Under the hood, deweather uses the parsnip
package. This package harmonises many different modelling engines,
meaning deweather can make it very easy to switch between
engines with a consistent API.
However, for users familiar with the engine packages themselves
(i.e., xgboost, ligthgbm, and
ranger), it can make it challenging to understand which
hyperparameter you’re actually tweaking when setting trees
or mtry or any of the other arguments in
build_dw_model().
The table in this article can be considered a ‘cheat sheet’ so you
can understand how different deweather arguments map onto
their engine equivalents.
| deweather/parsnip | xgboost | lightgbm | ranger |
|---|---|---|---|
| tree_depth | max_depth | max_depth | - |
| trees | nrounds | num_iterations | num.trees |
| learn_rate | eta | learning_rate | - |
| mtry | colsample_bynode | feature_fraction_bynode | mtry |
| min_n | min_child_weight | min_data_in_leaf | min.node.size |
| loss_reduction | gamma | min_gain_to_split | - |
| sample_size | subsample | - | - |
| stop_iter | early_stop | - | - |