The MultiOrderModel implementation is constructed with layers starting from 1. The model selection is currently implemented with some workarounds to account for the frequencies and probabilities of transitions in zeroth order.
Do we want to keep the current situation, or shall we add a zeroth order layer in the MultiOrderModel implementation?
The MultiOrderModel implementation is constructed with layers starting from 1. The model selection is currently implemented with some workarounds to account for the frequencies and probabilities of transitions in zeroth order.
Do we want to keep the current situation, or shall we add a zeroth order layer in the MultiOrderModel implementation?