"NaN" result when running multi class classification - tensorflow

Question

when i run these lines of code for binary classification it is running well without any problem and get a good result, but when i try to make it for many classes e.g 3 classes it give "NaN" in predict result
# Importing the Keras libraries and packages
from keras.models import Sequential
from keras.layers import Conv2D
from keras.layers import MaxPooling2D
from keras.layers import Flatten
from keras.layers import Dense
# Initialising the CNN
classifier = Sequential()
# Step 1 - Convolution
classifier.add(Conv2D(32, (3, 3), input_shape = (64, 64, 3), activation = 'relu'))
# Step 2 - Pooling
classifier.add(MaxPooling2D(pool_size = (2, 2)))
# Adding a second convolutional layer
classifier.add(Conv2D(32, (3, 3), activation = 'relu'))
classifier.add(MaxPooling2D(pool_size = (2, 2)))
classifier.add(Conv2enter code hereD(32, (3, 3), activation = 'relu'))
classifier.add(MaxPooling2D(pool_size = (2, 2)))
classifier.add(Conv2D(32, (3, 3), activation = 'relu'))
classifier.add(MaxPooling2D(pool_size = (2, 2)))
# Step 3 - Flattening
classifier.add(Flatten())
# Step 4 - Full connection
classifier.add(Dense(units = 128, activation = 'relu'))
classifier.add(Dense(units = 3, activation = 'sigmoid'))
# Compiling the CNN
classifier.compile(optimizer = 'adam', loss = 'categorical_crossentropy', metrics = ['accuracy'])
# Part 2 - Fitting the CNN to the images
from keras.preprocessing.image import ImageDataGenerator
train_datagen = ImageDataGenerator(rescale = 1./255,
shear_range = 0.2,
zoom_range = 0.2,
horizontal_flip = True)
test_datagen = ImageDataGenerator(rescale = 1./255)
training_set = train_datagen.flow_from_directory('data/train',
target_size = (64, 64),
batch_size = 32,
class_mode = 'categorical')
test_set = test_datagen.flow_from_directory('data/test',
target_size = (64, 64),
batch_size = 32,
class_mode = 'categorical')
classifier.fit_generator(training_set,
steps_per_epoch = 240 ,
epochs = 25,
validation_data = test_set,
validation_steps = 30)
import numpy as np
from keras.preprocessing import image
test_image = image.load_img('2.jpeg', target_size = (64, 64))
test_image = image.img_to_array(test_image)
test_image = np.expand_dims(test_image, axis = 0)
result = classifier.predict(test_image)
training_set.class_indices
i tried these lines of code with loss function "binary" with 2 classes it worked well without any problems and get a good result that helped me with my work and the accuracy approximately '93%' .
but my project based on multi class classification, so i tried to change the loss function to 'categorical_crossentropy' and the class mod in fit_generator to 'categorical' to make it multi class, the accuracy start with 60% and grows up to 99 and suddenly drop down to 33%.
the expected result the labels of the classes
the actual result is "NaN".
thanks in advance.

Answer 1

For multi-class classification, usually softmax is applied on the last dense layer instead of sigmoid. Change it to softmax to see whether the issue is still there.