def learn(self, evidence):
"""Quantize data into 'max_coincidence_count' clusters with maximal size
of 'max_distance'.
Args:
evidence: a numpy vector to be learned
"""
# input checking
if not utils.is_list(evidence) and not evidence:
raise ValueError("'evidence' has incorrect value")
# don't learn empty patterns
if self.ignore_background_pattern:
is_empty = np.min(evidence) == np.max(evidence) == self.background_color
if is_empty:
return False
if self.algorithm in ('product', 'sum'):
# keep only the indices of the maximal activation for each child
evidence = utils.get_max_indices(evidence)
# if there is no memorized coincidence, store it
if len(self.coincidences) == 0:
self.coincidences[0] = evidence
self.coincidences_stats[0] = 1
else:
# store the evidence iff d(evicence, coincidences) > maxDistance
# dist is a list of tuples [ (index, distance), ...]
dist = self._compute_dist_to_coincidences(evidence)
# if self.debug and min(dist, key=itemgetter(1))[1] > 0:
# print "-> The distance to the closest coincidence is %.2f" % min(dist, key=itemgetter(1))[1]
codebook_is_not_full = len(self.coincidences) < self.max_coincidence_count
nearest_coincidence = min(dist, key=itemgetter(1)) # tuple (id, distance)
if codebook_is_not_full and nearest_coincidence[1] > self.max_distance:
self.coincidences[len(self.coincidences)] = evidence # new coincidence
self.coincidences_stats[len(self.coincidences) - 1] = 1
# if self.debug: print "-> New coincidence added, count = %d" % len(self.coincidences)
else:
# Increase stats for the closest coincidence
# nearest_coincidence[0] is the index of the coincidence
self.coincidences_stats[nearest_coincidence[0]] += 1
def learn(self, evidence):
"""Quantize data into 'max_coincidence_count' clusters with maximal size
of 'max_distance'.
Args:
evidence: a numpy vector to be learned
"""
# input checking
if not utils.is_list(evidence) and not evidence:
raise ValueError("'evidence' has incorrect value")
# don't learn empty patterns
if self.ignore_background_pattern:
is_empty = np.min(evidence) == np.max(evidence) == self.background_color
if is_empty:
return False
if self.algorithm in ("product", "sum"):
# keep only the indices of the maximal activation for each child
evidence = utils.get_max_indices(evidence)
# if there is no memorized coincidence, store it
if len(self.coincidences) == 0:
self.coincidences[0] = evidence
self.coincidences_stats[0] = 1
else:
# store the evidence iff d(evicence, coincidences) > maxDistance
# dist is a list of tuples [ (index, distance), ...]
dist = self._compute_dist_to_coincidences(evidence)
# if self.debug and min(dist, key=itemgetter(1))[1] > 0:
# print "-> The distance to the closest coincidence is %.2f" % min(dist, key=itemgetter(1))[1]
codebook_is_not_full = len(self.coincidences) < self.max_coincidence_count
nearest_coincidence = min(dist, key=itemgetter(1)) # tuple (id, distance)
if codebook_is_not_full and nearest_coincidence[1] > self.max_distance:
self.coincidences[len(self.coincidences)] = evidence # new coincidence
self.coincidences_stats[len(self.coincidences) - 1] = 1
# if self.debug: print "-> New coincidence added, count = %d" % len(self.coincidences)
else:
# Increase stats for the closest coincidence
# nearest_coincidence[0] is the index of the coincidence
self.coincidences_stats[nearest_coincidence[0]] += 1
def testMaxIndices(self):
l = [[23, 1, 4], [2.4, 5, 11], [44, 22, 105, 18]]
r = utils.get_max_indices(l)
self.assertTrue(np.array_equal(r, np.array([0, 2, 2])))
def testMaxIndices(self):
l = [[23, 1, 4], [2.4, 5, 11], [44, 22, 105, 18]]
r = utils.get_max_indices(l)
self.assertTrue(np.array_equal(r, np.array([0, 2, 2])))