feature_finder: Add features one-by-one to find good small sets of features
In jeroenvdhoven/datapiper: datapiper

Description Usage Arguments Details Value

Add features one-by-one to find good small sets of features

1 2	feature_finder(train, test, response, model, difference, n_features, verbose = F)

`train`	The training dataset
`test`	The testing dataset
`response`	The response column as a string
`model`	A list containing at least a training function `.train` and a `.predict` function, plus optional named parameters to the train function. The `.train` function has to take a `data` argument that stores the training data and a `...` argument for the parameters. The `.predict` function needs to take two arguments, where the first is the model and the second the new dataset. You can use `model_trainer` as a wrapper for this list. It will also test your inputs to a certain degree.
`difference`	A function that calculates the difference between labels and predictions, for example `function(x,y) abs(x-y)`
`n_features`	The number of features to try before stopping the algorithm. Can be passed as a integer between 1 and `ncol(train)` or a fraction, in which case the number of features to try will be `n_features * ncol(train)`. Defaults to `ncol(train)`, only taking numeric columns into account.
`verbose`	Flag indicating if intermediate updates should be printed

This function will select features one-by-one depending on the correlation of a feature with the residuals of the train predictions to slowly build up a bigger model.

A dataframe, containing the mean difference for train and test set, as well as the feature that was added at each step.

jeroenvdhoven/datapiper documentation built on July 14, 2019, 9:34 p.m.