The actual tuning
The best hyperparameters make the best model for your data. Once you decided on a tuning grid, you need to train and evaluate models on every grid point to see which grid point gives the best model performance.
This can take some time, given that using k-fold cross-validation, an ensemble size of n
trees, and a tuning grid of t
combinations makes k * n * t
models to be trained in total.
It's your turn to perform the actual tuning! Pre-loaded are customers_train
and the results of the last exercise, boost_spec
and tunegrid_boost
:
# A tibble: 27 x 3
tree_depth learn_rate sample_size
<int> <dbl> <dbl>
1 1 0.0000000001 0.1
2 8 0.0000000001 0.1
3 15 0.0000000001 0.1
4 1 0.00000316 0.1
...
This exercise is part of the course
Machine Learning with Tree-Based Models in R
Exercise instructions
- Create six folds of the training data using
vfold_cv()
and save them asfolds
. - Use
tune_grid()
to tuneboost_spec
using your folds, your tuning grid, and theroc_auc
metric. Save the results astune_results
. - Plot the results to visualize the result of the tuning process.
Hands-on interactive exercise
Have a go at this exercise by completing this sample code.
# Create CV folds of training data
folds <- ___
# Tune along the grid
tune_results <- ___(___,
still_customer ~ .,
resamples = ___,
grid = ___,
metrics = metric_set(___))
# Plot the results
___(___)