Hooks are useful for defining additional checks that should be performed prior to and during training and prediction. For example, one might want to issue a warning if the user is predicting on rows that were used for training, or a sanity check might be present prior to training to ensure a dependent variable is present.
Add a hook to a tundraContainer.
run_hooks(hook_name) add_hook(hook_name, hook_function)
| hook_name | character. The hook to run. Must be one of the available hooks. |
|---|---|
| hook_function | function. The hook to execute. It will be provided
the |
The following hooks are available.
train_pre_mungeThis hook runs during a call to the
container's train method, just prior to invoking the
munge_procedure to clean up the dataset. It could be
useful for defining pre-conditions on the dataset to ensure
it can be munged successfully.
train_post_mungeThis hook runs during a call to the
container's train method, just after invoking the
munge_procedure to clean up the dataset. It could be
useful for defining post-conditions on the dataset to ensure
it was munged successfully.
train_finalizeThis hook runs just after the train
method calls the train_function. It could be used to
verify presence or validate properties of the trained model.
predict_pre_mungeThis hook runs during a call to the
container's predict method, just prior to invoking the
munge_procedure to clean up the dataset. It could be
useful for defining pre-conditions on the dataset to ensure
it can be munged successfully.
predict_post_mungeThis hook runs during a call to the
container's predict method, just after invoking the
munge_procedure to clean up the dataset. It could be
useful for defining post-conditions on the dataset to ensure
it was munged successfully.
Each hook will be provided the tundraContainer as input
(unless it has no arguments, in which case it will simply be called).