The partial_fit() method isn't available with the GridSearchCV class.
Instead, we can use its fit() method with a model that has a partial_fit()
method (or a PartialFitPipeline object) to find the best hyperparameters
in our search space. Then, we can grab the best model from the grid search
(best_estimator_) and use partial_fit() on it. Try this with the
PassiveAggressiveClassifier class from the sklearn.linear_model
module and a PartialFitPipeline object. This online learning classifier is
passive when it makes a correct prediction, but aggressive in correcting itself when
it makes an incorrect prediction. Don't worry about selecting a custom threshold.
Be sure to follow these steps:
a) Run a grid search using the January 2018 data for the initial training.
b) Grab the tuned model with the best_estimator_ attribute.
c) Evaluate the best estimator with the February 2018 data.
d) Make updates with the February 2018 data.
e) Evaluate the final model on March through June 2018 data.