def watermelon_example():
"""
GBDT的例子，以西瓜数据集为例
GBDT暂时只支持回归操作，不支持分类
:return:
"""
x, y = get_watermelon()
y = x[:, -1] # y为连续标签
x = x[:, :-1] # x为离散标签
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
gbdt = GBDT(learning_rate=1)
gbdt.fit(x_train, y_train)
print(gbdt.predict(x_test), y_test)
print("R square: %.4f" % gbdt.score(x_test, y_test))
x, y = get_wine()
y = x[:, -1] # y为连续标签
x = x[:, :-1] # x为离散标签
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
gbdt = GBDT(learning_rate=1)
gbdt.fit(x_train, y_train)
print(gbdt.predict(x_test), y_test)
print("R square: %.4f" % gbdt.score(x_test, y_test))
def moon_example():
"""
AdaBoost的例子，以月亮数据集为例
:return:
"""
x, y = get_wine()
x = x[(y == 0) | (y == 1)]
y = y[(y == 0) | (y == 1)]
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
# 采用Logistic回归作为子分类器的AdaBoost
ada = AdaBoost(classifier=ClassifierType.LR)
ada.fit(x_train, y_train)
print(ada.score(x_test, y_test))
ada.classify_plot(x_test, y_test, ", LR")
# 采用KNN作为子分类器的AdaBoost
ada = AdaBoost(classifier=ClassifierType.KNN)
ada.fit(x_train, y_train)
print(ada.score(x_test, y_test))
ada.classify_plot(x_test, y_test, ", KNN")
# 采用CART树为子分类器的AdaBoost
ada = AdaBoost(classifier=ClassifierType.CART)
ada.fit(x_train, y_train)
print(ada.score(x_test, y_test))
ada.classify_plot(x_test, y_test, ", CART")
def iris_example():
x, y = get_iris()
x = x[(y == 0) | (y == 1)]
y = y[(y == 0) | (y == 1)]
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
logistic = LogisticRegression()
logistic.fit(x_train, y_train)
print(logistic.w)
logistic.predict(x_test)
logistic.score(x_test, y_test)
logistic.classify_plot(x_test, y_test)
logistic.auc_plot(x_test, y_test)
lasso = Lasso()
lasso.fit(x_train, y_train)
print(lasso.w)
lasso.predict(x_test)
lasso.score(x_test, y_test)
lasso.classify_plot(x_test, y_test)
ridge = Ridge()
ridge.fit(x_train, y_train)
print(ridge.w)
ridge.predict(x_test)
ridge.score(x_test, y_test)
ridge.classify_plot(x_test, y_test)
def ID3_example():
x, y = get_watermelon()
x = x[:, :4]
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
id3 = ID3()
id3.fit(x_train, y_train)
print(id3.score(x_test, y_test))
def multi_svm_example():
x, y = get_wine() # get moon()
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
svm = SVM(kernel=KernelType.polynomial, d=2)
svm.fit(x_train, y_train)
y_predict = svm.predict(x_test)
print(y_predict, list(y_test))
def wine_example():
x, y = get_wine()
x = x[(y == 0) | (y == 1)]
y = y[(y == 0) | (y == 1)]
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
# 贝叶斯最小错误率
bme = BayesMinimumError()
bme.fit(x_train, y_train)
print(bme.score(x_test, y_test))
bme.classify_plot(x_test, y_test)
# 贝叶斯最小风险，需要给定风险矩阵
# 风险矩阵 [[0,100], [10,0]] 表示把0分为1（存伪）的损失为100，把1分为0（弃真）的损失为10
bmr = BayesMinimumRisk(np.array([[0, 100], [10, 0]]))
bmr.fit(x_train, y_train)
bmr.predict(x_test)
print(bmr.score(x_test, y_test))
bmr.classify_plot(x_test, y_test)
# 朴素贝叶斯
nb = NaiveBayes()
nb.fit(x_train, y_train)
nb.predict(x_test)
print(nb.score(x_test, y_test))
nb.classify_plot(x_test, y_test)
def iris_svm_example():
x, y = get_iris()
# x = x[(y == 0) | (y == 1)]
# y = y[(y == 0) | (y == 1)]
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
svm = SVM(c=0.1)
svm.fit(x_train, y_train)
print(svm.predict(x_test), y_test)
def wine_example():
x, y = get_wine()
# knn可以解决多分类问题
# x = x[(y == 0) | (y == 1)]
# y = y[(y == 0) | (y == 1)]
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
knn = KNN()
knn.fit(x_train, y_train)
print(knn.score(x_test, y_test))
knn.classify_plot(x_test, y_test)
def iris_example():
x, y = sc.get_iris()
y = x[:, 0]
x = x[:, 1:]
x_train, y_train, x_test, y_test = sd.train_test_split(x, y)
reg = MultiRegression()
reg.fit(x_train, y_train)
print(reg.beta)
print(reg.r_square)
print(reg.score(x_test, y_test))
reg.regression_plot(x_test, y_test, col_id=1)
def multi_class_example():
x, y = get_wine()
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
nn = NeuralNetwork(alpha=0.5, cost_func=CostFunction.square)
nn.clear_all()
nn.add_some_layers(2, 3, active_func=ActiveFunction.relu)
nn.fit(x_train, y_train)
# print(nn.predict_prob(x_test)) # raise error here
nn.classify_plot(x_test, y_test)
def wine_example():
x, y = get_wine()
#x = x[(y == 0) | (y == 1)]
#y = y[(y == 0) | (y == 1)]
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
cart = CART()
cart.fit(x_train, y_train)
print(cart.score(x_test, y_test))
cart.classify_plot(x_test, y_test)
y = x[:, -1]
x = x[:, :-1]
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
cart = CART()
cart.fit(x_train, y_train)
print(cart.score(x_test, y_test))
def random_forest_example():
x, y = get_wine()
# x = x[(y == 0) | (y == 1)]
# y = y[(y == 0) | (y == 1)]
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
rf = RandomForest(4, 50)
rf.fit(x_train, y_train)
print(rf.score(x_test, y_test))
rf.classify_plot(x_test, y_test)
def iris_svr_example():
_x, y = get_iris()
x = _x[:, 1:]
y = _x[:, 0]
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
svr = SVR(c=1.0, eps=0.9, kernel=KernelType.gaussian, sigma=2.0)
svr.fit(x_train, y_train)
y_predict = svr.predict(x_test)
print(np.corrcoef(y_predict.ravel(), y_test))
for i, j in enumerate(y_predict):
print(j, y_test[i])
svr.regression_plot(x_test, y_test, column_id=0)
def wine_example():
x, y = get_wine()
x = x[(y == 0) | (y == 1)]
y = y[(y == 0) | (y == 1)]
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
nn = NeuralNetwork(alpha=0.5, cost_func=CostFunction.square)
nn.clear_all()
nn.add_some_layers(2, 3, active_func=ActiveFunction.relu)
nn.fit(x_train, y_train)
print(nn.predict_prob(x_test))
nn.classify_plot(x_test, y_test)
nn.auc_plot(x_test, y_test)
def wine_example():
x, y = get_wine()
# x = x[(y == 2) | (y == 1)]
# y = y[(y == 2) | (y == 1)]
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 91)
model_list = [logistic.LogisticRegression(),
bayes.NaiveBayes()
] # , SVM(kernel_type=KernelTbbype.gaussian, sigma=1)]
stack = ensemble.Stacking(model_list, k_folder=5)
stack.fit(x_train, y_train)
print(stack.score(x_test, y_test))
print(stack.score_mat)
stack.classify_plot(x_test, y_test)
def multi_class_example():
x, y = get_wine()
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.5, 918)
logistic = LogisticRegression(has_intercept=True)
logistic.fit(x_train, y_train)
print(logistic.predict(x_test))
logistic.classify_plot(x_test, y_test)
logistic = Lasso(has_intercept=True)
logistic.fit(x_train, y_train)
print(logistic.predict(x_test))
logistic.classify_plot(x_test, y_test)
logistic = Ridge(has_intercept=True)
logistic.fit(x_train, y_train)
print(logistic.predict(x_test))
logistic.classify_plot(x_test, y_test)
def AdaBoost_multi_class_example():
x, y = get_wine()
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.3, 918)
ada = AdaBoost(classifier=ClassifierType.LR)
ada.fit(x_train, y_train)
print(ada.predict(x_test))
ada.classify_plot(x_test, y_test, ", LR")
ada = AdaBoost(classifier=ClassifierType.CART)
ada.fit(x_train, y_train)
print(ada.predict(x_test))
ada.classify_plot(x_test, y_test, ", CART")
ada = AdaBoost(classifier=ClassifierType.KNN)
ada.fit(x_train, y_train)
print(ada.predict(x_test))
ada.classify_plot(x_test, y_test, ", CART")
def cross_validation(model,
x,
y,
method=CrossValidationType.holdout,
test_size=0.3,
cv=5,
seed=918):
"""
交叉验证函数
:param model: 模型，继承predict和score方法
:param x: 特征
:param y: 标签
:param method: 交叉验证方法
:param test_size: 训练集占比，仅对holdout方法有用
:param cv: 交叉验证次数，如果是k_folder法，则k=cv
:param seed: 随机种子
:return:
"""
if not isinstance(x, np.ndarray):
raise FeatureTypeError
if x.shape[0] != len(y):
raise SampleNumberMismatchError
result = np.zeros(cv)
if method == CrossValidationType.holdout:
for i in range(cv):
x_train, x_test, y_train, y_test = train_test_split(
x, y, test_size)
model.fit(x_train, y_train)
result[i] = model.score(x_test, y_test)
return result
elif method == CrossValidationType.k_folder:
for i, (test_ids,
train_ids) in enumerate(get_k_folder_idx(x.shape[0], cv,
seed)):
x_test, y_test = x[test_ids], y[test_ids]
x_train, y_train = x[train_ids], y[train_ids]
model.fit(x_train, y_train)
result[i] = model.score(x_test, y_test)
return result
else:
raise CrossValidationTypeError
def wine_example():
x, y = get_wine()
x = x[(y == 0) | (y == 1)]
y = y[(y == 0) | (y == 1)]
x_train, y_train, x_test, y_test = train_test_split(x, y, 0.5, 918)
logistic = LogisticRegression(has_intercept=True)
logistic.fit(x_train, y_train)
logistic.score(x_test, y_test)
print(logistic.w)
logistic.classify_plot(x_test, y_test)
logistic.auc_plot(x_test, y_test)
lasso = Lasso()
lasso.fit(x_train, y_train)
print(lasso.w)
lasso.classify_plot(x_test, y_test)
lasso.auc_plot(x_test, y_test)
ridge = Ridge()
ridge.fit(x_train, y_train)
print(ridge.w)
ridge.classify_plot(x_test, y_test)
ridge.auc_plot(x_test, y_test)