The DecisionForestPredict function uses the model output by ML Engine DecisionForest function to analyze the input data and make predictions.
If your model table was created using a supported version of Aster Analytics on Aster Database, see AA 7.00 Usage Notes.
DecisionForestPredict outputs the probability that each observation is in the predicted class. To use DecisionForestPredict output as input to ML Engine ROC function, you must first transform it to show the probability that each observation is in the positive class. One way to do this is to change the probability to (1- current probability) when the predicted class is negative.
The prediction algorithm compares floating-point numbers. Due to possible inherent data type differences between ML Engine
and NewSQL Engine
executions, predictions can differ. Before calling the function, compute the relative error, using this formula:
relative_error = (abs(mle_prediction - td_prediction)/mle_prediction)*100
where mle_prediction is ML Engine prediction value and td_prediction is NewSQL Engine prediction value. Errors (e) follow Gaussian law; 0 < e < 3% is a negligible difference, with high confidence.