Consider a dataset where there is 10-to-1 ratio between class A and B in the dependent varaible. Assuming we wish to maintain a 1-to-1 ratio for each model (this can modified using the `ratio` parameter), each observation in larger class (say A here) will be randomly assigned a number between 1 and 10. In this example, 10 models will be estimated. Observations from the smaller class will be used in every model but observations in the larger class will be used in only one of the models. This is potentially advantageous since all data is used in the model.
Usage
downsample(formu, data, model_fun, ratio = 1, show_progress = TRUE, ...)
# S3 method for class 'downsample'
predict(object, newdata, ...)Arguments
- formu
an object of class "formula" (or one that can be coerced to that class): a symbolic description of the model to be fitted.
- data
data frame with the data to estimate the model from.
- model_fun
modeling function (e.g. `glm`).
- ratio
the ratio of small class to the larger class when downsampling.
- show_progress
if TRUE a progress bar will be displayed showing the status of the model estimations.
- ...
other parameters passed to `model_fun`.
- object
result of [downsample()].
- newdata
An optional data frame in which to look for variables with which to predict. If omitted, the fitted values are used.
Details
It should be noted that this function is simply a wrapper to any prediction function as long as the first parameter is model formula and the second parameter is a data frame. The `downsample` function will pass any parameters to the `model_fun` through the `...`.
The `predict` function will return a data frame where each row corresponds to the row in the training data frame or the `newdata` data frame and the columns are the predicted values from each of the trained models. Note that the `...` parameter will be passed to the appropriate `predict` function. For example, if the models were trained using `glm` for a logistic regression, the passing `type = 'response'` will provided the predicted probabilities.
Examples
data("pisa", package = "medley")
train_rows <- sample(nrow(pisa) * .7)
pisa_train <- pisa[train_rows,]
pisa_valid <- pisa[-train_rows,]
pisa_ds_out <- downsample(
formu = Public ~ .,
data = pisa_train,
model_fun = glm,
ratio = 2,
family = binomial(link = 'logit')
)
#>
|
| | 0%
|
|============== | 20%
|
|============================ | 40%
|
|========================================== | 60%
|
|======================================================== | 80%
|
|======================================================================| 100%
pisa_predictions_ds <- predict(pisa_ds_out, newdata = pisa_valid, type = 'response')
pisa_predictions_ds2 <- pisa_predictions_ds |> apply(1, mean)