ValueError：形状（50,50）和（3,1）不对齐：50（dim 1）！= 3（dim 0）

Question

I'm trying to create my second machinelearning, but i'm currently stock in a really annoying problem. I keep getting this error:

ValueError: shapes (50,50) and (3,1) not aligned: 50 (dim 1) != 3 (dim 0)

How can i fix the error?

My neural network

class neural_net:

def __init__(self):
    self.weight = 2 * np.random.random((3, 1)) - 1

def sigmoid(self, x):
    return 1/(1 + np.exp(-x))

def sigmoid_derivative(self, x):
    return x * (1 - x)

def train(self, training_input, training_output, iteration):
    for _ in range(iteration):
        output = self.think(training_input)
        error = training_output - output
        adjust = np.dot(training_input, error * self.sigmoid_derivative(output))
        self.weight += adjust

def think(self, input):
    input = input.astype(float)
    output = self.sigmoid(np.dot(input, self.weight))
    return output

X = train_data[len(train_data) - 1][0] -> [152 147 146 143 150 151 151 150 147 149 156 143 121  69  58 105  70  69

66 145 107 144 154 145 154 160 154 158 159 171 167 164 167 160 123 152 152 144 148 116 148 155 165 162 147 130 137 129 106 93] Y = train_data[len(train_data) - 1][1]

nn = neural_net()
nn.train(X, Y, 3)

Answer 1

To multiply two matrices the number of columns of the first matrix must be equal to number of rows of the second matrix. In your case columns of X should be equal to rows of self.weights . But the number of columns of X is 50 and the number of rows of self.weights is 3.

While defining weights for your neural network you should always consider the channels of the inputs and outupts.

If inputs have channels of n and outputs have channels of m the shape of the weights should be (n, m).

Assuming from the way you have accessed X ( train_data[len(train_data) - 1][0] ), X is a single sample of shape (50,50). If this is correct then X has to be converted into vector before feeding into neural network(Assuming your network's input layer is fully connected to hidden layer). The same goes to Y.

X = X.reshape(1, -1) # X.shape == (1, 2500) 
Y = Y.reshape(1, -1) # Y.shape == (1, 2)

Now the weight's shape should be (2500, 2)

self.weight = 2 * np.random.random((2500, 2)) - 1