_calculate_ms_and_mpd() about sompy HOT 2 CLOSED

The question is how to select the meta-parameters.
Heres, mpd deals with the number of iterations. mpd is the ratio between number of nodes and number of training data. trainlen is decided based on mpd but in 4 different ways:
(rough_train or finetune_train)x( initialization == 'random' or self.initialization == 'pca'). the coefficients are chosen logically and based on several experiments.
ms is to decide about the radius in neighborhood search.
for both of these factors, they are calculated differently if the SOM grid is one dimensional.

It would be much better if those final coefficients would be input parameters.

from sompy.

Thanks that makes sense with what I discovered in the meantime too. So I have modified my version of .train to allow training length and the initial and final radii to be set from outside the som and to allow multiple training sessions (i.e., weights not reset every time train is run).

from sompy.