Figure 5. Cross-validation of Nonnegative TCA on the Rodent Pre-frontal Data.

(A) Schematic of cross-validation procedure, in which elements of the data tensor are held out at random with fixed probability.

(B) Normalized reconstruction error in training set (red) and test set (black) for nonnegative TCA models with a training set comprising 80% of the tensor.

(C) Same as (B), except only 10% of the tensor was used as the training set.