Applications to nonparameter regression_table2.py

import numpy as np
from sklearn import tree
from scipy import stats


def N(n,p,se):
    X = np.random.uniform(0,1,(n,p))
    Y = np.random.randn(n)*se
    return np.mean(Y*Y),X,Y

def s(x1,x2):
    if (x1<=0.6):
        if (x2<=0.3):
            return 0
        else:
            if (x1<=0.3):
                return -2
            else:
                return -1
    else:
        if (x2<=0.8):
            return 2
        else:
            return 2.5

def T(n,p,se):
    X = np.random.uniform(0,1,(n,p))
    Y= np.zeros(n)
    noise = np.zeros(n)
    for i in range(n):
        Y[i] = s(X[i][0],X[i][1]) + np.random.randn()*se
        noise[i] = (Y[i] - s(X[i][0],X[i][1]))**2
    return np.mean(noise),X,Y

def D(T,X,Y,a,func):
    n = len(Y)
    delta_t = np.zeros(n)
    delta = np.zeros((T,n))
    u = np.zeros((T,n))
    for t in range(T):
        tmp = np.random.randn(n)*a
        delta_t = stats.norm.pdf(tmp/a,0,1)/a
        func.fit(X,Y+delta_t)
        u[t,:] = func.predict(X)
        delta[t,:] = delta_t
    ans = 0
    mean = np.mean(delta,axis=0)
    mean_u = np.mean(u,axis=0)
    for i in range(n):
        hi = np.dot(delta[:,i]-mean[i],u[:,i]-mean_u[i])/np.dot(delta[:,i]-mean[i],delta[:,i]-mean[i])
        ans += hi
    return ans
    
if __name__ == '__main__':   
    min_node = 2
    max_node = 6
    ite = 100
    print('\tModel\t1\t2\t3\t4\t5')
    GDF = np.zeros((max_node-min_node+1,ite))
    for i in range(ite):
        v = list()
        p = 10
        n = 100   
        se = 0.5
        s2,X,Y = N(n,p,se)
        T1 = 100
        a = 0.5*se
        for node in range(min_node,max_node+1):
            model = tree.DecisionTreeRegressor(max_leaf_nodes=node)
            GD = D(T1,X,Y,a,model)
            GDF[node-min_node][i] = GD
    print('\tN10', end='')
    for i in range(max_node-min_node):
        print("\t%.2f" % np.mean(GDF[i]), end='')
    print("\t%.2f" % np.mean(GDF[max_node-min_node]))     

    GDF = np.zeros((max_node-min_node+1,ite))      
    for i in range(ite):
        v = list()
        p = 10
        n = 100   
        se = 0.5
        s2,X,Y = T(n,p,se)
        T1 = 100
        a = 0.5*se
        for node in range(min_node,max_node+1):
            model = tree.DecisionTreeRegressor(max_leaf_nodes=node)
            GD = D(T1,X,Y,a,model)
            GDF[node-min_node][i] = GD
        
    print('\tT10', end='')
    for i in range(max_node-min_node):
        print("\t%.2f" % np.mean(GDF[i]), end='')
    print("\t%.2f" % np.mean(GDF[max_node-min_node]))