About tuning my own custom model

User 2013 | 8/19/2015, 5:54:32 PM

I am trying to understand how to tune the models with graphlab create. The documentation of https://dato.com/learn/userguide/modelparametersearch/models.html

Tuning your own custom model

is a very explanation for this. But I am confused for the explanation somehow. - In the ensemble() function, there is a score() function, and parameter for this function is test. Is this score not for training data-set? When the score() function will be called? Is there some template for customized-model function ? e.g. must have a score function, etc. - For the custom_evaluator() function, there is a function parameter as scorer. But also from the following examples, I can not find the implementation for it. Is that inside graph lab?

Thanks, Cong


User 19 | 8/21/2015, 8:03:22 PM

Hi Cong,

I agree that the User Guide is a bit unclear here. The ensemble() function must return a function that can make scores on a particular dataset. This returned value is the one that gets passed to customevaluator. Notice in the customevaluator() that this scorer function is used to make scores on the training data, then on the testing data; these scores are then returned by the custom_evaluator function, and this is done for each unique set of parameters.

Hope that helps, Chris