def given_real_data_test():
patients = pandas.DataFrame.from_csv(
'./data/training_SyncPatient.csv').reset_index()
transcripts = pandas.DataFrame.from_csv(
'./data/training_SyncTranscript.csv').reset_index()
transcripts = transcripts[transcripts['Height'] > 0]
transcripts = transcripts[transcripts['Weight'] > 0]
transcripts = transcripts[transcripts['BMI'] > 0]
joined_df = patients.merge(transcripts, on='PatientGuid', how='inner')
final_df = joined_df.groupby('PatientGuid').first().reset_index()
female_set = final_df.ix[np.random.choice(
final_df[final_df['Gender'] == 'F'].index, 500)]
male_set = final_df.ix[np.random.choice(
final_df[final_df['Gender'] == 'M'].index, 500)]
training_data = [(x[2], (x[8], x[9], x[10])) for x in female_set.values]
training_data += [(x[2], (x[8], x[9], x[10])) for x in male_set.values]
classifier = NaiveBayes.Classifier()
for class_label, input_data in training_data:
classifier.train(classification=class_label, observation=input_data)
# Manual verification
pprint.pprint(classifier._calculate_model_parameters())
# Men
print("Men")
print(classifier.classify(observation=(71.3, 210.0, 23.509)))
print(classifier.classify(observation=(66.0, 268.8, 27.241999999999997)))
print(classifier.classify(observation=(65.0, 284.0, 30.616)))
print("Women")
print(classifier.classify(observation=(60.5, 151.0, 29.002)))
print(classifier.classify(observation=(60.0, 148.0, 28.901)))
print(classifier.classify(observation=(60.0, 134.923, 26.346999999999998)))
assert True, "Always pass until we want to manually evaluate."
def given_one_observation_for_two_classes_test():
classifier = NaiveBayes.Classifier()
classifier.train(classification='a class', observation=0)
classifier.train(classification='b class', observation=100)
classification = classifier.classify(observation=23.2)
assert classification is None, "Should classify as the nearest class."
classification = classifier.classify(observation=73.2)
assert classification is None, "Should classify as the nearest class."
def given_multiple_observations_for_two_classes_with_roughly_same_variance_test(
):
classifier = NaiveBayes.Classifier()
classifier.batch_train([('a class', (0.0, )), ('a class', (1.0, )),
('a class', (75.0, )), ('b class', (25, )),
('b class', (99, )), ('b class', (100, ))])
classification = classifier.classify(observation=(25, ))
assert classification == 'a class', "Should classify as the best fit class."
classification = classifier.classify(observation=(75.0, ))
assert classification == 'b class', "Should classify as the best fit class."
def given_classes_of_different_likelihood_test():
classifier = NaiveBayes.Classifier()
observation = 3
observations = {
'class a': [1, 2, 3, 4, 5],
'class b': [1, 1, 2, 2, 3, 3, 4, 4, 5, 5]
}
results = classifier._probability_of_each_class_given_data(
observation, observations)
print results
assert results['class b'] > results[
'class a'], "Should classify as class b when class probability is taken into account."
def given_two_classes_with_identical_two_dimension_inputs_test():
classifier = NaiveBayes.Classifier()
observation = (3, 10)
observations = {
'class a': [(1, 10), (2, 5), (3, 12), (4, 10), (5, 5)],
'class b': [(1, 10), (2, 5), (3, 12), (4, 10), (5, 5)]
}
results = classifier._probability_of_each_class_given_data(
observation, observations)
print(results)
assert results['class a'] == 0.5, "There should be 50/50 chance of class a"
assert results['class b'] == 0.5, "There should be 50/50 chance of class b"
def given_two_classes_with_two_dimension_inputs_test():
classifier = NaiveBayes.Classifier()
observation = (3, 10)
observations = {
'class a': [(1, -1), (2, 0), (3, -1), (4, 1), (5, -1)],
'class b': [(1, 10), (2, 5), (3, 12), (4, 10), (5, 5)]
}
results = classifier._probability_of_each_class_given_data(
observation, observations)
print(results)
assert results['class b'] > results[
'class a'], "Should classify as class b because of dimension 2."
def given_multiple_observations_for_two_classes_with_roughly_same_variance_test(
):
classifier = NaiveBayes.Classifier()
classifier.train(classification='a class', observation=0.0)
classifier.train(classification='a class', observation=1.0)
classifier.train(classification='a class', observation=75.0)
classifier.train(classification='b class', observation=25)
classifier.train(classification='b class', observation=99)
classifier.train(classification='b class', observation=100)
classification = classifier.classify(observation=25)
assert classification == 'a class', "Should classify as the best fit class."
classification = classifier.classify(observation=75.0)
assert classification == 'b class', "Should classify as the best fit class."
def given_classes_of_different_likelihood_test():
classifier = NaiveBayes.Classifier()
observation = (3, )
observations = {
'class a': [(1, ), (2, ), (3, ), (4, ), (5, )],
'class b': [(1, ), (1, ), (2, ), (2, ), (3, ), (3, ), (4, ), (4, ),
(5, ), (5, )]
}
results = classifier._probability_of_each_class_given_data(
observation, observations)
print(results)
assert results['class b'] > results[
'class a'], "Should classify as class b when class probability is taken into account."
def given_data_test():
classifier = NaiveBayes.Classifier()
classifier.train(classification='class a', observation=(0.0, ))
classifier.train(classification='class a', observation=(1.0, ))
classifier.train(classification='class a', observation=(2.0, ))
classifier.train(classification='class b', observation=(50, ))
classifier.train(classification='class b', observation=(75, ))
classifier.train(classification='class b', observation=(100, ))
results = classifier._calculate_model_parameters()
assert results['class a'][0]['mean'] == 1.0
assert results['class a'][0]['variance'] == 2 / 3.
assert results['class b'][0]['mean'] == 75.
assert results['class b'][0]['variance'] == 416.66666666666669
def given_multiple_observations_for_two_classes_with_different_variance_test():
classifier = NaiveBayes.Classifier()
classifier.train(classification='a class', observation=0.0)
classifier.train(classification='a class', observation=1.0)
classifier.train(classification='a class', observation=2.0)
classifier.train(classification='b class', observation=50)
classifier.train(classification='b class', observation=75)
classifier.train(classification='b class', observation=100)
classifier.train(classification='c class', observation=0.0)
classifier.train(classification='c class', observation=-1.0)
classifier.train(classification='c class', observation=-2.0)
classification = classifier.classify(observation=15)
assert classification == 'b class', "Because of class b's variance this should be class b."
classification = classifier.classify(observation=2.5)
assert classification == 'a class', "Should classify as class a because of tight variance."
classification = classifier.classify(observation=-2.5)
assert classification == 'c class', "Should classify as class c because it's the only negative one."
def quantify_classifier_accuracy_test():
# Load and clean up the data
patients = pandas.DataFrame.from_csv(
'./data/training_SyncPatient.csv').reset_index()
transcripts = pandas.DataFrame.from_csv(
'./data/training_SyncTranscript.csv').reset_index()
transcripts = transcripts[transcripts['Height'] > 0]
transcripts = transcripts[transcripts['Weight'] > 0]
transcripts = transcripts[transcripts['BMI'] > 0]
joined_df = patients.merge(transcripts, on='PatientGuid', how='inner')
final_df = joined_df.groupby('PatientGuid').first().reset_index()
total_set = final_df.ix[np.random.choice(final_df.index,
7000,
replace=False)]
# Partition development and cross-validation datasets
training_count = 2000
training_data = map(lambda x: (x[2], (x[8], x[9], x[10])),
total_set.values[:training_count])
cross_validate_data = map(lambda x: (x[2], (x[8], x[9], x[10])),
total_set.values[training_count:])
# Train the classifier on the training data.
classifier = NaiveBayes.Classifier()
for class_label, input_data in training_data:
classifier.train(classification=class_label, observation=input_data)
# Test how well the classifier generalizes.
number_correct = 0
number_tested = 0
for class_label, input_data in cross_validate_data:
number_tested += 1
assigned_class = classifier.classify(observation=input_data)
if class_label == assigned_class:
number_correct += 1
correct_rate = number_correct / (1. * number_tested)
print("Correct rate: {0}, Total: {1}".format(correct_rate, number_tested))
pprint.pprint(classifier._calculate_model_parameters())
assert correct_rate > 0.6, "Should be significantly better than random."
def given_an_observation_for_a_single_class_test():
classifier = NaiveBayes.Classifier()
classifier.train(classification='a class', observation=0)
classification = classifier.classify(observation=23.2)
assert classification == 'a class', "Should always classify as given class if there is only one."
def no_observations_test():
classifier = NaiveBayes.Classifier()
classification = classifier.classify(observation=23.2)
assert classification is None, "Should not classify observations without training examples."
