def dt_classifier():
dt_clf = DecisionTreeClassifier(max_depth=25)
dt_clf.fit(data_train, target_train)
missing_data_rows, op = common.load_train_data_and_split(targetcol=6, file='data/processed_only_missing.csv', split=False)
preds = list(dt_clf.predict(missing_data_rows))
# print [[x,preds.count(x)] for x in set(preds)]
return preds
def dt_classifier():
dt_clf = DecisionTreeClassifier(max_depth=25)
dt_clf.fit(data_train, target_train)
missing_data_rows, op = common.load_train_data_and_split(
targetcol=6, file='data/processed_only_missing.csv', split=False)
preds = list(dt_clf.predict(missing_data_rows))
# print [[x,preds.count(x)] for x in set(preds)]
return preds
def main():
data_train, data_test, target_train, target_test = common.load_train_data_and_split(file='data/processed_missing_filled_in.csv')
data_train = np.asarray(data_train)
target_train = np.array(target_train)
target_train = target_train.astype(np.int32)
print(target_train)
data_train, target_train = smote.smote_data(data_train, target_train)
classify(data_train, target_train, data_test, target_test)
def main():
datafiles = ['data/processed_missing_filled_in.csv', 'data/processed_without_missing.csv', 'data/processed.csv']
datanames = ['md=imputed', 'md=deleted', 'md=0s']
num_samples_per_class = [-1] #, 6000]
nsnames = ['ns=all'] #, 'ns=6000']
num_classes = [2, 3]
cnames = ['nc=2', 'nc=3']
oversample = [True] #, False]
osnames = ["os=t"] #, "os=f"]
algnames = ["NN", "DT", "RandomForest", "AdaBoost", "GaussianNB", "LDA", "QDA", "SGD", "NNet"]
algs = [
KNeighborsClassifier(5),
DecisionTreeClassifier(max_depth=25),
RandomForestClassifier(max_depth=25, n_estimators=10, max_features=1),
AdaBoostClassifier(),
GaussianNB(),
LDA(),
QDA(),
SGDClassifier(penalty='elasticnet', alpha=0.1, loss='modified_huber'),
0
]
for alg, algname in zip(algs, algnames):
for dat, datname in zip(datafiles, datanames):
for numspl, sname in zip(num_samples_per_class, nsnames):
for numcls, cname in zip(num_classes, cnames):
for os, osname in zip(oversample, osnames):
algdesc = algname + "_" + datname + "_" + sname + "_" + cname + "_" + osname
print(algdesc)
input_train, input_test, output_train, output_test = common.load_train_data_and_split(file=dat, num_samples_per_class=numspl, num_classes=numcls, smote=os)
if algname is "NNet":
alg = NeuralNet(layers=[('input', InputLayer), ('dense0', DenseLayer), ('dropout0', DropoutLayer), ('dense1', DenseLayer), ('dropout1', DropoutLayer), ('output', DenseLayer)], input_shape=(None, input_train.shape[1]), dense0_num_units=300, dropout0_p=0.075, dropout1_p=0.1, dense1_num_units=750, output_num_units=numcls, output_nonlinearity=softmax, update=nesterov_momentum, update_learning_rate=0.001, update_momentum=0.99, eval_size=0.33, verbose=1, max_epochs=15)
model = alg.fit(input_train, output_train)
print("TRAIN ", algdesc)
predictions_train = model.predict(input_train)
save_results(output_train, predictions_train, algdesc+"_train", algname)
print("TEST ", algdesc)
predictions_test = model.predict(input_test)
save_results(output_test, predictions_test, algdesc+"_test", algname)
n_iter=10,
random_state=42,
n_jobs=-1,
average=True),
KNeighborsClassifier(3),
# SVC(kernel="linear", C=0.025),
# SVC(gamma=2, C=1),
DecisionTreeClassifier(max_depth=15),
RandomForestClassifier(max_depth=15, n_estimators=10, max_features=1),
AdaBoostClassifier(),
GaussianNB(),
LDA(),
QDA()
]
X_train, X_test, y_train, y_test = common.load_train_data_and_split(
file='data/processed_missing_filled_in.csv')
X_train = np.asarray(X_train)
y_train = np.array(y_train)
y_train = y_train.astype(np.int32)
X_train, y_train = smote.smote_data(X_train, y_train)
# iterate over classifiers
for name, clf in zip(names, classifiers):
print("Fitting " + name + "...")
predicted_test = clf.fit(X_train, y_train).predict(X_test)
test_p = ((y_test != predicted_test).sum()) / (len(X_test)) * 100
print("Error on test set: %d" % test_p)
('Percentile', SelectPercentile()): {
'Percentile__percentile': (1, 5)
},
('PCA', PCA()): {
'PCA__n_components': (2, 4, 8, 16, 32)
}
}
learners = {
('SGD', SGDClassifier()): {
'SGD__loss': ('hinge', 'squared_hinge', 'modified_huber'),
'SGD__penalty': ('l2', 'l1', 'elasticnet'),
'SGD__alpha': tuple([0.1 ** x for x in range(1, 5)])
}
}
data_train, data_test, target_train, target_test = common.load_train_data_and_split(num_samples_per_class=6000, file='data/processed_missing_filled_in.csv') # 0.21
for alg_name, pipeline, params in build_pipelines(learners, selectors):
grid = GridSearchCV(pipeline, params, cv=3, scoring='f1_weighted')
grid.fit(data_train, target_train)
predictions = grid.predict(data_test)
get_results.save_results(target_test, predictions, alg_name, alg_name)
print(grid.best_estimator_)
print(grid.best_params_)
print(grid.best_score_)
print(metrics.classification_report(target_test, predictions))
# learners = {
# ('sgd', SGDClassifier()): {
# 'sgd__loss': ('hinge', 'squared_hinge', 'modified_huber'),
# 'sgd__penalty': ('l2', 'l1', 'elasticnet'),
# 'sgd__kernel': ('rbf', 'sigmoid', 'linear'),
# 'sgd__alpha': tuple([0.1 ** x for x in range(1, 5)])
# }
# }
params = {
'base_estimator__loss': ['hinge', 'modified_huber'],
'base_estimator__penalty': ['l2', 'l1', 'elasticnet'],
'base_estimator__alpha': [0.1 ** x for x in range(1, 5)]
}
data_train, data_test, target_train, target_test = common.load_train_data_and_split()
sgd = SGDClassifier()
bagger = BaggingClassifier(sgd)
grid = GridSearchCV(bagger, params, cv=10)
grid.fit(data_train, target_train)
print(grid.best_estimator_)
print(grid.best_params_)
print(grid.best_score_)
predictions = grid.predict(data_test)
print(metrics.precision_recall_fscore_support(target_test, predictions))
# for score in grid.grid_scores_:
# learners = {
# ('sgd', SGDClassifier()): {
# 'sgd__loss': ('hinge', 'squared_hinge', 'modified_huber'),
# 'sgd__penalty': ('l2', 'l1', 'elasticnet'),
# 'sgd__kernel': ('rbf', 'sigmoid', 'linear'),
# 'sgd__alpha': tuple([0.1 ** x for x in range(1, 5)])
# }
# }
params = {
'base_estimator__loss': ['hinge', 'modified_huber'],
'base_estimator__penalty': ['l2', 'l1', 'elasticnet'],
'base_estimator__alpha': [0.1**x for x in range(1, 5)]
}
data_train, data_test, target_train, target_test = common.load_train_data_and_split(
)
sgd = SGDClassifier()
bagger = BaggingClassifier(sgd)
grid = GridSearchCV(bagger, params, cv=10)
grid.fit(data_train, target_train)
print(grid.best_estimator_)
print(grid.best_params_)
print(grid.best_score_)
predictions = grid.predict(data_test)
print(metrics.precision_recall_fscore_support(target_test, predictions))
# for score in grid.grid_scores_:
names = ["SGD", "Nearest Neighbors", # "Linear SVM", "RBF SVM",
"Decision Tree", "Random Forest", "AdaBoost", "Naive Bayes", "LDA", "QDA"]
classifiers = [
SGDClassifier(loss='hinge', penalty='l2', alpha=0.005, n_iter=10, random_state=42, n_jobs=-1, average=True),
KNeighborsClassifier(3),
# SVC(kernel="linear", C=0.025),
# SVC(gamma=2, C=1),
DecisionTreeClassifier(max_depth=15),
RandomForestClassifier(max_depth=15, n_estimators=10, max_features=1),
AdaBoostClassifier(),
GaussianNB(),
LDA(),
QDA()
]
X_train, X_test, y_train, y_test = common.load_train_data_and_split(file='data/processed_missing_filled_in.csv')
X_train = np.asarray(X_train)
y_train = np.array(y_train)
y_train = y_train.astype(np.int32)
X_train, y_train = smote.smote_data(X_train, y_train)
# iterate over classifiers
for name, clf in zip(names, classifiers):
print("Fitting " + name + "...")
predicted_test = clf.fit(X_train, y_train).predict(X_test)
test_p = ((y_test != predicted_test).sum())/(len(X_test))*100
print("Error on test set: %d" % test_p)
'loss': ['hinge', 'squared_hinge', 'modified_huber'],
# 'loss': ['hinge'],
'penalty': ['l2', 'l1', 'elasticnet'],
# 'penalty': ['l2', 'l1', 'elasticnet'],
'alpha': [0.1 ** x for x in range(1, 5)]
# 'alpha': [.001]
}
#data_train, data_test, target_train, target_test = common.load_test_train_as_two_class(f='data/processed_missing_filled_in.csv')
#data_train, data_test, target_train, target_test = common.load_test_train_as_two_class(f='data/processed_without_missing.csv')
#data_train, data_test, target_train, target_test = common.load_train_data_and_split() # 0.53
#data_train, data_test, target_train, target_test = common.load_train_data_and_split(num_samples_per_class=3000) # 0.24
#data_train, data_test, target_train, target_test = common.load_train_data_and_split(num_samples_per_class=6000, file='data/processed_missing_filled_in.csv') # 0.21
data_train, data_test, target_train, target_test = common.load_train_data_and_split(file='data/processed_missing_filled_in.csv') # 0.49
sgd = SGDClassifier()
grid = GridSearchCV(sgd, params, cv=10, verbose=10)
grid.fit(data_train, target_train)
print(grid.best_estimator_)
print(grid.best_params_)
print(grid.best_score_)
predictions = grid.predict(data_test)
np.save('data/predictions', predictions)
#print(metrics.precision_recall_fscore_support(target_test, predictions))
print(metrics.classification_report(target_test, predictions))
cm = confusion_matrix(target_test, predictions)
def main():
datafiles = [
'data/processed_missing_filled_in.csv',
'data/processed_without_missing.csv', 'data/processed.csv'
]
datanames = ['md=imputed', 'md=deleted', 'md=0s']
num_samples_per_class = [-1] #, 6000]
nsnames = ['ns=all'] #, 'ns=6000']
num_classes = [2, 3]
cnames = ['nc=2', 'nc=3']
oversample = [True] #, False]
osnames = ["os=t"] #, "os=f"]
algnames = [
"NN", "DT", "RandomForest", "AdaBoost", "GaussianNB", "LDA", "QDA",
"SGD", "NNet"
]
algs = [
KNeighborsClassifier(5),
DecisionTreeClassifier(max_depth=25),
RandomForestClassifier(max_depth=25, n_estimators=10, max_features=1),
AdaBoostClassifier(),
GaussianNB(),
LDA(),
QDA(),
SGDClassifier(penalty='elasticnet', alpha=0.1, loss='modified_huber'),
0
]
for alg, algname in zip(algs, algnames):
for dat, datname in zip(datafiles, datanames):
for numspl, sname in zip(num_samples_per_class, nsnames):
for numcls, cname in zip(num_classes, cnames):
for os, osname in zip(oversample, osnames):
algdesc = algname + "_" + datname + "_" + sname + "_" + cname + "_" + osname
print(algdesc)
input_train, input_test, output_train, output_test = common.load_train_data_and_split(
file=dat,
num_samples_per_class=numspl,
num_classes=numcls,
smote=os)
if algname is "NNet":
alg = NeuralNet(layers=[('input', InputLayer),
('dense0', DenseLayer),
('dropout0', DropoutLayer),
('dense1', DenseLayer),
('dropout1', DropoutLayer),
('output', DenseLayer)],
input_shape=(None,
input_train.shape[1]),
dense0_num_units=300,
dropout0_p=0.075,
dropout1_p=0.1,
dense1_num_units=750,
output_num_units=numcls,
output_nonlinearity=softmax,
update=nesterov_momentum,
update_learning_rate=0.001,
update_momentum=0.99,
eval_size=0.33,
verbose=1,
max_epochs=15)
model = alg.fit(input_train, output_train)
print("TRAIN ", algdesc)
predictions_train = model.predict(input_train)
save_results(output_train, predictions_train,
algdesc + "_train", algname)
print("TEST ", algdesc)
predictions_test = model.predict(input_test)
save_results(output_test, predictions_test,
algdesc + "_test", algname)
params = {
'loss': ['hinge', 'squared_hinge', 'modified_huber'],
# 'loss': ['hinge'],
'penalty': ['l2', 'l1', 'elasticnet'],
# 'penalty': ['l2', 'l1', 'elasticnet'],
'alpha': [0.1**x for x in range(1, 5)]
# 'alpha': [.001]
}
#data_train, data_test, target_train, target_test = common.load_test_train_as_two_class(f='data/processed_missing_filled_in.csv')
#data_train, data_test, target_train, target_test = common.load_test_train_as_two_class(f='data/processed_without_missing.csv')
#data_train, data_test, target_train, target_test = common.load_train_data_and_split() # 0.53
#data_train, data_test, target_train, target_test = common.load_train_data_and_split(num_samples_per_class=3000) # 0.24
#data_train, data_test, target_train, target_test = common.load_train_data_and_split(num_samples_per_class=6000, file='data/processed_missing_filled_in.csv') # 0.21
data_train, data_test, target_train, target_test = common.load_train_data_and_split(
file='data/processed_missing_filled_in.csv') # 0.49
sgd = SGDClassifier()
grid = GridSearchCV(sgd, params, cv=10, verbose=10)
grid.fit(data_train, target_train)
print(grid.best_estimator_)
print(grid.best_params_)
print(grid.best_score_)
predictions = grid.predict(data_test)
np.save('data/predictions', predictions)
#print(metrics.precision_recall_fscore_support(target_test, predictions))
print(metrics.classification_report(target_test, predictions))
cm = confusion_matrix(target_test, predictions)