def w_indicate(d):
myData = rg.artificial(200, 100, dataType=d)
x = myData.xTrain
x = np.insert(x, 1, 1, axis=0)
xs = myData.xTrain.shape[0]
x_t = x.transpose()
y = myData.yTrain
sum1 = np.zeros((xs, xs))
sum2 = np.zeros((xs, xs))
#pdb.set_trace()
if d == "1D":
for i in np.arange(myData.xTrain.shape[1]):
la1 = x[:, i]
la2 = x_t[i, :]
la1 = la1[np.newaxis, :]
la2 = la2[:, np.newaxis]
ans1 = la1 * la2
ans2 = y[i] * la1
#pdb.set_trace()
sum1 = sum1 + ans1
sum2 = sum2 + ans2
else:
for i in np.arange(myData.xTrain.shape[1]):
sum1 = np.append(sum1, x[:, i] * x_t[i, :])
sum2 = np.append(sum2, y[i] * x[:, i])
pdb.set_trace()
sum1_inv = np.linalg.inv(sum1)
w = sum1_inv * sum2
#pdb.set_trace()
print(w)
#self.w = np.zeros([self.xDim,1])
#------------------------------------
# クラスの定義終わり
#-------------------
#-------------------
# メインの始まり
if __name__ == "__main__":
#------------------------------------
#liner
# 1) 学習入力次元が2の場合のデーター生成
myData = rg.artificial(200, 100, dataType="1D")
#myData = rg.artificial(200,100, dataType="1D",isNonlinear=True)
# 2) 線形回帰モデル
regression = linearRegression(myData.xTrain, myData.yTrain)
#regression = linearRegression(myData.xTrain,myData.yTrain,kernelType="gaussian",kernelParam=1)
# 3) 学習（For文版）
sTime = time.time()
regression.train()
eTime = time.time()
print("train with for-loop: time={0:.4} sec".format(eTime - sTime))
# 4) 学習（行列版）
sTime = time.time()
regression.trainMat()
# coding: utf-8 import regressionData as rg import numpy as np import pdb if __name__ == "__main__": m = rg.artificial(200, 100, dataType="2D") xTrain = np.append(m.xTrain, np.ones((1, m.xTrain.shape[1])), axis=0) # w = np.matmul(np.linalg.inv(np.matmul(xTrain, xTrain.T)), np.sum([y * xTrain[:, index] for index, y in enumerate(m.yTrain)], axis=0)) w = np.matmul(np.linalg.inv(np.matmul(xTrain, xTrain.T)), np.sum(m.yTrain * xTrain, axis=1)) print(w) # pdb.set_trace()
import regressionData as rg
myData1 = rg.artificial(200, 100, dataType="1D")
myData2 = rg.artificial(200, 100, dataType="2D")
x1 = myData1.xTrain
x2 = myData2.xTrain
w1 = myData1.train(x1)
print("1次元\n",w1)
w2 = myData2.train(x2)
print("2次元\n",w2)
y = y[np.newaxis]
loss = np.sum(pow(y - f_x, 2)) / (y - f_x).shape[1]
return loss
#------------------------------------
# クラスの定義終わり
#-------------------
#-------------------
# メインの始まり
if __name__ == "__main__":
# 1) 学習入力次元が2の場合のデーター生成
myData = rg.artificial(200, 100, dataType="1D")
# myData = rg.artificial(200,100, dataType="1D",isNonlinear=True)
# myData = rg.artificial(200,100, dataType="2D")
# myData = rg.artificial(200,100, dataType="2D",isNonlinear=True)
# 2) 線形回帰モデル
#regression = linearRegression(myData.xTrain,myData.yTrain)
regression = linearRegression(myData.xTrain,
myData.yTrain,
kernelType="gaussian",
kernelParam=1)
# 3) 学習（For文版）
sTime = time.time()
# regression.trainMat()
eTime = time.time()
