machine learning - Searching for binary range coincidence over many dimensions -
i've clarified , simplified question:
i have data looks this:
011100111110100111 111111111111110010 111100001111000011
1d lanes of streams of data. each row signifies presence of type 1 or 0. types tend exist in chunks across stream. vertical order of rows doesn't matter.
i seeking patterns dimensions coincide @ 2 or more indices, both @ start of '1' group's, , where '1' groups overlap across rows/dimensions.
pattern can offset nearby adjacent indices, if it's proportion maintained.
d = dimension/row n = index in stream
(( d1(n), d25(n+4), d900(n-1) ), ( 3, 67, 90, 3000 ))
an example of pattern match migh occur @ multiple places. dimension01 @ n, dimension25 @ index n + 4, dimension900 @ index n -1 occur @ indices 3, 67, 90 , 3000.
the returned patterns:
- are ordered number of matching indices or number of dimensions in pattern.
- match @ least 2 dimensions @ at least 2 points
how can go this?
as far understand, appraoches frequent itemset mining might looking for. starting point famous apriori algorithm, 1 of basic techniques detect common blocks of "1" elements in example (assuming can interpret each column independent dimension, , there no specific semantic/relationship between dimension, modify problem since order of dimensions becomes meaningful).
Comments
Post a Comment