//
// Resampling will run resampling algorithm using values that has been defined
// in `Runtime` and return list of synthetic samples.
//
// The `dataset` must be samples of minority class not the whole dataset.
//
// Algorithms,
//
// (0) If oversampling percentage less than 100, then
// (0.1) replace the input dataset by selecting n random sample from dataset
// without replacement, where n is
//
// (percentage-oversampling / 100) * number-of-sample
//
// (1) For each `sample` in dataset,
// (1.1) find k-nearest-neighbors of `sample`,
// (1.2) generate synthetic sample in neighbors.
// (2) Write synthetic samples to file, only if `SyntheticFile` is not empty.
//
func (smote *Runtime) Resampling(dataset tabula.Rows) (e error) {
smote.Init()
if smote.PercentOver < 100 {
// (0.1)
smote.NSynthetic = (smote.PercentOver / 100.0) * len(dataset)
dataset, _, _, _ = dataset.RandomPick(smote.NSynthetic, false)
} else {
smote.NSynthetic = smote.PercentOver / 100.0
}
// (1)
for x := range dataset {
sample := dataset[x]
// (1.1)
neighbors := smote.FindNeighbors(&dataset, sample)
// (1.2)
smote.populate(sample, neighbors)
}
// (2)
if smote.SyntheticFile != "" {
e = resampling.WriteSynthetics(smote, smote.SyntheticFile)
}
return
}
//
// Write will write synthetic samples to file defined in `file`.
//
func (smote *Runtime) Write(file string) error {
return resampling.WriteSynthetics(smote, file)
}