def algorithm_analysis():
"""
Prepares a classifier based on non-validated data, and evaluates its performance in the valdated
portion of the dataset.
:return: Dictionary of classifiers
"""
raw_dataframe = preprocessing.load_original_dataframe()
issues_train, issues_train_std, priorities_train, issues_test, issues_test_std, priorities_test = get_training_datasets(
raw_dataframe)
unfiltered_dataframe = preprocessing.filter_issues_dataframe(raw_dataframe, priority_changer=False)
filtered_dataframe = preprocessing.filter_issues_dataframe(raw_dataframe, priority_changer=True)
train, test = train_test_split(unfiltered_dataframe, test_size=0.2, random_state=0)
test_valid = preprocessing.filter_issues_dataframe(test, priority_changer=True)
issues_test_valid, issues_test_valid_std, priorities_test_valid = prepare_for_classification(test_valid,
issues_train)
issues_valid, issues_valid_std, priorities_valid = prepare_for_classification(filtered_dataframe, issues_train)
results = []
estimators = {}
for algorithm, grid_search in selector.get_algorithms():
training_set = issues_train_std
test_set = issues_test_std
test_valid_set = issues_test_valid_std
valid_set = issues_valid_std
print "Current algorithm: ", algorithm
if algorithm == "RandomForest":
training_set = issues_train
test_set = issues_test
test_valid_set = issues_test_valid
valid_set = issues_valid
optimal_estimator, best_params = tuning.parameter_tuning(grid_search, training_set,
priorities_train)
estimators[algorithm] = optimal_estimator
result = selector.analyse_performance(optimal_estimator, best_params, grid_search, algorithm, training_set,
priorities_train,
test_set,
priorities_test)
print "Evaluating on the valid portion of the test dataset ..."
assigner.evaluate_performance(prefix=algorithm, classifier=optimal_estimator,
issues_test_std=test_valid_set, priority_test=priorities_test_valid)
print "Evaluating on the complete valid dataset ..."
assigner.evaluate_performance(prefix=algorithm, classifier=optimal_estimator,
issues_test_std=valid_set, priority_test=priorities_valid)
if result:
results.append(result + ("ALL", len(unfiltered_dataframe.index)))
selector.write_results("all_experiment_results.csv", results)
return estimators
def predict_priority():
"""
Trains a classifier and runs it on a dataset.
:return:
"""
# This values are product of the experiments
best_class_label = "Severe"
best_estimators = 51
best_depth = 21
classifier = RandomForestClassifier(n_estimators=best_estimators, max_depth=best_depth, random_state=0,
n_jobs=-1)
# classifier = SVC(kernel='linear', C=1.0, class_weight='balanced')
target_dataframe = preprocessing.load_original_dataframe()
training_dataframe = preprocessing.filter_issues_dataframe(target_dataframe)
issues_training, labels_traning = preprocessing.encode_and_split(issues_dataframe=training_dataframe,
class_label=best_class_label,
numerical_features=preprocessing.NUMERICAL_FEATURES,
nominal_features=[],
text_feature=None)
assigner.train_and_predict(classifier, target_dataframe, issues_training, labels_traning, best_class_label,
preprocessing.NUMERICAL_FEATURES, [])
def predict_priority():
"""
Trains a classifier and runs it on a dataset.
:return:
"""
# This values are product of the experiments
best_class_label = "Severe"
best_estimators = 51
best_depth = 21
classifier = RandomForestClassifier(n_estimators=best_estimators,
max_depth=best_depth,
random_state=0,
n_jobs=-1)
# classifier = SVC(kernel='linear', C=1.0, class_weight='balanced')
target_dataframe = preprocessing.load_original_dataframe()
training_dataframe = preprocessing.filter_issues_dataframe(
target_dataframe)
issues_training, labels_traning = preprocessing.encode_and_split(
issues_dataframe=training_dataframe,
class_label=best_class_label,
numerical_features=preprocessing.NUMERICAL_FEATURES,
nominal_features=[],
text_feature=None)
assigner.train_and_predict(classifier, target_dataframe, issues_training,
labels_traning, best_class_label,
preprocessing.NUMERICAL_FEATURES, [])
def main():
original_dataframe = preprocessing.load_original_dataframe()
issues_dataframe = preprocessing.filter_issues_dataframe(
original_dataframe)
# Plotting projects
figure, axes = plt.subplots(1, 1)
issues_dataframe['Git Repository'].value_counts(normalize=True).plot(
kind='bar', ax=axes)
plt.show()
issues_dataframe, encoded_priorities = preprocessing.encode_and_split(
issues_dataframe, preprocessing.CLASS_LABEL,
preprocessing.NUMERICAL_FEATURES, preprocessing.NOMINAL_FEATURES)
# Plotting priorities
figure, axes = plt.subplots(1, 1)
encoded_priorities.value_counts(normalize=True, sort=True).plot(kind='bar',
ax=axes)
plt.show()
issues_train, issues_test, priority_train, priority_test = train_test_split(
issues_dataframe, encoded_priorities, test_size=0.2, random_state=0)
print len(issues_train.index), " issues on the train set."
issues_train_std, issues_test_std = preprocessing.escale_numerical_features(
preprocessing.NUMERICAL_FEATURES, issues_train, issues_test)
logit_classifier = select_features_l1(issues_train_std, priority_train,
issues_test_std, priority_test)
knn_classifier = sequential_feature_selection(issues_train_std,
priority_train,
issues_test_std,
priority_test)
print "Building Random Forest Classifier ..."
rforest_classifier = RandomForestClassifier(n_estimators=10000,
random_state=0,
n_jobs=-1)
rforest_classifier.fit(issues_train, priority_train)
forest_classifier = feature_importance_with_forest(issues_train,
priority_train,
issues_test,
priority_test)
rforest_classifier = RandomForestClassifier(n_estimators=10000,
random_state=0,
n_jobs=-1)
train_and_predict(rforest_classifier, original_dataframe, issues_dataframe,
encoded_priorities, preprocessing.CLASS_LABEL,
preprocessing.NUMERICAL_FEATURES,
preprocessing.NOMINAL_FEATURES)
def main():
"""
Initial execution point.
:return: None.
"""
best_forest = algorithm_analysis()["RandomForest"]
raw_dataframe = preprocessing.load_original_dataframe()
issues_train, issues_train_std, priorities_train, issues_test, issues_test_std, priorities_test = get_training_datasets(
raw_dataframe)
assigner.feature_importance_with_forest(best_forest, issues_train, priorities_train, issues_test, priorities_test)
def main():
"""
Initial execution point.
:return: None.
"""
best_forest = algorithm_analysis()["RandomForest"]
raw_dataframe = preprocessing.load_original_dataframe()
issues_train, issues_train_std, priorities_train, issues_test, issues_test_std, priorities_test = get_training_datasets(
raw_dataframe)
assigner.feature_importance_with_forest(best_forest, issues_train,
priorities_train, issues_test,
priorities_test)
def main():
"""
Initial execution point
:return: None
"""
original_dataframe = preprocessing.load_original_dataframe()
repositories = get_all_repositories(original_dataframe)
results = []
for repository in repositories:
print "Working on repository ", repository, " ..."
project_dataframe = preprocessing.filter_issues_dataframe(
original_dataframe, repository=repository)
# Threslhold taking into account considering the scikit-learn cheat sheet
# http://scikit-learn.org/stable/tutorial/machine_learning_map/
minimum_threshold = 50
issues_found = len(project_dataframe.index)
print issues_found, " issues found on repository ", repository
if issues_found > minimum_threshold:
# The Git Repository feauture is not needed since it is filtered.
nominal_features = []
issues, priorities = preprocessing.encode_and_split(
project_dataframe, assigner.CLASS_LABEL,
assigner.NUMERICAL_FEATURES, nominal_features)
train_test = preprocessing.train_test_encode(
repository, issues, priorities)
if train_test:
issues_train_std, priority_train, issues_test_std, priority_test = train_test
result = run_algorithm_analysis(issues_train_std,
priority_train,
issues_test_std, priority_test,
repository, issues_found)
if result:
results.extend(result)
else:
print "Issues corresponding to repository ", repository, " are not enough for analysis."
write_results("project_experiment_results.csv", results)
def algorithm_analysis():
"""
Executes the analysis for finding the optimal class label and algorithm configuration.
:return: None.
"""
consolidated_results = []
try:
minimum_records = 50
class_labels = [
'Severe'
# 'Blocker'
# , 'Non-Severe', 'Trivial', 'Critical'
]
original_dataframe = preprocessing.load_original_dataframe()
repositories = []
# valid_dataframe = preprocessing.filter_issues_dataframe(original_dataframe)
# repositories.append(("VALID", valid_dataframe))
for repo_name in selector.get_all_repositories(original_dataframe):
project_dataframe = preprocessing.filter_issues_dataframe(
original_dataframe, repository=repo_name)
repositories.append((repo_name, project_dataframe))
for class_label in class_labels:
for repository_name, dataframe in repositories:
print "Using ", class_label, " as the class feature."
print "Working on repository ", repository_name, " with ", len(
dataframe.index), "Issues"
if len(dataframe.index) >= minimum_records:
results = execute_analysis(
dataframe, class_label + "-" + repository_name,
class_label)
consolidated_results.extend(results)
else:
print "Not enough issues for analysis: ", len(
dataframe.index)
finally:
selector.write_results("Binary_Classification.csv",
consolidated_results)
winsound.Beep(2500, 1000)
def main():
"""
Initial execution point
:return: None
"""
original_dataframe = preprocessing.load_original_dataframe()
repositories = get_all_repositories(original_dataframe)
results = []
for repository in repositories:
print "Working on repository ", repository, " ..."
project_dataframe = preprocessing.filter_issues_dataframe(original_dataframe, repository=repository)
# Threslhold taking into account considering the scikit-learn cheat sheet
# http://scikit-learn.org/stable/tutorial/machine_learning_map/
minimum_threshold = 50
issues_found = len(project_dataframe.index)
print issues_found, " issues found on repository ", repository
if issues_found > minimum_threshold:
# The Git Repository feauture is not needed since it is filtered.
nominal_features = []
issues, priorities = preprocessing.encode_and_split(project_dataframe, assigner.CLASS_LABEL,
assigner.NUMERICAL_FEATURES, nominal_features)
train_test = preprocessing.train_test_encode(repository, issues, priorities)
if train_test:
issues_train_std, priority_train, issues_test_std, priority_test = train_test
result = run_algorithm_analysis(issues_train_std, priority_train, issues_test_std, priority_test,
repository, issues_found)
if result:
results.extend(result)
else:
print "Issues corresponding to repository ", repository, " are not enough for analysis."
write_results("project_experiment_results.csv", results)
def main():
original_dataframe = preprocessing.load_original_dataframe()
issues_dataframe = preprocessing.filter_issues_dataframe(original_dataframe)
# Plotting projects
figure, axes = plt.subplots(1, 1)
issues_dataframe['Git Repository'].value_counts(normalize=True).plot(kind='bar', ax=axes)
plt.show()
issues_dataframe, encoded_priorities = preprocessing.encode_and_split(issues_dataframe, preprocessing.CLASS_LABEL,
preprocessing.NUMERICAL_FEATURES,
preprocessing.NOMINAL_FEATURES)
# Plotting priorities
figure, axes = plt.subplots(1, 1)
encoded_priorities.value_counts(normalize=True, sort=True).plot(kind='bar', ax=axes)
plt.show()
issues_train, issues_test, priority_train, priority_test = train_test_split(issues_dataframe,
encoded_priorities,
test_size=0.2, random_state=0)
print len(issues_train.index), " issues on the train set."
issues_train_std, issues_test_std = preprocessing.escale_numerical_features(preprocessing.NUMERICAL_FEATURES,
issues_train,
issues_test)
logit_classifier = select_features_l1(issues_train_std, priority_train, issues_test_std, priority_test)
knn_classifier = sequential_feature_selection(issues_train_std, priority_train, issues_test_std, priority_test)
print "Building Random Forest Classifier ..."
rforest_classifier = RandomForestClassifier(n_estimators=10000, random_state=0, n_jobs=-1)
rforest_classifier.fit(issues_train, priority_train)
forest_classifier = feature_importance_with_forest(issues_train, priority_train, issues_test, priority_test)
rforest_classifier = RandomForestClassifier(n_estimators=10000, random_state=0, n_jobs=-1)
train_and_predict(rforest_classifier, original_dataframe, issues_dataframe, encoded_priorities,
preprocessing.CLASS_LABEL,
preprocessing.NUMERICAL_FEATURES,
preprocessing.NOMINAL_FEATURES)
def algorithm_analysis():
"""
Executes the analysis for finding the optimal class label and algorithm configuration.
:return: None.
"""
consolidated_results = []
try:
minimum_records = 50
class_labels = [
'Severe'
# 'Blocker'
# , 'Non-Severe', 'Trivial', 'Critical'
]
original_dataframe = preprocessing.load_original_dataframe()
repositories = []
# valid_dataframe = preprocessing.filter_issues_dataframe(original_dataframe)
# repositories.append(("VALID", valid_dataframe))
for repo_name in selector.get_all_repositories(original_dataframe):
project_dataframe = preprocessing.filter_issues_dataframe(original_dataframe,
repository=repo_name)
repositories.append((repo_name, project_dataframe))
for class_label in class_labels:
for repository_name, dataframe in repositories:
print "Using ", class_label, " as the class feature."
print "Working on repository ", repository_name, " with ", len(dataframe.index), "Issues"
if len(dataframe.index) >= minimum_records:
results = execute_analysis(dataframe, class_label + "-" + repository_name, class_label)
consolidated_results.extend(results)
else:
print "Not enough issues for analysis: ", len(dataframe.index)
finally:
selector.write_results("Binary_Classification.csv", consolidated_results)
winsound.Beep(2500, 1000)
def algorithm_analysis():
"""
Prepares a classifier based on non-validated data, and evaluates its performance in the valdated
portion of the dataset.
:return: Dictionary of classifiers
"""
raw_dataframe = preprocessing.load_original_dataframe()
issues_train, issues_train_std, priorities_train, issues_test, issues_test_std, priorities_test = get_training_datasets(
raw_dataframe)
unfiltered_dataframe = preprocessing.filter_issues_dataframe(
raw_dataframe, priority_changer=False)
filtered_dataframe = preprocessing.filter_issues_dataframe(
raw_dataframe, priority_changer=True)
train, test = train_test_split(unfiltered_dataframe,
test_size=0.2,
random_state=0)
test_valid = preprocessing.filter_issues_dataframe(test,
priority_changer=True)
issues_test_valid, issues_test_valid_std, priorities_test_valid = prepare_for_classification(
test_valid, issues_train)
issues_valid, issues_valid_std, priorities_valid = prepare_for_classification(
filtered_dataframe, issues_train)
results = []
estimators = {}
for algorithm, grid_search in selector.get_algorithms():
training_set = issues_train_std
test_set = issues_test_std
test_valid_set = issues_test_valid_std
valid_set = issues_valid_std
print "Current algorithm: ", algorithm
if algorithm == "RandomForest":
training_set = issues_train
test_set = issues_test
test_valid_set = issues_test_valid
valid_set = issues_valid
optimal_estimator, best_params = tuning.parameter_tuning(
grid_search, training_set, priorities_train)
estimators[algorithm] = optimal_estimator
result = selector.analyse_performance(optimal_estimator, best_params,
grid_search, algorithm,
training_set, priorities_train,
test_set, priorities_test)
print "Evaluating on the valid portion of the test dataset ..."
assigner.evaluate_performance(prefix=algorithm,
classifier=optimal_estimator,
issues_test_std=test_valid_set,
priority_test=priorities_test_valid)
print "Evaluating on the complete valid dataset ..."
assigner.evaluate_performance(prefix=algorithm,
classifier=optimal_estimator,
issues_test_std=valid_set,
priority_test=priorities_valid)
if result:
results.append(result + ("ALL", len(unfiltered_dataframe.index)))
selector.write_results("all_experiment_results.csv", results)
return estimators