Cells of the year by series grid are numbered series by series, years within
a series, which is the order as.vector() lays out a [year,
series] matrix. A path from series j to series k at lag
L fills cell (t, k) of row and cell (t - L, j) of column
in \(B\), for every year t > L.
Value
List with n_cells, n_grid_yrs and, for \(I - B\) and
\(\Gamma\), a numbered sparse template (m), the entry filling each
stored slot (slot_entry), the arrow behind each entry
(entry_arrow, 0 for the diagonal of \(I - B\)), the row and
column of each entry (entry_row, entry_col) and its year
(entry_yr), which a moderated arrow reads its value at.
det_is_one says whether the cells can be ordered so that everything
an arrow comes from is set before what it points to,
needs_dense_logdet whether dgmrf will hold a random effect in the
precision, and, for a reduced rank model, project_k with the cells
of a solved series, obs_idx and unobs_idx splitting the
cells the way dsem does, and Q_oo the pattern
get_dsem_Q_oo fills.