3 replies to this topic

[mahdad]

Dear Colleagues

my problem seems to be a bit frustrating for me so Im here to ask for suggestions. I have a training set of pairs. each pair is labeled (1,-1). the features value is calculated based on a distance function (edit distance). each pair has a feature value equal to the distance between a pair. the distance is calculated based on sequences of deletion, insertion and substitution. each operation (deletion, insertion and substitution) can have a score value so based on that the distance is calculated. for example for a pair the distance is 5 which is a summation of the score of each operation.

now my problem is how to learn/estimate the best score of each operation without assigning it manually. which algorithm could be more suitable? note that each operation usually happens more than one time in each pair. Thanks in advance for your reply.

[niner]

my problem seems to be a bit frustrating for me so Im here to ask for suggestions. I have a training set of pairs. each pair is labeled (1,-1). the features value is calculated based on a distance function (edit distance). each pair has a feature value equal to the distance between a pair. the distance is calculated based on sequences of deletion, insertion and substitution. each operation (deletion, insertion and substitution) can have a score value so based on that the distance is calculated. for example for a pair the distance is 5 which is a summation of the score of each operation.

now my problem is how to learn/estimate the best score of each operation without assigning it manually. which algorithm could be more suitable? note that each operation usually happens more than one time in each pair. Thanks in advance for your reply.

There are a lot of things I don't understand here. How does the pair labeling relate to the distance? What are the meanings of the operations and how do they relate to the score? When you say you want to know the best score of each operation, do you mean over the entire set of pairs? Or for a given pair? Are you trying to find the sequence of operations that optimizes the score?

[Connor MacLeod]

Dear Colleagues

my problem seems to be a bit frustrating for me so Im here to ask for suggestions. I have a training set of pairs. each pair is labeled (1,-1). the features value is calculated based on a distance function (edit distance). each pair has a feature value equal to the distance between a pair. the distance is calculated based on sequences of deletion, insertion and substitution. each operation (deletion, insertion and substitution) can have a score value so based on that the distance is calculated. for example for a pair the distance is 5 which is a summation of the score of each operation.

now my problem is how to learn/estimate the best score of each operation without assigning it manually. which algorithm could be more suitable? note that each operation usually happens more than one time in each pair. Thanks in advance for your reply.

You could try doing a search on www.scholar.google.com for "weighted Damerau distance" or "weighted edit distance," but I'm fairly confident that optimization of the weights, to minimize classifcation error for example, is going to be very difficult; you'll probably have to go with simulated annealing or some related method.

[mahdad]

my problem seems to be a bit frustrating for me so Im here to ask for suggestions. I have a training set of pairs. each pair is labeled (1,-1). the features value is calculated based on a distance function (edit distance). each pair has a feature value equal to the distance between a pair. the distance is calculated based on sequences of deletion, insertion and substitution. each operation (deletion, insertion and substitution) can have a score value so based on that the distance is calculated. for example for a pair the distance is 5 which is a summation of the score of each operation.

now my problem is how to learn/estimate the best score of each operation without assigning it manually. which algorithm could be more suitable? note that each operation usually happens more than one time in each pair. Thanks in advance for your reply.

There are a lot of things I don't understand here. How does the pair labeling relate to the distance? What are the meanings of the operations and how do they relate to the score? When you say you want to know the best score of each operation, do you mean over the entire set of pairs? Or for a given pair? Are you trying to find the sequence of operations that optimizes the score?

Dear Niner

the edit dostance operation is priliminary set, for example del=1, Ins=1, sub=1 and then a distance is calculated as the cost of sequence of operations. for example dis(A,B) is del,del,del,Ins,sub,sub which is 6 since each operation cost 1. so instead of priliminary set opeartin cost to 1 I want to learn/fidn the best operation cost.