How to access class probabilities in keras?

Question

I am training a model for which I need to report class probabilities instead of a single classification. I have three classes and each training instance has either of the three classes assigned to it.

I am trying to use Keras to create an MLP. But I can't figure how to extract the final class probabilities for each class. I am using this as my base example: http://machinelearningmastery.com/regression-tutorial-keras-deep-learning-library-python/

Thanks !

Answer 1

In order to perform multi-class classification (nb_classes > 1) you have to prepare your model in a specific manner.

1. Make sure your labels are well-designed for multi-class classification. Have a look at the numpy_utils
2. You have to use the categorical_crossentropy as objective function for multi-class classification (see Keras objectives)
3. Your last layer must have the softmax activation function (guarantees the output to be between 0 and 1) and nb_classes neurons.
4. Train your model as usual
5. Use the predict function. You will receive a vector of size (nb_classes,1) containing the probabilities of each class.

Answer 2

You could use the predict method of your trained model

predict

predict(self, x, batch_size=32, verbose=0)

Generates output predictions for the input samples, processing the samples in a batched way.

Arguments

x: the input data, as a Numpy array (or list of Numpy arrays if the model has multiple outputs). batch_size: integer. verbose: verbosity mode, 0 or 1.

Returns a Numpy array of predictions.

model.predict(input_to_your_network)