next up previous index
Next: Comparison of trees built Up: VLMM results Previous: VLMM results   Index




Comparison of the Lidstone probability estimation with the maximum likelihood probability estimation

We have compared the effect of using the Lidstone and the maximum likelihood estimation of probability during model training. The comparison is done twice: once in conjunction with the Matusita distance and a second time in conjunction to the KL divergence. The experiments have been done on sequences of letters. The VLMM has been trained on English texts. The results are presented in a form of a tree where every element that has a probability higher than 0.003 is shown. The sequences should be read from the leaves to the root of the tree. For instance figure 6.3 shows the sequences ``rea'', ``ha'', ``la'' and so on, from left to right.



Subsections



franck 2006-10-01