Use DM_setRegRSquared to specify the maximum tolerated squared Pearson correlation coefficient r². This coefficient can be tolerated in the regression model for the error on verification data. If the training algorithms find that it is impossible to construct a model with r² below that threshold, a warning is issued, but a valid model is constructed. Note: r² (with 0 <= r² <= 1) measures the correlation between the prediction error on verification data and the actual target values. r²=0 means that the prediction error and the target field values are uncorrelated, r²=1 means that the two values are perfectly (positively or negatively) correlated.
Therefore, setting an upper threshold for r² means setting an upper limit for the acceptable systematic error of the model.