This message was deleted.

Do you have existing code?

No sir, I’m currently evaluating softwares and I think BentoML will be a good fit. Example:

from sklearn.preprocessing import OrdinalEncoder

enc = OrdinalEncoder()

enc.fit()

enc.transform()

I can pickle.dump this object but I was wondering what would be the best approach to load it in a bentoml service such as a runner etc

ya that could work too

Will i just load it globally and call it inside the service predict method?

ya that could work. You can initialize the encoder within

def __init__(self):
  self.enc = OrdinalEncoder()

Sweet thank you

Thank you, I appreciate the swift response.

Extra 2 cents, for inference, you might not want to initialize a new

OrdinalEncoder

object each time and fit + transform, but rather use the one you created during training so that you don’t run into new values. So it might be suitable to use

custom_objects

when you are saving the model to save the encoder at training time, then load it when you are creating BentoML services.