def analysis(x_tr, y_tr, x_te=None, y_te=None):
#print("Performing KNN Classification!")
# Create the classifier
knn = KNeighborsClassifier(n_neighbors=3)
# Train the model
knn.fit(x_tr, y_tr)
# Compute the training accuracy
acc = knn.score(x_tr, y_tr)
# Compute the CV scores
scores = cross_val_score(knn, x_tr, y_tr, cv=5)
print("\n")
print("KNN Accuracy = %3.4f" % (acc))
print("CV Accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2))
# Classify the data
test_score = 0
if x_te is not None:
yhat = knn.predict(x_te)
test_score, notneeded = hp.check_accuracy(yhat, y_te)
else:
yhat = None
data_scores = np.array([scores.mean(), scores.std(), acc, test_score])
return yhat, data_scores
def analysis(x_tr, y_tr, x_te=None, y_te=None):
#print("Performing Bagging Classification!")
# Create the classifier
clf = BaggingClassifier(n_estimators=100)
# Train the model
clf.fit(x_tr, y_tr)
# Compute the training accuracy
acc = clf.score(x_tr, y_tr)
# Compute the CV scores
scores = cross_val_score(clf, x_tr, y_tr, cv=5)
print("\n")
print("Bagging Accuracy = %3.4f" % (acc))
print("CV Accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2))
# Classify the data
test_score = 0
if x_te is not None:
yhat = clf.predict(x_te)
test_score, notneeded = hp.check_accuracy(yhat, y_te)
else:
yhat = None
data_scores = np.array([scores.mean(), scores.std(), acc, test_score])
return yhat, data_scores
def analysis(x_tr,y_tr,x_te=None,y_te=None):
#print("Performing Decision Tree Classification!")
# Create the classifier
decision_tree = DecisionTreeClassifier(max_depth=7)
# Train the model
decision_tree.fit(x_tr, y_tr)
# Compute the training accuracy
acc = decision_tree.score(x_tr, y_tr)
# Compute the CV scores
scores = cross_val_score(decision_tree, x_tr, y_tr, cv=5)
print("\n")
print("Decision Tree Accuracy = %3.4f" % (acc))
print("CV Accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2))
# Classify the data
test_score = 0
if x_te is not None:
yhat = decision_tree.predict(x_te)
test_score, notneeded = hp.check_accuracy(yhat,y_te)
else:
yhat = None
data_scores = np.array([scores.mean(),scores.std(),acc,test_score])
return yhat,data_scores
def analysis(x_tr, y_tr, x_te=None, y_te=None):
#print("Performing RBF SVM Classification!")
# Create the SVM classifier
svm = SVC(kernel='rbf', C=1)
# Train the models
svm.fit(x_tr, y_tr)
# Compute the training accuracy
acc = svm.score(x_tr, y_tr)
# Compute the CV scores
scores = cross_val_score(svm, x_tr, y_tr, cv=5)
print("\n")
print("RBF SVM Accuracy = %3.4f" % (acc))
print("CV Accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2))
# Classify the data
test_score = 0
if x_te is not None:
yhat = svm.predict(x_te)
test_score, notneeded = hp.check_accuracy(yhat, y_te)
else:
yhat = None
data_scores = np.array([scores.mean(), scores.std(), acc, test_score])
return yhat, data_scores
def analysis(x_tr, y_tr, x_te=None, y_te=None):
mode = 1
#print("Performing SSL Classification!")
# Perform K-fold cross validation
k_fold = KFold(n_splits=5)
clf = RandomForestClassifier(n_estimators=100)
NUM_ITER = 2
THRESHOLD = 0.9
scores = []
x_tr_values = x_tr.values
y_tr_values = y_tr.values
for train_indices, test_indices in k_fold.split(x_tr.values):
x_tr_kfold = x_tr_values[train_indices]
x_te_kfold = x_tr_values[test_indices]
y_tr_kfold = y_tr_values[train_indices]
y_te_kfold = y_tr_values[test_indices]
score, temp = train_ssl(
clf,
x_tr_kfold, #labeled
y_tr_kfold, #labeled
x_te.values,
x_te_kfold,
y_te_kfold,
NUM_ITER,
THRESHOLD,
mode)
scores.append(score)
scores = np.array(scores)
#print("Scoring Training accuracy")
acc, temp = train_ssl(clf, x_tr_values, y_tr_values, x_te.values,
x_tr_values, y_tr_values, NUM_ITER, THRESHOLD, mode)
print("\n")
print("SSL Accuracy = %3.4f" % (acc))
print("CV Accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2))
# Classify the data
test_score = 0
if x_te is not None:
mode = 0
temp, yhat = train_ssl(clf, x_tr_values, y_tr_values, x_te.values,
x_te.values, y_te.values, NUM_ITER, THRESHOLD,
mode)
test_score, notneeded = hp.check_accuracy(yhat, y_te)
else:
yhat = None
data_scores = np.array([scores.mean(), scores.std(), acc, test_score])
return yhat, data_scores
def analysis(x_tr, y_tr, x_te=None, y_te=None):
#print("Performing MLP Classification!")
# Create the classifier
clf = MLPClassifier(solver='adam',
alpha=1e5,
hidden_layer_sizes=(100, 100),
random_state=1,
max_iter=500,
momentum=0.9,
learning_rate_init=0.002)
# Create a scaler to scale the features (mean=0, var=1)
scaler = StandardScaler()
# Fit
scaler.fit(x_tr)
# Scale
x_tr = scaler.transform(x_tr)
x_te = scaler.transform(x_te)
# Train the model
clf.fit(x_tr, y_tr)
# Compute the training accuracy
acc = clf.score(x_tr, y_tr)
# Compute the CV scores
scores = cross_val_score(clf, x_tr, y_tr, cv=5)
print("\n")
print("MLP_SK Accuracy = %3.4f" % (acc))
print("CV Accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2))
# Classify the data
test_score = 0
if x_te is not None:
yhat = clf.predict(x_te)
test_score, notneeded = hp.check_accuracy(yhat, y_te)
else:
yhat = None
data_scores = np.array([scores.mean(), scores.std(), acc, test_score])
return yhat, data_scores
def analysis(x_tr,y_tr,x_te=None,y_te=None):
#print("Performing Linear SVM Classification!")
# Create the SVM classifier
svm = SVC(kernel='linear',C=1)
if (0):
# Genearate a feature matrix of polynomial combinations (Cover's Thm)
poly = PolynomialFeatures(degree=3)
# Map or transform training data to higher level
x_tr = poly.fit_transform(x_tr)
# Map or transform test data to higher level
x_te = poly.fit_transform(x_te)
# Train the models
svm.fit(x_tr, y_tr)
# Compute the training accuracy
acc = svm.score(x_tr, y_tr)
# Compute the CV scores
scores = cross_val_score(svm, x_tr, y_tr, cv=5)
print("\n")
print("Linear SVM Accuracy = %3.4f" % (acc))
print("CV Accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2))
# Classify the data
test_score = 0
if x_te is not None:
yhat = svm.predict(x_te)
test_score, notneeded = hp.check_accuracy(yhat,y_te)
else:
yhat = None
data_scores = np.array([scores.mean(),scores.std(),acc,test_score])
return yhat,data_scores
def train_ssl(clf, X_tr_lab, Y_tr_lab, X_tr_unlab, X_test, Y_test, NUM_ITER,
THRESHOLD, mode):
for i in range(NUM_ITER):
clf.fit(X_tr_lab, Y_tr_lab)
result = clf.predict(X_test)
#print("Accuracy after %d rounds: %.2f" % (i,accuracy))
idxs_d = []
if (X_tr_unlab.size > 0):
for j, point in enumerate(X_tr_unlab):
prob = clf.predict_proba(point.reshape(1, -1))
class_w = np.argmax(prob)
if (prob[0, class_w] > THRESHOLD):
# add data to labeled data
X_tr_lab = np.vstack((X_tr_lab, point))
Y_tr_lab = np.hstack((Y_tr_lab, class_w))
idxs_d.append(j)
# remove data from unlabed data
X_tr_unlab = np.delete(X_tr_unlab, idxs_d, 0)
#print("Added %d data after this round" % (len(idxs_d)))
#print("%.2f" % (accuracy))
Y_pred = clf.predict(X_test)
if (mode == 1):
accuracy, pass_fail = hp.check_accuracy(Y_pred, Y_test)
accuracy = accuracy / 100
else:
accuracy = None
return accuracy, Y_pred
def analysis(x_tr, y_tr, x_te=None, y_te=None):
# Set up to perform k-fold cross validation
k_fold = KFold(n_splits=5)
HIDDEN = [100, 100, 100, 100, 100]
NUM_STEPS = 1000
mode = 1
feature_columns = [
tf.feature_column.numeric_column("x", shape=[1, x_tr.values.shape[1]])
]
# Normalize the data for the DNN
for column in x_tr:
#print(column)
x_tr = ti.normalize_column(x_tr, column)
for column in x_te:
#print(column)
x_te = ti.normalize_column(x_te, column)
classifier = tf.estimator.DNNClassifier(
feature_columns=feature_columns,
hidden_units=HIDDEN,
optimizer=tf.train.AdamOptimizer(1e-4),
n_classes=2,
dropout=0.1)
scores = []
if mode == 1:
for train_indices, test_indices in k_fold.split(x_tr.values):
x_tr_values = x_tr.values
y_tr_values = y_tr.values
x_tr_kfold = x_tr_values[train_indices]
x_te_kfold = x_tr_values[test_indices]
y_tr_kfold = y_tr_values[train_indices]
y_te_kfold = y_tr_values[test_indices]
train_input_fn = tf.estimator.inputs.numpy_input_fn(
x={"x": x_tr_kfold},
y=y_tr_kfold,
num_epochs=None,
batch_size=50,
shuffle=True)
classifier.train(input_fn=train_input_fn, steps=NUM_STEPS)
train_input_fn = tf.estimator.inputs.numpy_input_fn(
x={"x": x_te_kfold}, y=y_te_kfold, num_epochs=1, shuffle=False)
scores.append(
classifier.evaluate(input_fn=train_input_fn)["accuracy"])
else:
scores = []
scores = np.array(scores)
# Set up to get training accuracy
train_input_fn = tf.estimator.inputs.numpy_input_fn(x={"x": x_tr.values},
y=y_tr.values,
num_epochs=None,
batch_size=50,
shuffle=True)
classifier.train(input_fn=train_input_fn, steps=NUM_STEPS)
train_input_fn = tf.estimator.inputs.numpy_input_fn(x={"x": x_tr.values},
y=y_tr.values,
num_epochs=1,
shuffle=False)
acc = classifier.evaluate(input_fn=train_input_fn)["accuracy"]
print("\n")
print("DNN Accuracy = %3.4f" % (acc))
print("CV Accuracy: %0.2f (+/- %0.2f)" % (scores.mean(), scores.std() * 2))
# Classify the data
test_score = 0
if x_te is not None:
train_input_fn = tf.estimator.inputs.numpy_input_fn(
x={"x": x_te.values}, num_epochs=1, shuffle=False)
predictions = classifier.predict(input_fn=train_input_fn)
y_pred = []
for i in predictions:
y_pred.append(int(i['classes']))
yhat = np.array(y_pred)
test_score, notneeded = hp.check_accuracy(yhat, y_te)
else:
yhat = None
data_scores = np.array([scores.mean(), scores.std(), acc, test_score])
return yhat, data_scores