Thank you for your reply. I just think HTM is a kind of mechanism that like human brain. Maybe it will do well in classification if it is good enough. The handwritten digits give me inspiration about processing disease data on HTM. Whatever, thank you for your reply.
------------------ 原始邮件 ------------------ 发件人: "cogmission";<[email protected]>; 发送时间: 2015年2月12日(星期四) 凌晨2:56 收件人: "天朗气清"<[email protected]>; 主题: Re: how to use HTM to study disease data It would seem that this particular problem would lend itself well to "Decision Tree" style technologies like Rule/Inference Engines where there are discrete/hard decision points. It would be hard to model the diagnostic nature of this problem in an "online" style where no pre-existing knowledge is present. That *would maybe* be something for an HTM technology far down the road where there are hierarchies capable of true cognition. For now, massive data entry with rulesets is the way to go imo. On Wed, Feb 11, 2015 at 12:24 PM, Matthew Taylor <[email protected]> wrote: I think I agree with Michael. NuPIC is best suited (in its current state) for fast moving temporal data. It seems like your problem is not temporal. --------- Matt Taylor OS Community Flag-Bearer Numenta On Wed, Feb 11, 2015 at 9:46 AM, Michael Klachko <[email protected]> wrote: > Hi yajingfu, your problem is rather simple, so you can probably get better > results by using traditional ML methods (simple neural net, svm, or even > regression). Why do you want to use HTM? > > On Wed, Feb 11, 2015 at 7:29 AM, 天朗气清 <[email protected]> wrote: >> >> hello all >> I sent an email to you last week. But I'm afraid that I didn't get my >> answer. It seems that I didn't explain my problem clearly, so I try to send >> this email again. >> I'm trying to use HTM to analyse disease data. The data have about 4000 >> lines, and each line have many columns, the first column represent for >> diease D, and the rest columns represent for many symptoms S1 S2 S3...Sn. We >> need do infer what kind of desease D is depending on the symptoms S1 S2 >> S3...Sn. I think it's same with the process of handwritten digits >> recognition: first encoding, then spatial pooler, last classification. So I >> regard one piece of disease data as an image. But I think maybe it is not >> the best way to process my data. So I want to know are there any better >> method to process such multifields category? Are there any example that can >> be used as a reference? >> >> >> Hoping for your reply. >> Best regards. >> yajingfu > > -- We find it hard to hear what another is saying because of how loudly "who one is", speaks...
