Pyhton inplementation of:

1. Facets: Fast Comprehensive Mining of Coevolving High-order Time Series [PDF],
2. Fast Mining of a Network of Coevolving Time Series [PDF].

Given a Network of High-order Time Series (Net-Hits), this algorithm can recover its missing parts indicated by the indicator tensor W or predict t time step after X.

class facets.Facets

__init__(self, X, ranks, weights)

• X: nd-array
  • tensor of shape N_1 x N_2 x ... x T
• ranks: int list
  • size of latent tensor Z (i.e., len(ranks) == tensor.ndim)
• weights: float list
  • weight of contextual information for each mode of X.
    if weight = 0, then the contextual information is ignored.
    if weight = 1, then only the contextual information included to learn observation tensor U.

The dynamic contextual matrix factorization algorithm encodes both contextual information and temporal dynamics. It is a special case of Facets algorithm where the order of input tensor M = 1.