object of type 'NoneType' has no len()

First time try with Hyperas. I get an error: object of type ‘NoneType’ has no len() at trials=Trials() in the optim.minimize(...) call. Not sure what this means. I am trying to optimize a network that takes three inputs in the form of a list X_train[], so X_train[0] has a set of input data, X_train[1] and X_train_[2] also. Not sure if that could be the cause for this error? Don’t know what Hyperas can handle.

Here’s some code:

def data(): 
    # load VGG16 output data, X_train and X_valid are lists with data channels
    X_train, Y_train, X_valid, Y_valid = load_VGG16_output(data_type='train', num_channels=3)
    return X_train, Y_train, X_valid, Y_valid

def model(X_train, Y_train, X_valid, Y_valid):

    in_0 = Input(shape=X_train[0].shape[1:])    
    x = Dense(256, activation='relu')(in_0)
    x = Dropout(0.5)(x)

    in_1 = Input(shape=X_train[1].shape[1:])
    y = Dense(64, activation='relu')(in_1) #, W_regularizer=l2(0.01))(in_1)
    y = Dropout(0.5)(y)

    in_2 = Input(X_train[2].shape[1:])
    z = Dense(32, activation='relu')(in_2) #, W_regularizer=l2(0.01))(in_2)
    z = Dropout(0.5)(z)

    x = merge([x, y, z], mode='concat', concat_axis=1)
    
    predict = Dense(1, activation='sigmoid')(x)
    model = Model(input=[in_0, in_1, in_2], output=predict)
   
    optimizer=Adam(lr=1e-4, decay=0.003)

    model.compile(loss='binary_crossentropy', optimizer=optimizer, metrics=['accuracy'])

    # load the class weight to balance the input positive / negative scores
    class_weight = load_class_weight()

    model.fit(X_train, Y_train,
              batch_size={{choice([32, 64, 128])}}, # optimize batch size
              nb_epoch=1, 
              verbose=2, 
              class_weight=class_weight,
              validation_data=(X_valid, Y_valid))

    score, acc = model.evaluate(X_valid, Y_valid, verbose=0)
    print('Valid accuracy: ', acc)
    return {'loss': -acc, 'status': STATUS_OK, 'model': model}

if __name__ == '__main__':
    
    X_train, Y_train, X_valid, Y_valid = data()

    best_run, best_model = optim.minimize(model=model,
                                          data=data,
                                          algo=tpe.suggest,
                                          max_evals=5,
                                          trials=Trials())
       
    print('Evalutation of best performing model: ')
    print(best_model.evaluate(X_valid, Y_valid))

[ileadall42]

I got it! There is an indent after the def data(): funtion as follow

[maxpumperla]

Sorry, I can’t reproduce this, but I think this has been resolved. @jdelange please re-open if the problem still persists.