def train_rls():
X_train, Y_train, X_test, Y_test = load_newsgroups()
#CGRLS does not support multi-output learning, so we train
#one classifier for the first column of Y. Multi-class learning
#would be implemented by training one CGRLS for each column, and
#taking the argmax of class predictions.
predictions = []
rls = CGRLS(X_train, Y_train[:, 0], regparam=100.0)
P = rls.predict(X_test)
perf = auc(Y_test[:, 0], P)
print("auc for task 1 %f" % perf)
def train_rls():
X_train, Y_train, X_test, Y_test = load_newsgroups()
# CGRLS does not support multi-output learning, so we train
# one classifier for the first column of Y. Multi-class learning
# would be implemented by training one CGRLS for each column, and
# taking the argmax of class predictions.
predictions = []
rls = CGRLS(X_train, Y_train[:, 0], regparam=100.0)
P = rls.predict(X_test)
perf = auc(Y_test[:, 0], P)
print("auc for task 1 %f" % perf)
def testCGRLS(self):
m, n = 100, 300
for regparam in [0.00000001, 1, 100000000]:
Xtrain = np.mat(np.random.rand(m, n))
Y = np.mat(np.random.rand(m, 1))
rpool = {}
rpool["train_features"] = Xtrain
rpool["train_labels"] = Y
rpool["regparam"] = regparam
rpool["bias"] = 1.0
rls = RLS.createLearner(**rpool)
rls.solve(regparam)
model = rls.getModel()
W = model.W
b = model.b
rls = CGRLS.createLearner(**rpool)
rls.train()
model = rls.getModel()
W2 = model.W
b2 = model.b
for i in range(W.shape[0]):
# for j in range(W.shape[1]):
# self.assertAlmostEqual(W[i,j],W2[i,j],places=5)
self.assertAlmostEqual(W[i], W2[i], places=5)
self.assertAlmostEqual(b, b2, places=5)
def testCGRLS(self):
m, n = 100, 300
for regparam in [0.00000001, 1, 100000000]:
Xtrain = np.mat(np.random.rand(m, n))
Y = np.mat(np.random.rand(m, 1))
rpool = {}
rpool['train_features'] = Xtrain
rpool['train_labels'] = Y
rpool['regparam'] = regparam
rpool["bias"] = 1.0
rls = RLS.createLearner(**rpool)
rls.solve(regparam)
model = rls.getModel()
W = model.W
b = model.b
rls = CGRLS.createLearner(**rpool)
rls.train()
model = rls.getModel()
W2 = model.W
b2 = model.b
for i in range(W.shape[0]):
#for j in range(W.shape[1]):
# self.assertAlmostEqual(W[i,j],W2[i,j],places=5)
self.assertAlmostEqual(W[i], W2[i], places=5)
self.assertAlmostEqual(b, b2, places=5)
def testCGRLS(self):
m, n = 100, 300
for regparam in [0.00000001, 1, 100000000]:
Xtrain = np.mat(np.random.rand(m, n))
Y = np.mat(np.random.rand(m, 1))
rpool = {}
rpool['X'] = Xtrain
rpool['Y'] = Y
rpool['regparam'] = regparam
rpool["bias"] = 2.0
rls = RLS(**rpool)
rls.solve(regparam)
model = rls.predictor
W = model.W
b = model.b
rls = CGRLS(**rpool)
model = rls.predictor
W2 = model.W
b2 = model.b
for i in range(W.shape[0]):
self.assertAlmostEqual(W[i], W2[i], places=5)
self.assertAlmostEqual(b, b2, places=5)
def evaluate_topic(X_train, y_train, X_test, classifier):
if classifier == 'svm':
# Create a linear SVM classifier. Instead of subsampling I change the value of the 'class_weight' parameter to 'balanced'
clf = svm.SVC(kernel='linear', class_weight=None, probability=True) # decision_function_shape='ovo'
clf.fit(X_train, y_train)
new_p, new_y_test_predict_log_proba = ([] for i in range(2))
#for val in clf.decision_function(X_test):
# new_p.append(val - min(clf.decision_function(X_test)))
for val in clf.predict_log_proba(X_test)[:, 1]:
new_y_test_predict_log_proba.append(math.exp(val))
y_test = clf.predict_proba(X_test)[:, 1] # The correct is with 1 apparently #y_test = clf.predict_proba(X_test)[:, 0]
new_p = new_y_test_predict_log_proba
return(y_test, new_p)
elif classifier == 'lgb':
regr = LGBMRegressor(boosting_type='gbdt')
regr.fit(X_train, y_train)
y_test = regr.predict(X_test)
''' if you want to use LGBMClassifier ucomment the lines below '''
#train_data = lgb.Dataset(X_train, label=y_train)
#clf = lgb.train(param, train_data)
#y_test = clf.predict(X_test)
#new_p = clf.predict_proba(X_test, num_iteration=clf.best_iteration_)[:, 1] # supported only for LGBMClassifier
new_p = None
return(y_test, new_p)
elif classifier == 'lsvr':
regr = LinearSVR(random_state=0, tol=1e-5)
regr.fit(X_train, y_train)
y_test = regr.predict(X_test)
return(y_test, None)
elif classifier == 'svr':
#svr = SVR(kernel='rbf', C=math.pow(2,C), gamma=math.pow(2,gamma), cache_size=2000, verbose=False, max_iter=-1, shrinking=False)
regr = svm.SVR(C=1.0, epsilon=0.2, probability=True)
regr.fit(X_train, y_train)
y_test = regr.predict(X_test)
return(y_test, None)
elif classifier == 'sgd':
clf = SGDClassifier(loss='hinge', penalty='l2',
alpha=1e-3, random_state=42, max_iter=5, tol=None)
clf.fit(X_train, y_train)
y_test = clf.predict(X_test)
return(y_test, None)
elif classifier == 'xgb':
clf = XGBClassifier()
clf.fit(X_train, y_train)
y_test = clf.predict(X_test)
return(y_test, None)
elif classifier == 'rls':
#clf = CGRLS(X_train, Y_train[:, 0], regparam=100.0)
clf = CGRLS(X_train, y_train, regparam=100.0)
y_test = clf.predict(X_test)
return(y_test, None)
elif classifier == 'dtc':
clf = DecisionTreeClassifier(random_state=0)
clf.fit(X_train, y_train)
y_test = clf.predict(X_test)
return(y_test, None)
elif classifier == 'rfc':
clf = RandomForestClassifier(
n_estimators=100, max_depth=2, random_state=0)
clf.fit(X_train, y_train)
y_test = clf.predict(X_test)
return(y_test, None)
elif classifier == 'kne':
clf = KNeighborsClassifier(n_neighbors=3)
clf = clf.fit(X_train, y_train)
y_test = clf.predict(X_test)
return(y_test, None)
elif classifier == 'smote':
smt = SMOTE()
smt = SMOTE(sampling_strategy='auto')
X_train_sampled, y_train_sampled = smt.fit_sample( X_train.toarray(), np.asarray(y_train))
clf = LinearSVC().fit(X_train_sampled, y_train_sampled)
y_test = clf.predict(X_test)
return(y_test, None)
elif classifier == 'lr':
clf = LogisticRegression()
clf.fit(X_train, y_train)
y_test = clf.predict(X_test)
return(y_test, None)
else:
raise Exception('Wrong classifier')
return(y_test, None)