How use Talos with custom model declaration f.ex. model=Model() and not model = Sequential() about talos HOT 7 CLOSED

Could you install from master and see what happens (at least you should get a different error message then):

pip install git+https://github.com/autonomio/talos.git

Also can you post your model code for reference.

from talos.

@Jakedismo just to clarify, this is a known issue and is addressed in the version of Talos that @mikkokotila posted. Basically there was a catch-all statement that spit out the error message you're seeing regardless of the error.

If what you posted above is the model function you're trying to import to Talos, it won't work because the arguments are not in the right format.

Also, in a more recent version of Talos (on the Master branch) you can manually specify the validation and training sets independently if you want.

Does this help at all?

from talos.

def DAE_CF(x_train, y_train, x_val, y_val, params):
    #model = Model()
    user_input = Input((train_x.shape[1],), name='user_input')
    #model.add(user_input)
    h_user = Dropout(1)(user_input)
    #model.add(h_user)
    #h_item = Embedding(input_dim=I, output_dim=K, input_length=1, W_regularizer=l2(l))(item_input)
    encoded = Dense(50, W_regularizer=l2(params['lr']), b_regularizer=l2(params['lr']))(h_user)
    #model.add(encoded)
    
    item_input = Input((1,), dtype='int32', name='item_input')
    #model.add(item_input)
    h_item = Embedding(input_dim=len(y_train), output_dim=50, input_length=1, W_regularizer=l2(params['lr']))(item_input)
    #model.add(h_item)
    decoded = Flatten()(h_item)
    #model.add(decoded)

    h = merge([encoded, decoded], mode='sum')
    if hidden_activation:
        h = Activation(params['hidden_activation'])(h)
    y = Dense(train_x.shape[1], activation=params['output_activation'])(h)
    #model.add(y)
    model = Model(input=[user_input, item_input], output=y) 
    #model.add(autoencoder)
    
    encoder = Model(user_input, encoded)
    encoded_input = Input((50,), name='encoded_input')
    decoder_layer = autoencoder.layers[-1]
    decoder = Model(encoded_input, decoder_layer(encoded_input))

    model.compile(loss=params['loss'], optimizer=params['optimizer'], metrics=[metrics.rmse, 'accuracy'])
    history = autoencoder.fit(x=x_train, y=y_train,
                       batch_size=batch_size, epochs=n_epochs, verbose=1,
                       validation_data=[x_val, y_val])
    return history, model

from talos.

Same problem here. I have not a Sequential model, but a Model(), a Model has not got a predict_classes(). I;m using Keras 2.2.0

File "/usr/local/lib/python3.5/dist-packages/talos/scan/Scan.py", line 142, in init
self._null = self.runtime()
File "/usr/local/lib/python3.5/dist-packages/talos/scan/Scan.py", line 147, in runtime
self = scan_run(self)
File "/usr/local/lib/python3.5/dist-packages/talos/scan/scan_run.py", line 29, in scan_run
self = rounds_run(self)
File "/usr/local/lib/python3.5/dist-packages/talos/scan/scan_run.py", line 61, in rounds_run
self._val_score = get_score(self)
File "/usr/local/lib/python3.5/dist-packages/talos/metrics/score_model.py", line 17, in get_score
y_pred = self.keras_model.predict_classes(self.x_val)
AttributeError: 'Model' object has no attribute 'predict_classes'

from talos.

@timvhamme, I found a solution for this problem: modify the score_model.py file as follows

from .performance import Performance

def get_score(self):

#y_pred = self.keras_model.predict_classes(self.x_val)  # not supported by Model()
y_prob = self.keras_model.predict(self.x_val)    # supported by Model()
y_pred = y_prob.argmax(axis=-1)  # return the class
return Performance(y_pred, self.y_val, self.shape, self.y_max).result

from talos.

This is related with #39, #42 and #64. It looks like in #64 we've just agreed to add this fix as a parameter into Talos. Should have it pushed to dev over the next few days. The way it will work is to to use:

experimental_functional_support=True

...in Scan()

from talos.

This is now fixed in the current dev branch.

from talos.