The TreeCatServer.mode(data) method would be useful for imputing maximum a posteriori values of missing cells, conditioned on partially observed cells. This method should input a table of partially-observed data (as a ragged array), and output a completed array of the same shape and dtype, just like ServerBase.median().
To implement this, we can follow a similar propagation algorithm as in .mean() or .sample() (note that those two methods vectorize in different ways). The .mode() method should be able to compute all MAP estimates for the entire table using only a single propagation pass. It should do this by vectorizing over rows of the table to be imputed, as .mean() does.
The
TreeCatServer.mode(data)method would be useful for imputing maximum a posteriori values of missing cells, conditioned on partially observed cells. This method should input a table of partially-observed data (as a ragged array), and output a completed array of the same shape and dtype, just likeServerBase.median().To implement this, we can follow a similar propagation algorithm as in
.mean()or.sample()(note that those two methods vectorize in different ways). The.mode()method should be able to compute all MAP estimates for the entire table using only a single propagation pass. It should do this by vectorizing over rows of the table to be imputed, as.mean()does.