CODEWITHSUNDEEP
python-3.xkeraspickle
Aakash aggarwal
Aakash aggarwal

Reputation: 443

AttributeError: 'Sequential' object has no attribute '_built'

I am doing transfer learning and I am using VGG16 model. I am fine tuning my model by feature extraction. Then I have trained my final model and I have pickled its weight. Here is my code

def prediction(array):
    size = 224
    array = image_array(filename , size)
    print(array.shape)
    array = np.array(array , dtype = np.float64)
    array = np.reshape(array, (1,224,224,3))
    print(array.shape)
    final_array = preprocess_input(array)
    vgg16 = VGG16(weights='imagenet', include_top=False)
    features = vgg16.predict(final_array)
    image = features.reshape(features.shape[0] , -1)
    #return image
    loaded_model = pickle.load(open('vgg16.sav', 'rb'))
    #print(image.shape)
    array = np.asarray(array)
    y_predict = loaded_model.predict(array)

when i call this function I am getting error in line y_predict = loaded_model.predict(array) I am getting

AttributeError: 'Sequential' object has no attribute '_built'

Upvotes: 1

Views: 3662

Answers (1)

nuric
nuric

Reputation: 11225

You shouldn't use picke.dump to save the weights and load as a model. Instead use the provided functions model.save(filename) or model.save_weights(filename) to save the model or just the weights respectively. In your case you can do:

vgg16.save('vgg16.h5')
# ...
loaded_model = keras.models.load_model('vgg16.h5')

You'll need h5py package to use these functions.

Upvotes: 3

Related Questions