Parameters | Predictive Modeling | Target rank (k) of the low-rank randomized SVD

Use the text box to specify the target rank (k) of the low-rank randomized singular value decomposition (SVD) of the additive and dominance matrices.The lower the target rank is, the faster is the computational time, however, possibly with larger approximation error. Target rank should be smaller than the number of observations (n) with non-missing values for the dependent variable. When k is greater than n/1.5, the randomized SVD might not be computationally efficient when compared to the truncated SVD. Therefore, when this condition is found, truncated SVD is computed instead of randomized SVD.

Make sure that the Compute near optimal low-rank randomized SVD for the additive and dominance matrices check box is checked.