重复错误：形状 (1,3) 和 (100,) 未对齐：3 (dim 1) != 100 (dim 0)

Question

在我执行的每个代码部分之后，我都会收到此错误，并且无法解决此问题。 以下是代码：请需要解决此问题

成本 Function

def compute_logistic_cost(theta, X, y):    
    m=len(y)
    J = (1/m) * np.sum((-y * np.log(h(theta,X))) - ((1-y)*np.log(1-h(theta,X))))
    eps = 1e-12
    h[h < eps] = eps
    eps = 1.0 - 1e-12
    h[h > eps] = eps

    return J


# Recall that for a model with inputs, we actually use 3 parameters, theta_0, theta_1 
# and theta_2.  The inputs X need to have an initial column of all 1's for the theta_0
# parameter.  So for our current data, X needs to be a 3xm shaped set of data, where 
# the first value in each column is 1.0, and the next value in each column is our raw inputs
X = np.ones( (3, m) )
X[1:,:] = X.T # the second column contains the raw inputs

一个错误：

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-99-596c409301b0> in <module>
4 # the first value in each column is 1.0, and the next value in each column is our raw 
inputs
5 X = np.ones( (3, m) )
----> 6 X[1:,:] = X.T # the second column contains the raw inputs

ValueError: could not broadcast input array from shape (3,3) into shape (2,3)

给定假设的计算阈值

theta = np.zeros( (3, 1) )
print(theta)
print(compute_logistic_cost(theta, X, y))


theta = np.array([[1.0],
              [1.0],
              [1.0]])

print(theta)

print(compute_logistic_cost(theta, X, y))

theta = np.array([[0.1],
              [0.1],
              [0.1]])
print(theta)
print(compute_logistic_cost(theta, X, y))

又报错了：

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-92-d07eafdd63a5> in <module>
  2 print(theta)
  3 print("Hello")
  ----> 4 print(compute_logistic_cost(theta, X, y))
  5 
  6 

 <ipython-input-90-88cf8ecf55dc> in compute_logistic_cost(theta, X, y)
 24     h = sigmoid(X.dot(theta))
 25 
 ---> 26     J = -1*(1/m)*(np.log(h).T.dot(y)+np.log(1-h).T.dot(1-y))
 27 
 28     if np.isnan(J[0]):

 ValueError: shapes (1,3) and (100,) not aligned: 3 (dim 1) != 100 (dim 0)

计算梯度 function：

def compute_logistic_cost_gradients(theta, X, y):
    m = y.size
    h = sigmoid(X.dot(theta.reshape(-1,1)))

    grad =(1/m)*X.T.dot(h-y)

    return(grad)


theta = np.zeros( (3, 1) )
print(compute_logistic_cost_gradients(theta, X, y))
# Expected Result >>> [ -0.1        -12.00921659 -11.26284221]
theta = np.array([[1.0],
              [1.0],
              [1.0]])
print(compute_logistic_cost_gradients(theta, X, y))
# Expected Result >>> [  0.4         20.81292044  21.84815684]
theta = np.array([[0.1],
              [0.1],
              [0.1]])
print(compute_logistic_cost_gradients(theta, X, y))
# Expected Result >>> [  0.39997223  20.81184964  21.84684953]

又是一个错误

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-108-f8672fbf30fe> in <module>
  1 theta = np.zeros( (3, 1) )
----> 2 print(compute_logistic_cost_gradients(theta, X, y))
  3 
  4 theta = np.array([[1.0],
  5                   [1.0],

<ipython-input-107-0af7d4f48cd7> in compute_logistic_cost_gradients(theta, X, y)
  1 def compute_logistic_cost_gradients(theta, X, y):
  2     m = y.size
----> 3     h = sigmoid(X.dot(theta.reshape(-1,1)))
  4 
  5     grad =(1/m)*X.T.dot(h-y)

ValueError: shapes (100,2) and (3,1) not aligned: 2 (dim 1) != 3 (dim 0)

参数使用 SciPy 优化

from scipy.optimize import minimize
theta = np.zeros( 2 ) # initial theta parameters to start gradient descent from
res = minimize(compute_logistic_cost(theta, X, y), theta, method='CG', 
           jac=compute_logistic_cost_gradients(theta, X, y), 
           options={'disp': True})
print(res.x)

一个错误

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-93-bc65484d35ef> in <module>
  1 from scipy.optimize import minimize
  2 theta = np.zeros( 2 ) # initial theta parameters to start gradient descent from
 ----> 3 res = minimize(compute_logistic_cost(theta, X, y), theta, method='CG', 
  4                jac=compute_logistic_cost_gradients(theta, X, y),
  5                options={'disp': True})

 <ipython-input-90-88cf8ecf55dc> in compute_logistic_cost(theta, X, y)
 22     """
 23     m = y.size
 ---> 24     h = sigmoid(X.dot(theta))
 25 
 26     J = -1*(1/m)*(np.log(h).T.dot(y)+np.log(1-h).T.dot(1-y))

 ValueError: shapes (3,3) and (2,) not aligned: 3 (dim 1) != 2 (dim 0)

尝试 plot 分散 plot：

res = minimize(compute_logistic_cost(theta, X, y), theta, method='CG', 
           jac=compute_logistic_cost_gradients(theta, X, y), 
           options={'disp': True})

plt.plot(x[val1, 0], X[val1, 1], linestyle='', marker='^', color='r')
plt.plot(x[val2, 0], X[val2, 1], linestyle='', marker='o', color='y')
# Assigning the calculated θ to a variable
gradBFGS = res['x']

# Calculating x and y for the decision boundary
plot_x = np.array([np.min(X[:, 2])-1, np.max(X[:, 2])+1])

# From the decision boundary calculations x2 = (-1 / θ2) * (θ0 * x1 + θ0)
plot_y = (-1 / gradBFGS[2]) * (gradBFGS[1] * plot_x + gradBFGS[0])
plt.scatter(45, 85, s=30, c='r', marker='x')

# Plotting the data
plotData(X[:,1:], y, 'Exam 1 score', 'Exam 2 score', 'Admitted', 'Not Admitted')
plt.plot(plot_x, plot_y, c='b');

一个错误

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-94-a46335302bf6> in <module>
----> 1 res = minimize(compute_logistic_cost(theta, X, y), theta, method='CG', 
  2                jac=compute_logistic_cost_gradients(theta, X, y),
  3                options={'disp': True})
  4 
  5 # Assigning the calculated θ to a variable

<ipython-input-90-88cf8ecf55dc> in compute_logistic_cost(theta, X, y)
 22     """
 23     m = y.size
---> 24     h = sigmoid(X.dot(theta))
 25 
 26     J = -1*(1/m)*(np.log(h).T.dot(y)+np.log(1-h).T.dot(1-y))

ValueError: shapes (3,3) and (2,) not aligned: 3 (dim 1) != 2 (dim 0)

Answer 1

所有这些错误都是由于您的 numpy arrays 的尺寸与您尝试的操作不兼容。 例如，在第一部分中，您尝试将形状为 (3, 3) 的数组分配给形状为 (2, 3) 的数组。

 ----> 6 X[1:,:] = X.T # the second column contains the raw inputs

ValueError: could not broadcast input array from shape (3,3) into shape (2,3)

请注意，这些 arrays 索引为 0，因此X[1:,:]仅抓取该矩阵的第 2 行和第 3 行。 所有错误都是因为这些形状不匹配，在其他情况下是因为点积中的形状不兼容。

如果您尝试向原始数据添加截距，那么这可能就是您要寻找的

intercept = np.ones((3, 1))

raw_data = np.zeros((3, 3))

X = np.hstack([intercept, raw_data])