Specifying Custom Missing Value Encodings

There may be cases in which you have missing values in your data, but you encode these values in a special way (other than the standard NaN).

In such cases, Fiddler offers a way to specify your own missing value encodings for each column.


You can create a "fall back" dictionary, which holds the values you would like to have treated as missing for each column. Then just pass that dictionary into your fdl.ModelInfo object before registering your model.

fall_back = {
  'column_1': [-999, 'missing'],
  'column_2': [-1, '?', 'na']
}

model_info = fdl.ModelInfo.from_dataset_info(
  ...
  fall_back=fall_back
)

Did this page help you?