python - Returning probabilities in a classification prediction in Keras?

Question

I am trying to make a simple proof-of-concept where I can see the probabilities of different classes for a given prediction.

However, everything I try seems to only output the predicted class, even though I am using a softmax activation. I am new to machine learning, so I'm not sure if I am making a simple mistake or if this is a feature not available in Keras.

I'm using Keras + TensorFlow. I have adapted one of the basic examples given by Keras for classifying the MNIST dataset.

My code below is exactly the same as the example, except for a few (commented) extra lines that exports the model to a local file.

'''Trains a simple deep NN on the MNIST dataset.
Gets to 98.40% test accuracy after 20 epochs
(there is *a lot* of margin for parameter tuning).
2 seconds per epoch on a K520 GPU.
'''

from __future__ import print_function

import keras
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout
from keras.optimizers import RMSprop

import h5py # added import because it is required for model.save
model_filepath = 'test_model.h5' # added filepath config

batch_size = 128
num_classes = 10
epochs = 20

# the data, shuffled and split between train and test sets
(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
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')

# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)

model = Sequential()
model.add(Dense(512, activation='relu', input_shape=(784,)))
model.add(Dropout(0.2))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.2))
model.add(Dense(num_classes, activation='softmax'))

model.summary()

model.compile(loss='categorical_crossentropy',
          optimizer=RMSprop(),
          metrics=['accuracy'])

history = model.fit(x_train, y_train,
                batch_size=batch_size,
                epochs=epochs,
                verbose=1,
                validation_data=(x_test, y_test))
score = model.evaluate(x_test, y_test, verbose=0)
print('Test loss:', score[0])
print('Test accuracy:', score[1])

model.save(model_filepath) # added saving model
print('Model saved') # added log

Then the second part of this is a simple script that should import the model, predict the class for some given data, and print out the probabilities for each class. (I am using the same mnist class included with the Keras codebase to make an example as simple as possible).

import keras
from keras.datasets import mnist
from keras.models import Sequential
import keras.backend as K

import numpy

# loading model saved locally in test_model.h5
model_filepath = 'test_model.h5'
prev_model = keras.models.load_model(model_filepath)

# these lines are copied from the example for loading MNIST data
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = x_train.reshape(60000, 784)

# for this example, I am only taking the first 10 images
x_slice = x_train[slice(1, 11, 1)]

# making the prediction
prediction = prev_model.predict(x_slice)

# logging each on a separate line
for single_prediction in prediction:
    print(single_prediction)

If I run the first script to export the model, then the second script to classify some examples, I get the following output:

[ 1.  0.  0.  0.  0.  0.  0.  0.  0.  0.]
[ 0.  0.  0.  0.  1.  0.  0.  0.  0.  0.]
[ 0.  1.  0.  0.  0.  0.  0.  0.  0.  0.]
[ 0.  0.  0.  0.  0.  0.  0.  0.  0.  1.]
[ 0.  0.  1.  0.  0.  0.  0.  0.  0.  0.]
[ 0.  1.  0.  0.  0.  0.  0.  0.  0.  0.]
[ 0.  0.  0.  1.  0.  0.  0.  0.  0.  0.]
[ 0.  1.  0.  0.  0.  0.  0.  0.  0.  0.]
[ 0.  0.  0.  0.  1.  0.  0.  0.  0.  0.]
[ 0.  0.  0.  1.  0.  0.  0.  0.  0.  0.]

This is great for seeing which class each is predicted to be, but what if I want to see the relative probabilities of each class for each example? I am looking for something more like this:

[ 0.94 0.01 0.02 0. 0. 0.01 0. 0.01 0.01 0.]
[ 0. 0. 0. 0. 0.51 0. 0. 0. 0.49 0.]
...

In other words, I need to know how sure each prediction is, not just the prediction itself. I thought seeing the relative probabilities was a part of using a softmax activation in the model, but I can't seem to find anything in the Keras documentation that would give me probabilities instead of the predicted answer. Am I making some kind of silly mistake, or is this feature not available?

Answer 1

So it turns out that the problem was I was not fully normalizing the data in the prediction script.

My prediction script should have had the following lines:

# these lines are copied from the example for loading MNIST data
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = x_train.reshape(60000, 784)
x_train = x_train.astype('float32') # this line was missing
x_train /= 255 # this line was missing too

Because the data was not cast to float, and divided by 255 (so it would be between 0 and 1), it was just showing up as 1s and 0s.