def experiment14_17():
for fi in range(2 , -11, -1):
X, Y = create_muldata_fromfile(train_file)
X_train = X[:train_num]
Y_train = Y[:train_num]
X_val = X[train_num:]
Y_val = Y[train_num:]
W = train_ridge_regression_model(X_train, Y_train, 10 ** fi)
Y_p = np.dot(X_train, W).transpose()[0]
etrain = error_01(Y_p, Y_train)
Y_p = np.dot(X_val, W).transpose()[0]
eval = error_01(Y_p, Y_val)
X, Y = create_muldata_fromfile(test_file)
Y_p = np.dot(X, W).transpose()[0]
eout_ = error_01(Y_p, Y)
fi_ex14 = [-4, -6, -2, -10, -8]
fi_ex15 = [-9, -1, -5, -7, -3]
fi_ex16 = [-6, -8, -4, -2, 0]
fi_ex17 = [-3, -8, -6, -9, 0]
if fi in fi_ex17:
print 'fi : %d, errortrain -> %f errorval -> %f errorout -> %f' % (fi, etrain, eval, eout_)
def experiment14_17():
for fi in range(2, -11, -1):
X, Y = create_muldata_fromfile(train_file)
X_train = X[:train_num]
Y_train = Y[:train_num]
X_val = X[train_num:]
Y_val = Y[train_num:]
W = train_ridge_regression_model(X_train, Y_train, 10**fi)
Y_p = np.dot(X_train, W).transpose()[0]
etrain = error_01(Y_p, Y_train)
Y_p = np.dot(X_val, W).transpose()[0]
eval = error_01(Y_p, Y_val)
X, Y = create_muldata_fromfile(test_file)
Y_p = np.dot(X, W).transpose()[0]
eout_ = error_01(Y_p, Y)
fi_ex14 = [-4, -6, -2, -10, -8]
fi_ex15 = [-9, -1, -5, -7, -3]
fi_ex16 = [-6, -8, -4, -2, 0]
fi_ex17 = [-3, -8, -6, -9, 0]
if fi in fi_ex17:
print 'fi : %d, errortrain -> %f errorval -> %f errorout -> %f' % (
fi, etrain, eval, eout_)
def error_with_fi(fi):
X, Y = create_muldata_fromfile(train_file)
W = train_ridge_regression_model(X, Y, fi)
# print(X.shape, W.shape)
Y_p = np.dot(X, W).transpose()[0]
print error_01(Y_p, Y)
X, Y = create_muldata_fromfile(test_file)
Y_p = np.dot(X, W).transpose()[0]
print error_01(Y_p, Y)
def error_with_fi(fi):
X, Y = create_muldata_fromfile(train_file)
W = train_ridge_regression_model(X, Y, fi)
# print(X.shape, W.shape)
Y_p = np.dot(X, W).transpose()[0]
print error_01(Y_p, Y)
X, Y = create_muldata_fromfile(test_file)
Y_p = np.dot(X, W).transpose()[0]
print error_01(Y_p, Y)
def ex_19_20():
X, Y = create_muldata_fromfile(train_data)
X_train = X[:400]
Y_train = Y[:400]
X_test = X[400:]
Y_test = Y[400:]
gammas = [32, 2, 0.125]
rus = [0.001, 1, 1000]
for gamma in gammas:
for ru in rus:
beta = get_beta(X_train, Y_train, gamma, ru)
# X_new = [[4, 6], [3, 5]]
# print 'Beta get'
r = predict_by_X(beta, X_test, X_train, gamma)
print 'gamma|ru|error %f|%f|%f' % (gamma, ru, error_01(r, Y_test))
def avg_error_pertree(self, X, y):
all_error = 0
for i in range(0, self.__n_estimator):
cur_crtree = self.__rf_tree[i]
y_p = cur_crtree.predict(X)
all_error += error_01(y_p, y)
return all_error / self.__n_estimator
def avg_error_pertree(self, X, y):
all_error = 0
for i in range(0, self.__n_estimator):
cur_crtree = self.__rf_tree[i]
y_p = cur_crtree.predict(X)
all_error += error_01(y_p, y)
return all_error / self.__n_estimator
fi_ex17 = [-3, -8, -6, -9, 0]
if fi in fi_ex17:
print 'fi : %d, errortrain -> %f errorval -> %f errorout -> %f' % (
fi, etrain, eval, eout_)
cv_num = 5
X, Y = create_muldata_fromfile(train_file)
group_num = len(X) / cv_num
for fi in [-4, 0, -2, -6, -8]:
all_error = 0
for i in range(cv_num):
X_train = np.vstack((X[0:i * group_num], X[(i + 1) * group_num:]))
Y_train = np.vstack((Y[0:i * group_num], Y[(i + 1) * group_num:]))
X_val = X[i * group_num:(i + 1) * group_num]
Y_val = Y[i * group_num:(i + 1) * group_num]
W = train_ridge_regression_model(X_train, Y_train, 10**fi)
Y_p = np.dot(X_val, W).transpose()[0]
ecv = error_01(Y_p, Y_val)
all_error += ecv
print 'fi : %d, errorcv -> %f ' % (fi, all_error / cv_num)
error_with_fi(1)
if r - 0.5 > 0:
zo_result.append(1)
else:
zo_result.append(-1)
result = zo_result
return result
X, Y = create_muldata_fromfile(train_file)
# X = np.array([[1, 1, 0], [1, 0, 1]])
# Y = np.array([1, -1])
W = train_logistic_model(X, Y, l_n = 0.01, is_sgd=True)
# print(W)
X, Y = create_muldata_fromfile(test_file)
ra = predict(W, X)
print(error_01(ra, Y))
'''
fix learning rate : n = 0.001 0.475
fix learning rate : n = 0.01 0.22 cost:68s
'''
fi_ex16 = [-6, -8, -4, -2, 0]
fi_ex17 = [-3, -8, -6, -9, 0]
if fi in fi_ex17:
print 'fi : %d, errortrain -> %f errorval -> %f errorout -> %f' % (fi, etrain, eval, eout_)
cv_num = 5
X, Y = create_muldata_fromfile(train_file)
group_num = len(X) / cv_num
for fi in [-4, 0, -2, -6, -8]:
all_error = 0
for i in range(cv_num):
X_train = np.vstack((X[0: i * group_num], X[(i + 1) * group_num:]))
Y_train = np.vstack((Y[0: i * group_num], Y[(i + 1) * group_num:]))
X_val = X[i * group_num:(i + 1) * group_num]
Y_val = Y[i * group_num:(i + 1) * group_num]
W = train_ridge_regression_model(X_train, Y_train, 10 ** fi)
Y_p = np.dot(X_val, W).transpose()[0]
ecv = error_01(Y_p, Y_val)
all_error += ecv
print 'fi : %d, errorcv -> %f ' % (fi, all_error / cv_num)
error_with_fi(1)
if r - 0.5 > 0:
zo_result.append(1)
else:
zo_result.append(-1)
result = zo_result
return result
X, Y = create_muldata_fromfile(train_file)
# X = np.array([[1, 1, 0], [1, 0, 1]])
# Y = np.array([1, -1])
W = train_logistic_model(X, Y, l_n=0.01, is_sgd=True)
# print(W)
X, Y = create_muldata_fromfile(test_file)
ra = predict(W, X)
print(error_01(ra, Y))
'''
fix learning rate : n = 0.001 0.475
fix learning rate : n = 0.01 0.22 cost:68s
'''
result = 1 if score > 0 else -1
p_ys.append(result)
return p_ys
datas = get_datas(train_file)
X = [dd[:-1] for dd in datas]
Y = [dd[-1] for dd in datas]
models = train_stump_tree_adaboost(datas, 300)
# models.sort(key=lambda x:x[-1])
# print(models)
yps = adaboost_model_predict(models, X)
print error_01(Y, yps) #ein
datas = get_datas(test_file)
X = [dd[:-1] for dd in datas]
Y = [dd[-1] for dd in datas]
yps = adaboost_model_predict(models, X)
print error_01(Y, yps) #eout
# X = [[1, 25], [0, 3], [2, 5]]
# Y = [1, 1, -1]
# datas = [[1, 25, 1], [0, 3, 1], [2, 5, -1]]
score += ((x[i] - theta) * s) * w
result = 1 if score > 0 else -1
p_ys.append(result)
return p_ys
datas = get_datas(train_file)
X = [dd[:-1] for dd in datas]
Y = [dd[-1] for dd in datas]
models = train_stump_tree_adaboost(datas, 300)
# models.sort(key=lambda x:x[-1])
# print(models)
yps = adaboost_model_predict(models, X)
print error_01(Y, yps) #ein
datas = get_datas(test_file)
X = [dd[:-1] for dd in datas]
Y = [dd[-1] for dd in datas]
yps = adaboost_model_predict(models, X)
print error_01(Y, yps) #eout
# X = [[1, 25], [0, 3], [2, 5]]
# Y = [1, 1, -1]
# datas = [[1, 25, 1], [0, 3, 1], [2, 5, -1]]