Not really, We don't need to change the minsup at any point really. But I will investigate some more.
I'm currently testing a different technique for building conditional trees (trying to remove the bonsai pruning completely) and collecting/outputting closed FP's. The current implementation uses way too many heaps to collect patterns (their creation is expensive, and this is a bottleneck as we are creating a new heap (where the PriorityQueue creation takes lots of time) every time we are mining a single-path tree). We can limit ourselves to having just N heaps, where N is the number of frequent singleton patterns. Seems like I can even use a different thread for pattern collection purposes, as the mining and pattern collection can overlap. If I get any significant results, will let you know. yarco;) On Mon, Jun 27, 2011 at 5:38 PM, Robin Anil <[email protected]> wrote: > I dont remember about the (/2) part. But the line was supposed to get the > min support of the mined conditional tree and raise the minSupportValue for > subsequent mining of conditional trees up the header table. Does that make > sense? > > Robin > > On Tue, Jun 28, 2011 at 3:32 AM, Yarco Hayduk <[email protected]> wrote: > > > Hi, > > > > I'm having trouble understanding some of the code in the FP-Growth > > algorithm. > > > > What is the purpose of the > > > > minSupportValue = Math.max(minSupportValue, > > minSupport.longValue() / 2); > > > > line. When exactly can the minsup value change? > > > > yarco;) > > >
