Hello AlexT,
Thanks for your asnwer.
As I mentioned at the begging of my post, I used Dataiku interface for deep learning.
Which means I build a model using TensorFlow functional API.
Dataiku documentation doesn't tell you to implement predict_proba() method when you're using that, as you do when you create a class deriving from sklearn BaseEstimator.
I don't think that's the case with TensorFlow, right?
I did a test, I used default code that Dataiku puts in the "Architecture" tab of deep learning GUI, which is:
def build_model(input_shapes, n_classes=None):
# This input will receive all the preprocessed features
# sent to 'main'
input_main = Input(shape=input_shapes["main"], name="main")
x = Dense(64, activation='relu')(input_main)
x = Dense(64, activation='relu')(x)
predictions = Dense(n_classes, activation='softmax')(x)
# The 'inputs' parameter of your model must contain the
# full list of inputs used in the architecture
model = Model(inputs=[input_main], outputs=predictions)
return model
As you can see, final layer uses softmax, which returns probabilities per class.
Still, the only thing I can see in the GUI is "No model-specific details available".
I also checked the logs and there is no "Cannot draw background rows and compute column importance: model is not probabilistic" there.
