Calculate raw model predictions on float features and string categorical feature values
@param calcer model handle
@param docCount object count
@param floatFeatures array of array of float (first dimension is object index, second is feature index)
@param floatFeaturesSize float feature count
@param catFeatures array of array of char* categorical value pointers.
String pointer should point to zero terminated string.
@param catFeaturesSize categorical feature count
@param result pointer to user allocated results vector
@param resultSize result size should be equal to modelApproxDimension * docCount
(e.g. for non multiclass models should be equal to docCount)
@return false if error occured
Use this method only if you really understand what you want.
Calculate raw model predictions on flat feature vectors
Flat here means that float features and categorical feature are in the same float array.
@param calcer model handle
@param docCount number of objects
@param floatFeatures array of array of float (first dimension is object index, second if feature index)
@param floatFeaturesSize float values array size
@param result pointer to user allocated results vector
@param resultSize Result size should be equal to modelApproxDimension * docCount
(e.g. for non multiclass models should be equal to docCount)
@return false if error occured
Calculate raw model prediction on float features and string categorical feature values for single object
@param calcer model handle
@param floatFeatures array of float features
@param floatFeaturesSize float feature count
@param catFeatures array of char* categorical feature value pointers.
Each string pointer should point to zero terminated string.
@param catFeaturesSize categorical feature count
@param result pointer to user allocated results vector (or single double)
@param resultSize result size should be equal to modelApproxDimension
(e.g. for non multiclass models should be equal to 1)
@return false if error occured
Calculate raw model predictions on float features and string categorical feature values
@param calcer model handle
@param docCount object count
@param floatFeatures array of array of float (first dimension is object index, second is feature index)
@param floatFeaturesSize float feature count
@param catFeatures array of array of char* categorical value pointers.
String pointer should point to zero terminated string.
@param catFeaturesSize categorical feature count
@param textFeatures array of array of char* text value pointers.
String pointer should point to zero terminated string.
@param textFeaturesSize text feature count
@param result pointer to user allocated results vector
@param resultSize result size should be equal to modelApproxDimension * docCount
(e.g. for non multiclass models should be equal to docCount)
@return false if error occured
Calculate raw model predictions on float features and hashed categorical feature values
@param calcer model handle
@param docCount object count
@param floatFeatures array of array of float (first dimension is object index, second if feature index)
@param floatFeaturesSize float feature count
@param catFeatures array of array of integers - hashed categorical feature values.
@param catFeaturesSize categorical feature count
@param result pointer to user allocated results vector
@param resultSize result size should be equal to modelApproxDimension * docCount
(e.g. for non multiclass models should be equal to docCount)
@return false if error occured
Check if model metadata holds some value for provided key
@param calcer model handle
Create empty data wrapper
@return
Use CUDA gpu device for model evaluation
Get expected categorical feature count for model
@param calcer model handle
Get number of dimensions in model
@param calcer model handle
If error occured will return stored exception message.
If no error occured, will return invalid pointer
@return
Get expected float feature count for model
@param calcer model handle
Special case for hash calculation - integer hash.
Internally we cast value to string and then calulcate string hash function.
Used in ClickHouse for catboost model evaluation on integer cat features.
@param val integer cat feature value
@return hash value
Get model metainfo for some key. Returns const char* pointer to inner string. If key is missing in model metainfo storage this method will return nullptr
@param calcer model handle
Get model metainfo value size for some key. Returns 0 both if key is missing in model metadata and if it is really missing
@param calcer model handle
Get number of dimensions for current prediction
For default APT_RAW_FORMULA_VAL
, APT_EXPONENT
, APT_PROBABILITY
, APT_CLASS
prediction type GetPredictionDimensionsCount == GetDimensionsCount
For APT_RMSE_WITH_UNCERTAINTY
- returns 2 (value prediction and predicted uncertainty)
@param calcer model handle
Get hash for given string value
@param data we don’t expect data to be zero terminated, so pass correct size
@param size string length
@return hash value
Get number of trees in model
@param calcer model handle
Load model from memory buffer into given model handle
@param calcer
@param binaryBuffer pointer to a memory buffer where model file is mapped
@param binaryBufferSize size of the buffer in bytes
@return false if error occured
Load model from file into given model handle
@param calcer
@param filename
@return false if error occured
Create empty model handle
@return
Delete model handle
@param calcer
Methods equivalent to the methods above
only returning a prediction for the specific class
@param classId number of the class should be in [0, modelApproxDimension - 1]
@param resultSize result size should be equal to docCount
Set prediction type for model evaluation
Set prediction type for model evaluation with string constant