges=self.ranges,
cal_width=self.width,
cal_height=self.height,
reference=self.reference,
temperature=5.0,
appd_range_bins=10,
init_jobs=4,
appd_range_dicovery_samples=HISTOGRAM_TEST_SAMPLES,
width=int(self.width / 4),
height=int(self.height / 4),
min_cropped_size=(int(self.width / 4 / 1.5),
int(self.height / 4 / 1.5)))
sampler = ParallelBufferedSampler(sampler=sampler,
n_jobs=4,
buffer_size=50)
print(
"ParallelBufferedSampler with UniformAPPDSampler initialization time: %.02f"
% (time.time() - t))
t = time.time()
bins, hist_edges, appds = sampler.histogram(
self.reference,
n_samples=HISTOGRAM_TEST_SAMPLES,
return_values=True,
n_bins=10,
width=int(self.width / 4),
height=int(self.height / 4),
normalized=True)
print(
"ParallelBufferedSampler with UniformAPPDSampler time for %d normalized samples (1/16 res): %.02f, AVG: %.04f"
% (HISTOGRAM_TEST_SAMPLES, time.time() - t,
(time.time() - t) / HISTOGRAM_TEST_SAMPLES))
plt.bar(hist_edges[:-1],
bins,
width=np.diff(hist_edges),
ec="k",
align="edge",
label="UniformAPPDSampler",
alpha=0.5)
sampler.stop()
plt.grid()
plt.legend()
plt.savefig(
path_here(
'ParallelBufferedSampler_UniformAPPDSampler_test_histogram_normalized_small.png'
))
plt.close()
# Save a plot that shows the sample sequence
plt.figure()
plt.plot(appds, 'x')
plt.grid()
plt.savefig(
path_here(
'ParallelBufferedSampler_UniformAPPDSampler_test_histogram_normalized_sequence.png'
))
plt.close()
def test_compare_noncached_and_cached_TriangularParallelBufferedSampler(
self):
plt.figure()
# UniformAPPDSampler
sampler = UniformAPPDSampler(
ranges=self.ranges,
cal_width=self.width,
cal_height=self.height,
sampler='triangular',
reference=self.reference,
temperature=5.0,
appd_range_bins=20,
init_jobs=4,
appd_range_dicovery_samples=HISTOGRAM_TEST_SAMPLES,
width=int(self.width / 4),
height=int(self.height / 4),
min_cropped_size=(int(self.width / 4 / 1.5),
int(self.height / 4 / 1.5)))
t = time.time()
# # NONCACHED -----------------------------------------
# noncached_sampler = ParallelBufferedSampler(sampler=sampler, n_jobs=4, buffer_size=50)
# print("Non-cached TriangularParallelBufferedSampler with UniformAPPDSampler initialization time: %.02f" % (time.time()-t))
#
# appds = list()
# t = time.time()
# for i in range(2*HISTOGRAM_TEST_SAMPLES):
# # The sampler will enforce the min correct size
# c = noncached_sampler.next()
# appds.append(c.appd(reference=self.reference, width=self.width, height=self.height, normalized=True))
# _ = c.rectify(image=self.test_image, mode='preserving')
# print("Non-cached TriangularParallelBufferedSampler with UniformAPPDSampler, APPD calculation and rectification for %d normalized samples (full res): %.02f, AVG: %.04f" %
# (HISTOGRAM_TEST_SAMPLES, time.time()-t, (time.time()-t)/HISTOGRAM_TEST_SAMPLES))
# plt.hist(appds, ec="k", label='non-cached, t=%ds' % int(time.time()-t))
# noncached_sampler.stop()
# CACHED -----------------------------------------
t = time.time()
cached_sampler = ParallelBufferedSampler(sampler=sampler,
n_jobs=1,
buffer_size=5,
cache_size=200)
print(
"Cached TriangularParallelBufferedSampler with UniformAPPDSampler initialization time: %.02f"
% (time.time() - t))
appds = list()
t = time.time()
for i in range(2 * HISTOGRAM_TEST_SAMPLES):
# The sampler will enforce the min correct size
c = cached_sampler.next()
appds.append(
c.appd(reference=self.reference,
width=self.width,
height=self.height,
normalized=True))
_ = c.rectify(image=self.test_image, mode='preserving')
print(
"Cached TriangularParallelBufferedSampler with UniformAPPDSampler, APPD calculation and rectification for %d normalized samples (full res): %.02f, AVG: %.04f"
% (HISTOGRAM_TEST_SAMPLES, time.time() - t,
(time.time() - t) / HISTOGRAM_TEST_SAMPLES))
plt.hist(appds, ec="k", label='cached, t=%ds' % int(time.time() - t))
cached_sampler.stop()
# ----------------------------------------------
plt.grid()
plt.legend()
plt.savefig(
path_here(
'non_cached_vs_cached_TriangularParallelBufferedSampler.png'))
plt.close()
plt.figure()
plt.plot(appds, 'x')
plt.grid()
plt.savefig(
path_here(
'non_cached_vs_cached_TriangularParallelBufferedSampler_sequence.png'
))
plt.close()