SAX-VSM suboptimal optimization

[snvv]

Hello
In the webpage example 5.0 SAX-VSM classifier, the un-optimized parameters:

w <- 60 # the sliding window size
p <- 6 # the PAA size
a <- 6 # the SAX alphabet size
are optimal, since the error term is 0 (zero miss-classifications rate)

However in the example 6.0 SAX-VSM parameters optimization the optimal parameters are "65 12 4 -> 0":

w <- 65 # the sliding window size
p <- 12 # the PAA size
a <- 4 # the SAX alphabet size
and the miss-classification error is [1] 0.004444444 with 4 misclassified series ([1] 187 589 766 860)

What I am missing?

Also, I tried the optimization with my own data set and I get the following error

Error in bags[[unique(train_labels)[j]]] <- manyseries_to_wordbag((train_data[set_train, :
attempt to select less than one element in OneIndex

Thank you in advance for your help