def naive_bayes(X_train, y_train, X_test, y_test):
# naive bayes:
nb = GaussianNB().fit(X_train, y_train)
pred = nb.predict(X_test)
err = evaluate.error_margin(pred, y_test) / 10.0
# print(nb.get_params())
return err, pred
def logistic_reg_op_params(X_train, y_train, X_test, y_test):
parameters = {
"solver": ['newton-cg', 'lbfgs', 'liblinear'],
"multi_class": ['auto'],
"penalty": ['l2'],
"C": [0.01, 0.1, 1, 10, 100]
}
lg = LogisticRegression()
scoring = {
'err_margin':
metrics.make_scorer(evaluate.error_margin, greater_is_better=False)
}
# clf = RandomizedSearchCV(estimator=lg, param_distributions=parameters, n_iter=50, cv=3, verbose=2,
# random_state=1, n_jobs=2, scoring=scoring, refit='err_margin')
clf = GridSearchCV(estimator=lg,
param_grid=parameters,
verbose=2,
n_jobs=2,
scoring=scoring,
refit='err_margin')
clf.fit(X_train, y_train)
pred = clf.predict(X_test)
err = evaluate.error_margin(pred, y_test) / 10.0
print("Best params:", clf.best_params_)
return err
def svm_clf_op_params(X_train, y_train, X_test, y_test):
# parameters = {"kernel": ['linear', 'poly', 'rbf', 'sigmoid', 'precomputed'],
# "gamma": ['scale', 'auto'],
# "C": [x for x in np.linspace(start=1.0, stop=10.0, num=100, dtype=float)],
# "shrinking": [True, False]}
parameters = {
'C': [0.1, 1, 10, 100],
'gamma': ['auto', 'scale'],
'kernel': ['rbf']
}
svc = SVC()
scoring = {
'err_margin':
metrics.make_scorer(evaluate.error_margin, greater_is_better=False)
}
# clf = RandomizedSearchCV(estimator=svc, param_distributions=parameters, n_iter=20, cv=3, verbose=2,
# random_state=1, n_jobs=2, scoring=scoring, refit='err_margin')
clf = GridSearchCV(estimator=svc,
param_grid=parameters,
n_jobs=2,
scoring=scoring,
refit='err_margin',
verbose=2)
clf.fit(X_train, y_train)
pred = clf.predict(X_test)
err = evaluate.error_margin(pred, y_test) / 10.0
print("Best params:", clf.best_params_)
return err
def svm_clf(X_train, y_train, X_test, y_test):
# svc
# svc = make_pipeline(StandardScaler(), SVC(gamma="auto"))
svc = SVC(gamma="auto")
svc.fit(X_train, y_train)
pred = svc.predict(X_test)
err = evaluate.error_margin(pred, y_test) / 10.0
return err, pred
def random_forest_reg(X_train, y_train, X_test, y_test):
# random forest reg:
random_forest = RandomForestRegressor(n_estimators=100,
min_samples_split=5,
min_samples_leaf=2)
random_forest.fit(X_train, y_train)
pred = random_forest.predict(X_test)
err = evaluate.error_margin(pred, y_test) / 10.0
return err, pred
def logistic_reg(X_train, y_train, X_test, y_test):
# logistic regression
lg = LogisticRegression(random_state=0,
solver='newton-cg',
multi_class='ovr',
max_iter=200,
penalty='l2',
n_jobs=-1)
lg.fit(X_train, y_train)
pred = lg.predict(X_test)
err = evaluate.error_margin(pred, y_test) / 10.0
return err, pred
def knn(X_train, y_train, X_test, y_test):
# knn:
# k = int(math.sqrt(len(X_train[0])))
# k = k if k & 1 else k + 1 # k will be odd
knn = KNeighborsClassifier(algorithm='auto',
n_jobs=-1,
n_neighbors=15,
weights='distance',
metric='euclidean')
knn.fit(X_train, y_train)
pred = knn.predict(X_test)
err = evaluate.error_margin(pred, y_test) / 10.0
return err, pred
def svm_reg(X_train, y_train, X_test, y_test):
# svr
# svr = make_pipeline(StandardScaler(), SVR(gamma="scale", C=1.1, epsilon=1.1))
svr = SVR(gamma="scale", C=1.1, epsilon=1.1)
svr.fit(X_train, y_train)
pred = svr.predict(X_test)
err = evaluate.error_margin(pred, y_test) / 10.0
# print(svr.get_params())
# print(svr.predict(X_example))
# file = open(PREDICTION_FILE, "w")
# for i in range(len(pred)):
# file.write(names_test[i] + ": " + str(pred[i]) + "\n")
# file.close()
return err, pred
def ridge_clf_op_params(X_train, y_train, X_test, y_test):
parameters = {
"alpha": [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0],
"solver": ['svd', 'cholesky', 'lsqr', 'sparse_cg', 'sag', 'saga']
}
ridge = RidgeClassifier()
scoring = {
'err_margin':
metrics.make_scorer(evaluate.error_margin, greater_is_better=False)
}
clf = GridSearchCV(estimator=ridge,
param_grid=parameters,
verbose=2,
n_jobs=2,
scoring=scoring,
refit='err_margin')
clf.fit(X_train, y_train)
pred = clf.predict(X_test)
err = evaluate.error_margin(pred, y_test) / 10.0
print("Best params:", clf.best_params_)
return err
def knn_op_params(X_train, y_train, X_test, y_test):
parameters = {
'weights': ['distance', 'uniform'],
'algorithm': ['auto', 'ball_tree', 'kd_tree', 'brute'],
'n_neighbors': [1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21],
'metric': ['euclidean']
}
scoring = {
'err_margin':
metrics.make_scorer(evaluate.error_margin, greater_is_better=False)
}
knn = KNeighborsClassifier()
clf = GridSearchCV(knn,
parameters,
scoring=scoring,
refit='err_margin',
n_jobs=-1,
verbose=2)
clf.fit(X_train, y_train)
pred = clf.predict(X_test)
err = evaluate.error_margin(pred, y_test) / 10.0
print("Best params:", clf.best_params_)
return err
def random_forest_reg_op_params(X_train, y_train, X_test, y_test):
# parameters = {'bootstrap': [True, False],
# 'max_depth': [10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 110, None],
# 'max_features': ['auto', 'sqrt'],
# 'min_samples_leaf': [1, 2, 4],
# 'min_samples_split': [2, 5, 10],
# 'n_estimators': [100, 150, 200]}
parameters = {
"n_estimators": [10, 100, 500, 1000],
"max_depth": [5, 8, 15, 25, 30],
"min_samples_split": [2, 5, 10, 15, 100],
"min_samples_leaf": [1, 2, 5, 10],
"max_features": list(range(1,
len(X_train[0]) // 2 + 1))
}
random_forest = RandomForestRegressor()
scoring = {
'err_margin':
metrics.make_scorer(evaluate.error_margin, greater_is_better=False)
}
clf = RandomizedSearchCV(estimator=random_forest,
param_distributions=parameters,
n_iter=20,
cv=3,
verbose=2,
random_state=42,
n_jobs=-1,
scoring=scoring,
refit='err_margin')
# clf = GridSearchCV(estimator=random_forest, param_grid=parameters, verbose=2, n_jobs=2,
# scoring=scoring, refit='err_margin')
clf.fit(X_train, y_train)
pred = clf.predict(X_test)
err = evaluate.error_margin(pred, y_test) / 10.0
print("Best params:", clf.best_params_)
return err
def ridge_clf(X_train, y_train, X_test, y_test):
ridge = RidgeClassifier(alpha=0.9, solver='lsqr')
ridge.fit(X_train, y_train)
pred = ridge.predict(X_test)
err = evaluate.error_margin(pred, y_test) / 10.0
return err, pred
def ridge_reg(X_train, y_train, X_test, y_test):
ridge = Ridge(alpha=0.5, solver='sag')
ridge.fit(X_train, y_train)
pred = ridge.predict(X_test)
err = evaluate.error_margin(pred, y_test) / 10.0
return err, pred
