/// <summary>
/// Evaluates the model on the single input.
/// <seealso cref="EvaluateSingle(float[], string[])"/>
/// </summary>
/// <param name="floatFeatures">Array of float features</param>
/// <param name="catFeatures">Array of categorical features</param>
/// <returns>Array storing model prediction for the input</returns>
public double[] EvaluateSingle(float[] floatFeatures, string[] catFeatures)
{
uint resultSize = CatboostNativeInterface.GetDimensionsCount(ModelContainer.ModelHandler);
double[] results = new double[resultSize];
bool res = true;
unsafe
{
fixed(float *floatFeaturesPtr = floatFeatures)
{
res = CatboostNativeInterface.CalcModelPredictionSingle(
ModelContainer.ModelHandler,
floatFeatures, (uint)floatFeatures.Length,
catFeatures, (uint)catFeatures.Length,
results, resultSize
);
}
}
if (res)
{
return(results);
}
else
{
var msg = CatboostNativeInterface.GetErrorStringConst(ModelContainer.ModelHandler);
throw new CatBoostException(
"An error has occurred in the CalcModelPredictionSingle() method in catboostmodel library.\n" +
$"Returned error message: {msg}"
);
}
}
/// <summary>
/// Evaluates the model on the input batch.
/// <seealso cref="EvaluateSingle(float[], string[])"/>
/// </summary>
/// <param name="floatFeatures">
/// 2D array of float features.
/// Should have the same <c>.GetLength(0)</c> as <paramref name="catFeatures"/>
/// </param>
/// <param name="catFeatures">
/// 2D array of categorical features encoded as strings.
/// Should have the same <c>.GetLength(0)</c> as <paramref name="floatFeatures"/>
/// </param>
/// <returns>2D array with model predictions for all samples in the batch</returns>
public double[,] EvaluateBatch(float[,] floatFeatures, string[,] catFeatures)
{
if (floatFeatures.GetLength(0) != catFeatures.GetLength(0))
{
if (floatFeatures.GetLength(0) > 0 && catFeatures.GetLength(0) > 0)
{
throw new CatBoostException("Inconsistent EvaluateBatch arguments:" +
$"got {floatFeatures.GetLength(0)} samples for float features " +
$"but {catFeatures.GetLength(0)} samples for cat features");
}
}
uint docs = (uint)Math.Max(floatFeatures.GetLength(0), catFeatures.GetLength(0));
uint dim = CatboostNativeInterface.GetDimensionsCount(ModelContainer.ModelHandler);
uint resultSize = dim * docs;
double[] results = new double[resultSize];
bool res = false;
unsafe
{
fixed(float *floatFeaturesPtr = floatFeatures)
{
using (var catFeatureHolder = new StringPointerHolder(catFeatures)) {
float *[] floatFeaturesBeginPtrs = new float *[docs];
for (int i = 0; i < floatFeatures.GetLength(0); ++i)
{
floatFeaturesBeginPtrs[i] = floatFeaturesPtr + i * floatFeatures.GetLength(1);
}
fixed(float **fff = floatFeaturesBeginPtrs)
{
res = CatboostNativeInterface.CalcModelPrediction(
ModelContainer.ModelHandler,
docs,
fff, (uint)floatFeatures.GetLength(1),
catFeatureHolder.MainPointer, (uint)catFeatures.GetLength(1),
results, resultSize
);
}
}
}
}
if (res)
{
double[,] resultMatrix = new double[docs, dim];
for (int doc = 0; doc < docs; ++doc)
{
for (int d = 0; d < dim; ++d)
{
resultMatrix[doc, d] = results[dim * doc + d];
}
}
return(resultMatrix);
}
else
{
var msg = CatboostNativeInterface.GetErrorStringConst(ModelContainer.ModelHandler);
throw new CatBoostException(
"An error has occurred in the CalcModelPredictionSingle() method in catboostmodel library.\n" +
$"Returned error message: {msg}"
);
}
}