CODEWITHSUNDEEP
python-3.xkerasconv-neural-networkgrid-searchhyperparameters
Swe
Swe

Reputation: 11

Grid search with GridSearchCV - scikit-learn (Hyperparameter Tuning) using ImageDataGenerator (keras)?

How can I perform hyperparameter tuning when my image inputs are through ImageDataGenerator? My training and test data are not in the form of arrays (X_train, Y_train etc). I want to tune my hyperparameters using GridSearchCV from sklearn and ImageDataGenerator from keras.

These are a few snippets from the code I've attempted!

#(5) Train
train_datagen = ImageDataGenerator(rescale=1./255)

validation_datagen = ImageDataGenerator(rescale=1./255)

train_batchsize = 15
val_batchsize = 10

train_generator = train_datagen.flow_from_directory(
        train_dir,
        batch_size=train_batchsize,
        class_mode='categorical',
        shuffle=False)

validation_generator = validation_datagen.flow_from_directory(
        validation_dir,
        target_size=(image_size, image_size1),
        batch_size=val_batchsize,
        class_mode='categorical')

#Function for Creating Model
def create_model():
    .....................
    return model

model = KerasClassifier(build_fn=create_model, batch_size=1000, epochs=10, verbose = 1) 

# Use scikit-learn to grid search 
activation =  ['relu', 'tanh', 'sigmoid', 'hard_sigmoid', 'linear'] # softmax, softplus, softsign 
momentum = [0.0, 0.2, 0.4, 0.6, 0.8, 0.9]
neurons = [1, 5, 10, 15, 20, 25, 30]
init = ['uniform', 'lecun_uniform', 'normal', 'zero', 'glorot_normal', 'glorot_uniform', 'he_normal', 'he_uniform']
optimizer = [ 'SGD', 'RMSprop', 'Adagrad', 'Adadelta', 'Adam', 'Adamax', 'Nadam']

param_grid = dict(epochs=epochs, batch_size=batch_size)
##############################################################
grid = GridSearchCV(estimator=model, param_grid=param_grid, n_jobs=-1)
grid_result = grid.fit_generator(train_generator, validation_generator)

Getting an error in this line : grid_result = grid.fit_generator(train_generator, validation_generator)

Upvotes: 1

Views: 1272

Answers (1)

Jakob Lindskog
Jakob Lindskog

Reputation: 130

Sklearn GridSearchCV doesn't expose a fit_generator method. You're probably confusing it with Keras (now deprecated) fit_generator.

This means that it is non-trivial to gridsearch a Keras model if you get your training data from generators. I found two related questions on SO:

So for now, you have to resort to workarounds.

Upvotes: 0

Related Questions