如何在 CNN 中找到模型的准确性？

Question

I want to find the accuracy of a customised CNN model. I have weights(w), loss value(l) and test data(x_test) with the class variable(y_test). The weights can't be adjusted, they should remain the same. It will be like a single layer feedforward neural network.

I want to complete this function to give the same accuracy as in Keras.

Edit 1: Binary class classification problem.

  def accuracy(x_test,y_test,w,l):
     y_pred=numpy.dot(x_test,w)
     acc=...
     return acc

How to complete the accuracy statement as they do in Keras or any other API?.

Answer 1

If you want something similar to keras , you just care about y_test and y_pred ( y_test is y_true ):

def acc(y_true, y_pred):
    return np.equal(np.argmax(y_true, axis=-1), np.argmax(y_pred, axis=-1)).mean()

This is my POC:

import numpy as np

# y_test onehot encoded
y_test = np.array([[1, 0, 0],[0, 1, 0],[0, 0, 1], [0, 1, 0], [1, 0, 0]])
y_pred = np.random.random((5,3))

print("y_true: " + str(np.argmax(y_test, axis=-1)))
print("y_pred: " + str(np.argmax(y_pred, axis=-1)))

def acc(y_true, y_pred):
    return np.equal(np.argmax(y_true, axis=-1), np.argmax(y_pred, axis=-1)).mean()

print("accuracy: " + str(acc(y_test, y_pred)))

Result:

y_real: [0 1 2 1 0]
y_pred: [1 1 0 1 0]
accuracy: 0.6

Update 1: Since it is for binary classification the function will be this:

def acc(y_true, y_pred):
    return np.equal(y_true, np.round(y_pred)).mean()

POC:

import numpy as np

y_test = np.array([1, 0, 0, 1, 0])
y_pred = np.random.random((5))

print("y_true: " + str(y_test))
print("y_pred: " + str(np.round(y_pred).astype(int)))

def acc(y_true, y_pred):
    return np.equal(y_true, np.round(y_pred)).mean()

print("accuracy: " + str(acc(y_test, y_pred)))

Answer 2

If I'm not wrong you can pass your actual and predicted values to scoring function available inside mlxtend library.

Look Scoring - mlxtend for more explanation.