for e.g. the input values can take finite states only, say 10 symbols {A, B, C, D, E, F, G, H, I, J}
@Yuri Please elaborate...
A naive way would be to assign them number in sequential order say A=1, B=2, ..., J=10
Finite automats described by disjunctive and conjunctive Boolean forms – this the most general description.
The way you will use the data can sometimes mean that the best approach to storing the data states is in the initial symbolic form and you can store a reference table that will be used to cross reference the input symbols with the target output function. To enable guys help you better you should state what functions or final output you would like to get. Some approaches discourage the storage of the data states as numeric and the proposed output would be by using various functions to compute the offset between the target input state and the stored data states.
The people of libsvm recommend 1-out-of-k vectors, i.e. a vector the length of your input with a 1 in the dimension corresponding to the state and a zero in each other states.
@Matthias I am dealing with a sequence of finite symbols. However, the length of the sequence is variable. I could use a one vs all vector, but the problem is that I need to consider a window of the sequnece each time and that would complicate matters.
@Sumit: are you reading a finite window of an infinite stream of tokens from a finite alphabet, i.e. what is known as online learning?
@Matthias: Yes. I am trying to do online classification of a variable length sequence of finite symbols
On possible encoding (and feature map) is the frequency of subsequences of finite length smaller than a fix k.
- Badges
- Science topic