Tensorflow - I don't get the right shapes - `ValueError: Shapes (100, 10, 10) and (100, 10) are incompatible`

Question

I am working on the mnist classification code. Such errors continue to occur in the code 코드 below.

import tensorflow as tf

(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()
print(x_train.shape) # (60000, 28, 28)
print(y_train.shape)

import matplotlib.pyplot as plt

print("Y[0] : ",  y_train[0])
plt.imshow(x_train[0], cmap=plt.cm.gray_r, interpolation = "nearest")

x_train = x_train.reshape(-1,28*28)
x_test = x_train / 255.0
x_test / 255.0

y_train = tf.keras.utils.to_categorical(y_train, 10)
y_test = tf.keras.utils.to_categorical(y_test, 10)

This is the error occurrence code.

model = tf.keras.Sequential()
model.add(tf.keras.layers.Dense(units=10, input_dim=784, activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer=tf.optimizers.Adam(learning_rate=0.001), metrics=['accuracy'])
model.summary()

model.fit(x_train, y_train, batch_size=100, epochs=10, validation_data=(x_test, y_test))

ValueError: Shapes (100, 10, 10) and (100, 10) are incompatible

This is my error message.

Initially, a reshape error occurred, so x_trial.reshape (-1,28*28) was added to the code. Then, this error occurs. How should I change the shape?

Answer 1

When I ran the code you provided I do not get the mentioned error, but an error related to an differing number of samples for the x and y variables. This was caused by the line in your code where you are normalizing the data from x_train but assign the output to x_test, leaving you with 60000 observations in x_test but only 10000 in y_test. After changing this the code runs fine:

import tensorflow as tf
import matplotlib.pyplot as plt

(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()

x_train = x_train.reshape(-1, 28 * 28)
x_train = x_train / 255.0 # changed to save output to x_train instead of x_test
x_test = x_test.reshape(-1, 28 * 28) # reshaped x_test to get the correct dimensions
x_test = x_test / 255.0 # also apply normalization to the test data

y_train = tf.keras.utils.to_categorical(y_train, 10)
y_test = tf.keras.utils.to_categorical(y_test, 10)

model = tf.keras.Sequential()
model.add(tf.keras.layers.Dense(units=10, input_dim=784, activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer=tf.optimizers.Adam(learning_rate=0.001), metrics=['accuracy'])
model.summary()

model.fit(x_train, y_train, batch_size=100, epochs=10, validation_data=(x_test, y_test))

Answer 2

The error occurs because of the x_test shape. In your code, you set it actually to x_train. [x_test = x_train / 255.0] Furthermore, if you feed the data as a vector of 784 you also have to transform your test data. So change the line to x_test = (x_test / 255.0).reshape(-1,28*28).