def test_with_two_characters(self):
#test with just two letters so A and B are copied to a
#special dir that is deleted after the test
base_dir = File("../../character_examples")
test_dir = File(base_dir,"test")
a_dir = File(base_dir,"A")
b_dir = File(base_dir,"B")
shutil.copytree(a_dir.getPath(), File(test_dir,"A").getPath())
shutil.copytree(b_dir.getPath(), File(test_dir,"B").getPath())
extracor = SimpleImageFeatureExtractor(nr_of_divisions=7,
size_classification_factor=1.3)
#Extract features
training_examples, test_examples = extracor.extract_training_and_test_examples(test_dir.getPath(), 90, 10)
#print("training examples", training_examples)
#print("testing examples", test_examples)
classifier = CharacterClassifier(training_examples,
nr_of_hmms_to_try = 1,
fraction_of_examples_for_test = 0.3,
feature_extractor=extracor,
train_with_examples=False)
before = classifier.test(test_examples)
#Test serialization
classifier_string = classifier.to_string()
reborn_classifier = CharacterClassifier(from_string_string=classifier_string)
reborn_classifier_test_result = reborn_classifier.test(test_examples)
if(reborn_classifier_test_result==before):
pass
else:
raise "Something is wrong with the test result"
classifier.train()
after = classifier.test(test_examples)
print("test_with_two_characters", "before", before, "after", after)
shutil.rmtree(test_dir.getPath())
def test_init_method_different_parameters(self):
test_dir = File("../../character_examples").getPath()
nr_of_training_examples = 90
nr_of_test_examples = 10
for size_classification_factor in drange(0.7, 6.0, 0.3):
print str(size_classification_factor) + ' &',
for nr_of_segs in range(4, 13):
#print(nr_of_segs)
test_scores = []
for test_nr in range(10):
#print(test_nr)
extracor = SimpleImageFeatureExtractor(
nr_of_divisions=nr_of_segs,
size_classification_factor=size_classification_factor)
training_examples, test_examples = extracor.extract_training_and_test_examples(
test_dir, nr_of_training_examples, nr_of_test_examples)
classifier = CharacterClassifier(
training_examples,
nr_of_hmms_to_try=1,
fraction_of_examples_for_test=0,
train_with_examples=False,
initialisation_method=SpecializedHMM.InitMethod.
count_based)
test_scores.append(classifier.test(test_examples))
score = sum(test_scores) / len(test_scores)
print ' $' + str(score) + '$ ',
if nr_of_segs == 12:
print '\\\\'
else:
print '&',
def test_init_method(self,
nr_of_segments=7,
size_classification_factor=1.3,
only_count_based_init=False):
'''Test with different number of training examples and compare
random init with count based init'''
test_dir = File("../../character_examples").getPath()
nr_of_training_examples = 90
nr_of_test_examples = 10
extracor = SimpleImageFeatureExtractor(nr_of_divisions=nr_of_segments,
size_classification_factor=size_classification_factor)
training_examples, test_examples = extracor.extract_training_and_test_examples(test_dir,
nr_of_training_examples,
nr_of_test_examples)
def get_examples(nr_of_examples):
if(nr_of_examples!=90):
raise "Illegal amount of examples"
else:
return training_examples
def get_character_classifier_with_init_method(traing_examples, init_method):
return CharacterClassifier(traing_examples,
nr_of_hmms_to_try=1,
fraction_of_examples_for_test=0,
train_with_examples=False,
initialisation_method=init_method)
self.test_init_method_with_classifier(get_examples,
get_character_classifier_with_init_method,
test_examples,
[90],
only_count_based_init=only_count_based_init)
def test_with_two_characters(self):
#test with just two letters so A and B are copied to a
#special dir that is deleted after the test
base_dir = File("../../character_examples")
test_dir = File(base_dir, "test")
a_dir = File(base_dir, "A")
b_dir = File(base_dir, "B")
shutil.copytree(a_dir.getPath(), File(test_dir, "A").getPath())
shutil.copytree(b_dir.getPath(), File(test_dir, "B").getPath())
extracor = SimpleImageFeatureExtractor(nr_of_divisions=7,
size_classification_factor=1.3)
#Extract features
training_examples, test_examples = extracor.extract_training_and_test_examples(
test_dir.getPath(), 90, 10)
#print("training examples", training_examples)
#print("testing examples", test_examples)
classifier = CharacterClassifier(training_examples,
nr_of_hmms_to_try=1,
fraction_of_examples_for_test=0.3,
feature_extractor=extracor,
train_with_examples=False)
before = classifier.test(test_examples)
#Test serialization
classifier_string = classifier.to_string()
reborn_classifier = CharacterClassifier(
from_string_string=classifier_string)
reborn_classifier_test_result = reborn_classifier.test(test_examples)
if (reborn_classifier_test_result == before):
pass
else:
raise "Something is wrong with the test result"
classifier.train()
after = classifier.test(test_examples)
print("test_with_two_characters", "before", before, "after", after)
shutil.rmtree(test_dir.getPath())
def test_init_method(self,
nr_of_segments=7,
size_classification_factor=1.3,
only_count_based_init=False):
'''Test with different number of training examples and compare
random init with count based init'''
test_dir = File("../../character_examples").getPath()
nr_of_training_examples = 90
nr_of_test_examples = 10
extracor = SimpleImageFeatureExtractor(
nr_of_divisions=nr_of_segments,
size_classification_factor=size_classification_factor)
training_examples, test_examples = extracor.extract_training_and_test_examples(
test_dir, nr_of_training_examples, nr_of_test_examples)
def get_examples(nr_of_examples):
if (nr_of_examples != 90):
raise "Illegal amount of examples"
else:
return training_examples
def get_character_classifier_with_init_method(traing_examples,
init_method):
return CharacterClassifier(traing_examples,
nr_of_hmms_to_try=1,
fraction_of_examples_for_test=0,
train_with_examples=False,
initialisation_method=init_method)
self.test_init_method_with_classifier(
get_examples,
get_character_classifier_with_init_method,
test_examples, [90],
only_count_based_init=only_count_based_init)
def create_character_classification_count_matrix():
'''
This function do the following:
1. Creates a character classifier with 90 training examples
2. Runs a test with 10 test examples for every character. The results are put in a
matrix M so element M[i][j] contains the number of test examples for the character
with index i that are classified to be the character with index j.
'''
extracor = SimpleImageFeatureExtractor(nr_of_divisions=11,
size_classification_factor=3.4)
training_examples, test_examples = extracor.extract_training_and_test_examples(File(File(File(".."),".."),"character_examples").getCanonicalPath(),
nr_of_training_examples=90,
nr_of_test_examples=10)
classifier = CharacterClassifier(training_examples,
nr_of_hmms_to_try=1,
fraction_of_examples_for_test=0,
train_with_examples=False,
initialisation_method=SpecializedHMM.InitMethod.count_based)
alphabet = get_example_alphabet()
classification_count_matrix = zeros(len(alphabet),len(alphabet))
for label, examples in test_examples:
character = label.lower()
character_index = alphabet.index(character)
for example in examples:
classified_character = classifier.classify_character_string(example).lower()
classified_character_index = alphabet.index(classified_character)
count = classification_count_matrix[character_index][classified_character_index]
classification_count_matrix[character_index][classified_character_index] = count + 1
return classification_count_matrix
def test_init_method_different_parameters(self):
test_dir = File("../../character_examples").getPath()
nr_of_training_examples = 90
nr_of_test_examples = 10
for size_classification_factor in drange(0.7, 6.0, 0.3):
print str(size_classification_factor) + ' &',
for nr_of_segs in range(4,13):
#print(nr_of_segs)
test_scores = []
for test_nr in range(10):
#print(test_nr)
extracor = SimpleImageFeatureExtractor(nr_of_divisions=nr_of_segs,
size_classification_factor=size_classification_factor)
training_examples, test_examples = extracor.extract_training_and_test_examples(test_dir,
nr_of_training_examples,
nr_of_test_examples)
classifier = CharacterClassifier(training_examples,
nr_of_hmms_to_try=1,
fraction_of_examples_for_test=0,
train_with_examples=False,
initialisation_method=SpecializedHMM.InitMethod.count_based)
test_scores.append(classifier.test(test_examples))
score = sum(test_scores) / len(test_scores)
print ' $' + str(score) +'$ ',
if nr_of_segs == 12:
print '\\\\'
else:
print '&',
def create_character_classification_count_matrix():
'''
This function do the following:
1. Creates a character classifier with 90 training examples
2. Runs a test with 10 test examples for every character. The results are put in a
matrix M so element M[i][j] contains the number of test examples for the character
with index i that are classified to be the character with index j.
'''
extracor = SimpleImageFeatureExtractor(nr_of_divisions=11,
size_classification_factor=3.4)
training_examples, test_examples = extracor.extract_training_and_test_examples(
File(File(File(".."), ".."), "character_examples").getCanonicalPath(),
nr_of_training_examples=90,
nr_of_test_examples=10)
classifier = CharacterClassifier(
training_examples,
nr_of_hmms_to_try=1,
fraction_of_examples_for_test=0,
train_with_examples=False,
initialisation_method=SpecializedHMM.InitMethod.count_based)
alphabet = get_example_alphabet()
classification_count_matrix = zeros(len(alphabet), len(alphabet))
for label, examples in test_examples:
character = label.lower()
character_index = alphabet.index(character)
for example in examples:
classified_character = classifier.classify_character_string(
example).lower()
classified_character_index = alphabet.index(classified_character)
count = classification_count_matrix[character_index][
classified_character_index]
classification_count_matrix[character_index][
classified_character_index] = count + 1
return classification_count_matrix
def extract_test_examples_to_file(self):
extractor = SimpleImageFeatureExtractor(nr_of_divisions=self.feature_extraction_number_of_segments,
size_classification_factor=self.feature_extraction_classification_factor)
examples_dir = File(File(File(File(str(inspect.getfile( inspect.currentframe() ))).getParent(),".."),".."),"word_examples_for_test").getCanonicalPath()
empty, character_test_examples = extractor.extract_training_and_test_examples(examples_dir, #character_examples word_examples_for_test
nr_of_training_examples=0,
nr_of_test_examples=10)
output = open('datatest_segments_' +
str(self.feature_extraction_number_of_segments) + '_cf_'+
str(self.feature_extraction_classification_factor).replace('.','_')+
'.pkl', 'wb')
pickle.dump(character_test_examples, output)
output.close()
def __init__(self,
characters_with_examples=None,
nr_of_hmms_to_try=3,
fraction_of_examples_for_test=0.1,
train_with_examples=True,
initialisation_method=SpecializedHMM.InitMethod.count_based,
feature_extractor=None,
from_string_string=None):
'''
See WordClassifier
'''
if from_string_string != None:
#init from string
#"\n\n"+ in the next row is for jython bug 1469
feature_extractor_parameters, classifer_string = eval(
"\n\n" + from_string_string)
nr_of_divisions, size_classification_factor = feature_extractor_parameters
self.feature_extractor = SimpleImageFeatureExtractor(
nr_of_divisions, size_classification_factor)
self.nr_of_segments = nr_of_divisions
super(CharacterClassifier,
self).__init__(from_string_string=classifer_string)
return
#Feature extractor is voluntary but is necessary if the classify_image
#method shall be used
self.feature_extractor = feature_extractor
#Get the number of segments created by the feature extractor
#by looking at the length of a training example
label, examples = characters_with_examples[0]
self.nr_of_segments = len(examples[0])
new_characters_with_examples = []
for label, examples in characters_with_examples:
new_characters_with_examples.append(
(label * self.nr_of_segments, examples))
super(CharacterClassifier,
self).__init__(new_characters_with_examples,
nr_of_hmms_to_try,
fraction_of_examples_for_test,
train_with_examples,
initialisation_method,
alphabet=SimpleImageFeatureExtractor.feature_ids)
def create_character_classifier(save_to_file_path):
example_dir = File("../../character_examples").getPath()
nr_of_training_examples = 100
nr_of_test_examples = 0
extractor = SimpleImageFeatureExtractor(nr_of_divisions=7,
size_classification_factor=1.3)
training_examples, test_examples = extractor.extract_training_and_test_examples(example_dir,
nr_of_training_examples,
nr_of_test_examples)
classifier = CharacterClassifier(training_examples,
nr_of_hmms_to_try=1,
fraction_of_examples_for_test=0,
train_with_examples=False,
initialisation_method=SpecializedHMM.InitMethod.count_based,
feature_extractor=extractor)
classifier_string = classifier.to_string()
file = open(save_to_file_path,'w')
file.write(classifier_string)
file.close()
def create_character_classifier(save_to_file_path):
example_dir = File("../../character_examples").getPath()
nr_of_training_examples = 90
nr_of_test_examples = 10
extractor = SimpleImageFeatureExtractor(nr_of_divisions=11,
size_classification_factor=4.6)
training_examples, test_examples = extractor.extract_training_and_test_examples(example_dir,
nr_of_training_examples,
nr_of_test_examples)
classifier = CharacterClassifier(training_examples,
nr_of_hmms_to_try=1,
fraction_of_examples_for_test=0,
train_with_examples=True,
initialisation_method=SpecializedHMM.InitMethod.count_based,
feature_extractor=extractor)
test_result = str(classifier.test(test_examples))
print('Prediction ratio:', test_result)
def extract_test_examples_to_file(self):
extractor = SimpleImageFeatureExtractor(
nr_of_divisions=self.feature_extraction_number_of_segments,
size_classification_factor=self.
feature_extraction_classification_factor)
examples_dir = File(
File(
File(
File(str(inspect.getfile(
inspect.currentframe()))).getParent(), ".."), ".."),
"word_examples_for_test").getCanonicalPath()
empty, character_test_examples = extractor.extract_training_and_test_examples(
examples_dir, #character_examples word_examples_for_test
nr_of_training_examples=0,
nr_of_test_examples=10)
output = open(
'datatest_segments_' +
str(self.feature_extraction_number_of_segments) + '_cf_' +
str(self.feature_extraction_classification_factor).replace(
'.', '_') + '.pkl', 'wb')
pickle.dump(character_test_examples, output)
output.close()
def create_character_classifier(save_to_file_path):
example_dir = File("../../character_examples").getPath()
nr_of_training_examples = 100
nr_of_test_examples = 0
extractor = SimpleImageFeatureExtractor(nr_of_divisions=11,
size_classification_factor=4.6)
training_examples, test_examples = extractor.extract_training_and_test_examples(
example_dir, nr_of_training_examples, nr_of_test_examples)
classifier = CharacterClassifier(
training_examples,
nr_of_hmms_to_try=1,
fraction_of_examples_for_test=0,
train_with_examples=False,
initialisation_method=SpecializedHMM.InitMethod.count_based,
feature_extractor=extractor)
#test_result = str(classifier.test(test_examples))
#print(test_result)
classifier_string = classifier.to_string()
file = open(save_to_file_path + ".dat", 'w')
file.write(classifier_string)
file.close()
def __init__(self,
characters_with_examples=None,
nr_of_hmms_to_try=3,
fraction_of_examples_for_test=0.1,
train_with_examples=True,
initialisation_method=SpecializedHMM.InitMethod.count_based,
feature_extractor=None,
from_string_string=None):
'''
See WordClassifier
'''
if from_string_string != None:
#init from string
#"\n\n"+ in the next row is for jython bug 1469
feature_extractor_parameters,classifer_string = eval("\n\n"+from_string_string)
nr_of_divisions,size_classification_factor = feature_extractor_parameters
self.feature_extractor = SimpleImageFeatureExtractor(nr_of_divisions,
size_classification_factor)
self.nr_of_segments = nr_of_divisions
super(CharacterClassifier,self).__init__(from_string_string=classifer_string)
return
#Feature extractor is voluntary but is necessary if the classify_image
#method shall be used
self.feature_extractor = feature_extractor
#Get the number of segments created by the feature extractor
#by looking at the length of a training example
label,examples = characters_with_examples[0]
self.nr_of_segments = len(examples[0])
new_characters_with_examples = []
for label,examples in characters_with_examples:
new_characters_with_examples.append((label*self.nr_of_segments,examples))
super(CharacterClassifier,self).__init__(new_characters_with_examples,
nr_of_hmms_to_try,
fraction_of_examples_for_test,
train_with_examples,
initialisation_method,
alphabet=SimpleImageFeatureExtractor.feature_ids)
class CharacterClassifier(WordClassifier):
'''
Works as WordClassifier with some extra features for character classification
'''
def __init__(self,
characters_with_examples=None,
nr_of_hmms_to_try=3,
fraction_of_examples_for_test=0.1,
train_with_examples=True,
initialisation_method=SpecializedHMM.InitMethod.count_based,
feature_extractor=None,
from_string_string=None):
'''
See WordClassifier
'''
if from_string_string != None:
#init from string
#"\n\n"+ in the next row is for jython bug 1469
feature_extractor_parameters, classifer_string = eval(
"\n\n" + from_string_string)
nr_of_divisions, size_classification_factor = feature_extractor_parameters
self.feature_extractor = SimpleImageFeatureExtractor(
nr_of_divisions, size_classification_factor)
self.nr_of_segments = nr_of_divisions
super(CharacterClassifier,
self).__init__(from_string_string=classifer_string)
return
#Feature extractor is voluntary but is necessary if the classify_image
#method shall be used
self.feature_extractor = feature_extractor
#Get the number of segments created by the feature extractor
#by looking at the length of a training example
label, examples = characters_with_examples[0]
self.nr_of_segments = len(examples[0])
new_characters_with_examples = []
for label, examples in characters_with_examples:
new_characters_with_examples.append(
(label * self.nr_of_segments, examples))
super(CharacterClassifier,
self).__init__(new_characters_with_examples,
nr_of_hmms_to_try,
fraction_of_examples_for_test,
train_with_examples,
initialisation_method,
alphabet=SimpleImageFeatureExtractor.feature_ids)
def classify_character_string(self, string):
classification = super(CharacterClassifier, self).classify(string)
return classification[0]
def classify_image(self, buffered_image):
string = self.feature_extractor.extract_feature_string(buffered_image)
return self.classify_character_string(string)
def test(self, test_examples):
'''
See WordClassifier.test()
'''
new_test_examples = []
for label, examples in test_examples:
new_test_examples.append((label * self.nr_of_segments, examples))
return super(CharacterClassifier, self).test(new_test_examples)
def to_string(self):
if self.feature_extractor == None:
raise "feature_extractor must be given if the character classifier shall be stringified"
else:
feature_extractor_parameters = (
self.feature_extractor.nr_of_divisions,
self.feature_extractor.size_classification_factor)
word_classifier_string = super(CharacterClassifier, self).to_string()
return str((feature_extractor_parameters, word_classifier_string))
class CharacterClassifier(WordClassifier):
'''
Works as WordClassifier with some extra features for character classification
'''
def __init__(self,
characters_with_examples=None,
nr_of_hmms_to_try=3,
fraction_of_examples_for_test=0.1,
train_with_examples=True,
initialisation_method=SpecializedHMM.InitMethod.count_based,
feature_extractor=None,
from_string_string=None):
'''
See WordClassifier
'''
if from_string_string != None:
#init from string
#"\n\n"+ in the next row is for jython bug 1469
feature_extractor_parameters,classifer_string = eval("\n\n"+from_string_string)
nr_of_divisions,size_classification_factor = feature_extractor_parameters
self.feature_extractor = SimpleImageFeatureExtractor(nr_of_divisions,
size_classification_factor)
self.nr_of_segments = nr_of_divisions
super(CharacterClassifier,self).__init__(from_string_string=classifer_string)
return
#Feature extractor is voluntary but is necessary if the classify_image
#method shall be used
self.feature_extractor = feature_extractor
#Get the number of segments created by the feature extractor
#by looking at the length of a training example
label,examples = characters_with_examples[0]
self.nr_of_segments = len(examples[0])
new_characters_with_examples = []
for label,examples in characters_with_examples:
new_characters_with_examples.append((label*self.nr_of_segments,examples))
super(CharacterClassifier,self).__init__(new_characters_with_examples,
nr_of_hmms_to_try,
fraction_of_examples_for_test,
train_with_examples,
initialisation_method,
alphabet=SimpleImageFeatureExtractor.feature_ids)
def classify_character_string(self,string):
classification = super(CharacterClassifier, self).classify(string)
return classification[0]
def classify_image(self,buffered_image):
string = self.feature_extractor.extract_feature_string(buffered_image)
return self.classify_character_string(string)
def test(self,test_examples):
'''
See WordClassifier.test()
'''
new_test_examples = []
for label, examples in test_examples:
new_test_examples.append((label * self.nr_of_segments, examples))
return super(CharacterClassifier, self).test(new_test_examples)
def to_string(self):
if self.feature_extractor == None:
raise "feature_extractor must be given if the character classifier shall be stringified"
else:
feature_extractor_parameters = (self.feature_extractor.nr_of_divisions,
self.feature_extractor.size_classification_factor)
word_classifier_string = super(CharacterClassifier,self).to_string()
return str((feature_extractor_parameters,
word_classifier_string))