def train(X,
y,
max_depth,
n_estimators,
learning_rate,
subsample=1.0,
K=2,
siGn=0):
Trees = []
m = shape(X)[0]
F0 = 0.5 * log((1 + mean(y)) / (1 - mean(y)))
Fx = ones(m) * F0
threNum = threN(m, max_depth) ##叶节点包含的最小数据个数
while n_estimators > 0:
residual = quasi_residual(y, Fx) ##更新残差
if subsample < 1.0: ##执行亚采样
index = random.choice(range(m),
size=int(subsample * m),
replace=False)
else:
index = array(range(m))
if siGn == 1: ##影响力剪枝
index = influTrim(residual[index])
tree = createTree(X[index], residual[index], max_depth, threNum, K)
thisPred = [] ##本轮预测
Trees.append(tree)
for i in range(m):
thisPred.append(predict(tree, X[i], learning_rate))
Fx += array(thisPred)
n_estimators -= 1
return Trees
def predct(Tree,X_test,y_test,lr):
n=len(Tree)
m=shape(X_test)[0]
pred=zeros(m)
for i in range(m):
for j in range(n):
pred[i]+=predict(Tree[j],X_test[i])*lr
return pred
def validate(Tree,X_test,y_test):
n=len(Tree)
m=shape(X_test)[0]
pred=zeros(m)
for i in range(m):
for j in range(n):
pred[i]+=predict(Tree[j],X_test[i])
return R2_score(pred,y_test)
def test(X, Forest, learning_rate, K):
m = shape(X)[0]
n = len(Forest)
pred = zeros((m, K))
for i in range(m):
for k in range(K):
for j in range(n):
pred[i, k] += predict(Forest[j][k], X[i], learning_rate)
return argmax(pred, axis=1)
def test(X, Trees, learning_rate):
m = shape(X)[0]
n = len(Trees)
pred = zeros(m)
for i in range(m):
for j in range(n):
pred[i] += predict(Trees[j], X[i], learning_rate)
pred[pred > 0] = 1
pred[pred <= 0] = -1
return pred
def get_predict(trees_result, trees_feature, data_train):
m_tree = len(trees_result)
m = np.shape(data_train)[0]
result = []
for i in range(m_tree):
clf = trees_result[i]
feature = trees_feature[i]
data = split_data(data_train, feature)
result_i = []
for i in range(m):
result_i.append(list(predict(data[i][0:-1], clf).keys())[0])
result.append(result_i)
final_predict = np.sum(result, axis=0)
return final_predict
def bdt(X,y,eps=1000,depth=6,lr=0.1):
Tree=[]
m,n=shape(X)
residual=y.copy() ## 残差
while sum(residual**2)>eps:
pred=[]
tree=createTree(X,residual,depth)
Tree.append(tree)
for i in range(m):
pred.append(predict(tree,X[i]))
residual-=array(pred)*lr ##学习率
print(sum(residual**2))
print('Train MSE:',sum(residual**2)/len(X))
return Tree
def train(X, y, max_depth, n_estimators, learning_rate, lamda, eta):
Trees = []
m = shape(X)[0]
F = zeros(m)
while n_estimators > 0:
grad, hes = GraHes(y, F)
tree = createTree(X, max_depth, grad, hes, lamda, eta)
pred = []
Trees.append(tree)
for i in range(m):
pred.append(predict(tree, X[i], learning_rate))
F += array(pred)
## print(F[:10])
n_estimators -= 1
return Trees
def train(X, y, max_depth, n_estimators, learning_rate, K):
Forest = [] ##每个元素都是K个字典组成的列表
m = shape(X)[0]
Fx = zeros((m, K))
threNum = threN(m, max_depth)
Forcast = zeros((m, K))
while n_estimators > 0:
Trees = []
eFx = exp(Fx)
Px = eFx / sum(eFx, axis=1)[:, None] ##按行求和
residual = y - Px
n_estimators -= 1
for k in range(K): ##构造K棵树，分别拟合K类残差
Tree = createTree(X, residual[:, k], max_depth, threNum, K)
Trees.append(Tree)
for i in range(m):
Forcast[i, k] = predict(Tree, X[i], learning_rate)
Fx += Forcast
Forest.append(Trees)
return Forest