Error while using the Precision metric for MNIST digit classification

Question

I am trying to achieve a high Precision score on the MNIST dataset using Tensorflow and Keras. My code is working if I set the metric to accuracy but when I set it to precision, it gives the following error:

ValueError: Shapes (32, 10) and (32, 1) are incompatible

Here is my code:

import tensorflow as tf 
import keras
from tensorflow.keras.datasets import mnist

def bulid_model(n = 1, neuron=30,lr = 3e-3,input_shape=(784,)):
    model = keras.models.Sequential()
    model.add(keras.layers.InputLayer(input_shape=input_shape))
    for layer in range(n):
        model.add(keras.layers.Dense(neuron, activation = 'relu'))
    model.add(keras.layers.Dense(10,activation='softmax'))
    optimizer = keras.optimizers.Adam(lr = lr)
    model.compile(loss = 'sparse_categorical_crossentropy',optimizer=optimizer,metrics = [keras.metrics.Precision()])
    return model

if __name__ == "__main__":
    (X_train,Y_train),(X_test,Y_test) = mnist.load_data()

    X_train = X_train.reshape(60000, 784)
    X_test = X_test.reshape(10000, 784)
    X_train = X_train.astype('float32')
    X_test = X_test.astype('float32')

    X_train /= 255
    X_test /= 255

    model = bulid_model(3,20,0.0156)

    history = model.fit(X_train,Y_train,epochs=50)

Can anyone help me with this?

Answer 1

Precision, is a metric for binary classification. It computes true_positives and false_positives then simply divides true_positives by the sum of true_positives and false_positives.

But Accuracy metric can be used for multi-class classification like MNIST, because it calculates how often predictions equal labels.