class GradientDescent():
#def __init__(self, X, y, theta, alpha, batch_size):
def __init__(self):
self.rawdata = LoadData('.\ex2data1.txt')
self.X, self.y, self.batch_size, rawdata = self.rawdata.loadTXT()
self.theta = np.array([[0.],[0.],[0.]])
self.alpha = 0.01
self.costlst = []
self.thetalst = []
self.rawdata_p = rawdata[np.where(rawdata[:,2]==1)]
self.rawdata_n = rawdata[np.where(rawdata[:,2]==0)]
# print(self.theta.shape)
# print(self.batch_size)
# print(self.rawdata_p)
# print(self.rawdata_n)
self.y = self.y[0]
def sigmoid(self,inputs):
inputs = np.array(inputs)
sigmoid_scores = [1/float(1 + np.exp(-x)) for x in inputs[0]]
return sigmoid_scores
def costFunction(self):
h = np.matrix(self.sigmoid(self.theta.transpose()*self.X))
print(-self.y.transpose()*np.log(h))
print((1-self.y.transpose())*np.log(1-h))
return 1/(self.batch_size) * np.sum(-self.y.transpose()*np.log(h) - (1-self.y.transpose())*np.log(1-h))
def gradientDescent(self):
diff = self.theta.transpose()*self.X - self.y
self.theta -= self.alpha*(1/self.batch_size) * self.X*diff.transpose()
self.thetalst.append(self.theta)
self.costlst.append(self.costFunction())
def plotCostJ(self):
x = [i for i in range(len(self.costlst))]
plt.plot(x, self.costlst)
def plotCostJTheta(self):
fig = plt.figure()
ax = fig.gca(projection='3d')
x = self.thetalst[0]
y = self.thetalst[1]
z = self.costlst
ax.plot(x, y, z, label='parametric curve')
ax.legend()
plt.show()
class GradientDescent():
#def __init__(self, X, y, theta, alpha, batch_size):
def __init__(self):
self.rawdata = LoadData('.\ex1data1.txt')
self.X, self.y, self.batch_size = self.rawdata.loadTXT()
self.theta = np.array([[0.], [0.]])
self.alpha = 0.01
self.costlst = []
self.thetalst = []
print(self.theta.shape)
print(self.batch_size)
print(np.sum(self.X[:, 1]))
print(np.sum(self.y))
def costFunction(self):
return 1 / (2 * self.batch_size) * np.sum(
np.square(self.theta.transpose() * self.X - self.y))
def gradientDescent(self):
diff = self.theta.transpose() * self.X - self.y
self.theta -= self.alpha * (
1 / self.batch_size) * self.X * diff.transpose()
self.thetalst.append(self.theta)
self.costlst.append(self.costFunction())
def plotCostJ(self):
x = [i for i in range(len(self.costlst))]
plt.plot(x, self.costlst)
def plotCostJTheta(self):
fig = plt.figure()
ax = fig.gca(projection='3d')
x = self.thetalst[0]
y = self.thetalst[1]
z = self.costlst
ax.plot(x, y, z, label='parametric curve')
ax.legend()
plt.show()
