This library implements imputation of incomplete 3-way tensors using CP decomposition.
Below we show how to impute a tensor of shape (m,n,p) with a mask missing_mask.
missing_mask also of shape (m,n,p) and has entries 0 when the data is missing, 1 when available.
a,b,c are the factor matrices of shape mx7 nx7 and px7 respectively.
re gives the reconstruction error and ie gives the imputation error.
Both the errors are relative/normalized mean squared errors with respect to the original tensor and the missing entries respectively.
from unknownev.imputer import impute
a,b,c,re, ie = impute(tensor, rank=7, tol=1e-10, n_iters=10000, mask=missing_mask)