KerasClassifier - Sci-Kit learn pipeline [Attribute Error]

Question

I am using Sci-kit learn's pipeline with a KerasClassfier at the end. The classifier would load in the trained model for prediction. But after adding the classifier to the pipeline (3 components in total), I received an error AttributeError: 'KerasClassifier' object has no attribute 'model' after calling pipeline.predict_proba. I think it expects me to fit it but I am importing a trained model. I can't find anything relevant online. Your kind help is greatly appreciated. The following is relevant part of my code:

def buildEngModelByLoading():
  # load json and create model
  json_file = open('saved_model/cnnModel.json', 'r')
  loaded_model_json = json_file.read()
  json_file.close()
  loaded_model = model_from_json(loaded_model_json)
  # load weights into new model
  loaded_model.load_weights("saved_model/cnnModel.h5")
  print("Loaded classifier model")
  return loaded_model

engSklearnCnn = KerasClassifier(build_fn=buildEngModelByLoading, epochs=20, batch_size=batchSize, verbose=1)
#Append classfier to one pipeline
pipeline.steps.insert(2,['classifier',engSklearnCnn])`

Mihai Alexandru-Ionut · Accepted Answer · 2018-04-02 09:20:38Z

1

That's because you forgot to use compile method first before predict function.

buildEngModelByLoading().compile(optimizer = 'classifier_optimizer', loss = 'loss_function', metrics = 'metrics')

Then just replace classifier_optimizer, loss_function, metrics with your parameters used.

edited Apr 2, 2018 at 9:20

answered Apr 2, 2018 at 8:24

Mihai Alexandru-Ionut

48.6k14 gold badges106 silver badges132 bronze badges

Sign up to request clarification or add additional context in comments.

6 Comments

soulless Over a year ago

Thanks for the help. I tried your suggestion and the error says KerasClassifier has no attribute compile. I guess the wrapper makes it no longer the native Keras model. So I tried calling compile within the build function on loaded_model after loading the weights, then it gives the same error of not having attribute 'model'

Mihai Alexandru-Ionut Over a year ago

@soulless, I updated my answer.You should use compile method for the imported model

soulless Over a year ago

Do u mean I should have it like this? Doesn't seem to to work engSklearnCnn = KerasClassifier(build_fn=buildEngModelByLoading().compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy']), epochs=20, batch_size=batchSize, verbose=1)

soulless Over a year ago

I am supposed to call predict with pipeline.predict_proba after wrapping the keras model with the KerasClassifier wrapper

Mihai Alexandru-Ionut Over a year ago

Yes, you have to use compile method before creating kerasclassifier

|

Collectives™ on Stack Overflow

KerasClassifier - Sci-Kit learn pipeline [Attribute Error]

1 Answer 1

6 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

6 Comments

Your Answer

Sign up or log in

Post as a guest

Related