PRESEE: An MDL/MML Algorithm to Time-Series Stream Segmenting

Algorithm 1

Segment in slide window.

Input: points in slideWindow Seg = ⟨𝑝₁,𝑝₂,𝑝₃⋯𝑝_𝑙𝑒𝑛⟩,

    batchSize

Output: character points set C = ⟨𝑐_𝑖1,𝑐_𝑖2⋯⟩

(1)   Put 𝑝₁ into C

(2)   𝑐𝑢𝑟𝑆𝑖𝑧𝑒=0,startIndex = 1, length = 1;

(3)   While  startIndex + length < len do

(4)    curIndex = startIndex + length;

(5)    𝐶𝑜𝑠𝑡_seg = MDL_seg(𝑆_{𝑠𝑡𝑎𝑟𝑡𝐼𝑛𝑑𝑒𝑥},𝑠_{𝑐𝑢𝑟𝐼𝑛𝑑𝑒𝑥}); 𝐶𝑜𝑠𝑡_noseg =   

     MDL_noseg(𝑐_{𝑠𝑡𝑎𝑟𝑡𝐼𝑛𝑑𝑒𝑥}, 𝑐_{𝑐𝑢𝑟𝐼𝑛𝑑𝑒𝑥});

(6)     If  𝑐𝑜𝑠𝑡_seg >𝑐𝑜𝑠𝑡_noseg  then

(7)     Put 𝑝_{𝑐𝑢𝑟𝐼𝑛𝑑𝑥𝑒−1} into C;

(8)     ++curSize;

(9)    If  curSize == batchSize then

     // enough batch has been processed

(10)     Return  C;

(11)      Else  startIndex = curIndex – 1; length = 1;

(12)      Else  length = length + 1;

(13)      If  C has only one point do

(14)     Return NULL;

(15)      Else   Put_{𝑝𝑙𝑒𝑛} into C;