def end_training (self) :
transcripts.printf_information ('Ending training...')
self.end_training_1 ()
self.training_stepper.destroy ()
self.training_stepper = None
self.executed_training = True
transcripts.printf_information ('Ended training.')
transcripts.printf_output ('Prototypes == %d;', self.get_prototype_count ())
transcripts.printf_output ('Vector length == %d;', self.get_vector_length ())
def print_similarity_matrix (classifier, similarity) :
matrix = compute_similarity_matrix (classifier, similarity)
matrix_ = list ()
for line in matrix :
line_ = list ()
for similarity in line :
line_.append ("%.4f" % (similarity))
line_ = ' '.join (line_)
matrix_.append (line_)
for line_ in matrix_ :
transcripts.printf_output (line_)
def statistic_1 (self) :
self.class_entropies = dict ()
self.classes_entropy = 0.0
self.class_purities = dict ()
self.classes_purity = 0.0
for clas, partitions in self.class_to_partition.items () :
size = self.classes_size[clas]
entropy = 0.0
max_count = 0
for partition in partitions :
count = partitions[partition]
entropy += float (count) / float (size) * math.log (float (count) / float (size))
if max_count < count :
max_count = count
entropy *= -1.0 / math.log (len (self.classes))
self.class_entropies[clas] = entropy
self.classes_entropy += float (size) / float (self.patterns) * entropy
purity = 1.0 / float (size) * float (max_count)
self.class_purities[clas] = purity
self.classes_purity += float (size) / float (self.patterns) * purity
transcripts.printf_output ('Entropy == %.4f:', self.classes_entropy)
class_entropies = self.class_entropies.items ()
class_entropies.sort (key = lambda pair : pair[0])
for clas, entropy in class_entropies :
transcripts.printf_output ('\-- %s == %.4f;', clas, entropy)
transcripts.printf_output ('Purity == %.4f:', self.classes_purity)
class_purities = self.class_purities.items ()
class_purities.sort (key = lambda pair : pair[0])
for clas, purity in class_purities :
transcripts.printf_output ('\-- %s == %.4f;', clas, purity)
def load_from_stream (self, stream) :
prototypes = vectors.input_ (stream)
if len (prototypes) > 0 :
vector_length = len (prototypes[0])
for prototype in prototypes :
if vector_length != len (prototype) :
raise Exception ()
else :
vector_length = None
self.pre_load (prototypes, vector_length)
self.prototypes = prototypes
self.vector_length = vector_length
self.post_load ()
transcripts.printf_information ('Loaded from stream.')
transcripts.printf_output ('Prototypes == %d;', self.get_prototype_count ())
transcripts.printf_output ('Vector length == %d;', self.get_vector_length ())
def end_training_1 (self) :
prototypes = list ()
for index in xrange (self.prototype_count) :
numerator = self.prototype_numerators[index]
denominator = self.prototype_denominators[index]
if denominator == 0 :
denominator = 1
prototype = numerator / denominator
prototype /= vectors.magnitude (prototype)
prototypes.append (prototype)
if self.prototypes is not None :
prototypes_delta = 0
for index in xrange (self.prototype_count) :
old_prototype = self.prototypes[index]
new_prototype = prototypes[index]
prototype_delta = self.distance (old_prototype, new_prototype)
prototypes_delta += prototype_delta
self.training_delta = prototypes_delta
transcripts.printf_output ('Training delta == %.4f;', self.training_delta)
else :
self.training_delta = None
self.prototypes = prototypes
def statistic_2 (self) :
transcripts.printf_output ('Distribution:')
classes = self.class_to_partition.items ()
classes.sort (key = lambda pair : pair[0])
for clas, partitions in classes :
transcripts.printf_output ('\-- %s -> %d:', clas, self.classes_size[clas])
if partitions is not None :
partitions = partitions.items ()
partitions.sort (key = lambda pair : pair[1], reverse = True)
for partition, count in partitions :
transcripts.printf_output (' \-- %s -> %d;', partition, count)
def end_validation_1 (self) :
self.ssp = float (self.ss) / float (self.tt)
self.sdp = float (self.sd) / float (self.tt)
self.dsp = float (self.ds) / float (self.tt)
self.ddp = float (self.dd) / float (self.tt)
self.rand_statistic = float (self.ss + self.dd) / float (self.ss + self.sd + self.ds + self.dd)
self.jaccard_coefficient = float (self.ss) / float (self.ss + self.sd + self.ds)
self.folkes_mallows_index = math.sqrt ((float (self.ss) / float (self.ss + self.sd)) * (float (self.ss) / float (self.ss + self.ds)))
transcripts.printf_output ('SS (C1 == C2, P1 == P2) == %.4f, %d;', self.ssp, self.ss)
transcripts.printf_output ('SD (C1 == C2, P1 != P2) == %.4f, %d;', self.sdp, self.sd)
transcripts.printf_output ('DS (C1 != C2, P1 == P2) == %.4f, %d;', self.dsp, self.ds)
transcripts.printf_output ('DD (C1 != C2, P1 != P2) == %.4f, %d;', self.ddp, self.dd)
transcripts.printf_output ('Rand statistic == %.4f;', self.rand_statistic)
transcripts.printf_output ('Jaccard coefficient == %.4f;', self.jaccard_coefficient)
transcripts.printf_output ('Folkes and Mallows index == %.4f;', self.folkes_mallows_index)
def print_statistics (self) :
transcripts.printf_output ('Prototype to prototype distances:')
prototype_identifiers = list (self.prototype_identifiers)
prototype_identifiers.sort ()
for prototype_identifier in self.prototype_identifiers :
transcripts.printf_output ('\-- %s:', prototype_identifier)
transcripts.printf_output (' \-- sum == %.4f;', self.prototype_to_prototype_distance_sum[prototype_identifier])
transcripts.printf_output (' \-- avg == %.4f;', self.prototype_to_prototype_distance_avg[prototype_identifier])
transcripts.printf_output (' \-- min == %.4f;', self.prototype_to_prototype_distance_min[prototype_identifier])
transcripts.printf_output (' \-- max == %.4f;', self.prototype_to_prototype_distance_max[prototype_identifier])
transcripts.printf_output ('Pattern to prototype distances:')
for prototype_identifier in self.prototype_identifiers :
transcripts.printf_output ('\-- %s:', prototype_identifier)
transcripts.printf_output (' \-- count == %d;', self.pattern_to_prototype_count[prototype_identifier])
transcripts.printf_output (' \-- sum == %.4f;', self.pattern_to_prototype_distance_sum[prototype_identifier])
transcripts.printf_output (' \-- avg == %.4f;', self.pattern_to_prototype_distance_avg[prototype_identifier])
transcripts.printf_output (' \-- min == %.4f;', self.pattern_to_prototype_distance_min[prototype_identifier])
transcripts.printf_output (' \-- max == %.4f;', self.pattern_to_prototype_distance_max[prototype_identifier])
transcripts.printf_output ('Pattern to prototypes distances:')
transcripts.printf_output ('\-- count == %d;', self.pattern_count)
transcripts.printf_output ('\-- sum == %.4f;', self.pattern_to_prototypes_distance_sum)
transcripts.printf_output ('\-- avg == %.4f;', self.pattern_to_prototypes_distance_avg)
transcripts.printf_output ('\-- min == %.4f;', self.pattern_to_prototypes_distance_min)
transcripts.printf_output ('\-- max == %.4f;', self.pattern_to_prototypes_distance_max)
transcripts.printf_output ('Pattern to center distances:')
transcripts.printf_output ('\-- count == %d;', self.pattern_count)
transcripts.printf_output ('\-- sum == %.4f;', self.pattern_to_center_distance_sum)
transcripts.printf_output ('\-- avg == %.4f;', self.pattern_to_center_distance_avg)
transcripts.printf_output ('\-- min == %.4f;', self.pattern_to_center_distance_min)
transcripts.printf_output ('\-- max == %.4f;', self.pattern_to_center_distance_max)
transcripts.printf_output ('PBM index == %.4f;', self.pbm_index)