Custom quantization borders and NaN modes


Custom quantization borders and the method for processing NaN values for the dataset features.

  • Each line contains information regarding a single border and optionally the NaN mode settings for the corresponding feature.
  • Different NaN mode settings can not be specified for a single feature. The value either has to be set only on one line or should be the same on different lines that contain information regarding a single feature.
  • Supported NaN modes:
    • Forbidden — NaN values are not supported, their presence raises an exception.
    • Min — Each NaN float feature is processed as the minimum value from the dataset.
    • Max — Each NaN float feature is processed as the maximum value from the dataset.
  • The NaN mode for the feature is defined by the value of the --nan-mode (nan_mode) training parameter if not specified in this file.
  • Format of a single line:
    <zero-based feature ID><\t><border value><\t><nan mode (optional)>
The following description contains two borders for features indexed 0 and 2 and NaN mode settings for each of these features: