How can I implement confidence level in a CNN with tensorflow?

Question

My CNN outputs an array of values that I have to check for the biggest one and take it as the predicted class. Example:

-148.7290802 , -133.90687561,  -90.850914  , -135.78356934,
    -128.6325531 , -125.76812744,  -85.41909027,  -72.3269577 ,
    -103.51300812

For class index 6.

Now, how can I get the confidence of that result?

My setup is:

predict_op = [tf.argmax(py_x,1), py_x]
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(py_x, Y))
train_op = tf.train.RMSPropOptimizer(learningRate, decayRate).minimize(cost) 

Updated code now returning: [[ 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0.]]

predict_op = tf.nn.softmax(py_x)
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(py_x, Y))
train_op = tf.train.RMSPropOptimizer(learningRate, decayRate).minimize(cost) 

Answer 1

Apply softmax in the last stage; this will yield posterior probabilities at the final stage. You're already using softmax in the set-up; just use it on the final vector to convert it to RMS probabilities. The confidence of that prediction is simply the probability of the top item.

For a quick illustration, see the Wikipedia page under Generalization and Statistics. This section also describes the confidence of the model overall.