ISAXUtils missing?

[vaske maskinsen]

Hi everyone again,

I just wanted to try the iSAX-example from here: http://code.google.com/p/jmotif/wiki/iSAXJavaPrimer - but the ISAXUtils class seems to be missing. I couldn't find it anywhere in the code. Has this changed, so the example is not up to date? Should I use SAXFactory somehow instead? Or was iSAX removed completely, then why?

Greets,
Vaske

[Pavel Senin]

Hi Vaske:

At some point there were some internal compatibility issues with code, so I moved it into "isax" branch - you can get all the sources from there.

Sorry about the issues.

--
You received this message because you are subscribed to the Google Groups "jmotif-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to jmotif-discus...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

--
Mahalo, Pavel.

[vaske maskinsen]

Thank you seninp!
I'll try that...

Greets,
Vaske

[vaske maskinsen]

Hi Pavel,

it's a lot of work getting it to run... I guess I have to combine branches isax-2010-backup and isax, since I cannot find ISAXUtils in the (newer) isax branch. I'm really curious: Why did you throw that code out? Was it not so useful or unstable? Will you reinclude it in the future? As far as I understand iSAX, it's mostly useful for fast searches, right? Or what reason would I have to prefer iSAX? Because SAX is left in the trunk... I am confused. I read a lot of papers, but that's just theory :) . However, iSAX 2.0 (http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=5693959) looks promising. Was the previous iSAX code influenced by that one or "just" the "1.0"?
Maybe you can explain it. Or maybe Josh?

Thank you very much in advance,

Vaske

Am Dienstag, 7. Mai 2013 14:00:13 UTC+2 schrieb seninp:

[Pavel Senin]

Hi Vaske:

I don't think that code was unstable or if it wasn't performing as intended. 

I think you are pretty safe to go with it, there might be some lost connections with JMotif, but this can be repaired. (Josh built his code upon SAX, so, I assume, his code is not broken at all, it just need to be connected by fixing syntax of calls to evolved JMotif API. - I really had no time at that point to do that. Sorry.)

Now, personally, I never worked with billions of series since I can't figure out how to handle hundreds of those :). Thus I simply couldn't advise you about on what is better, but to some extend could point some information: SAX-based methods are not exact, iSAX, in my opinion, approximating things up even more, so it becomes possible to tackle really large data. Thus you have to make decision on your own - if you can afford to loose some of the good results and willing to deal with false positives.

I use SAX to convert timeseries from long real-valued time series of numbers into a dictionary (corpus) of SAX words (terms), then I use tf*idf to build a vector space which, in turn, is used for classification and knowledge discovery (I am after this knowledge - so I can't afford loosing something or getting wrong answers). Within last months I pushed a lot of changes making this possible. In oncoming time, I plan to settle down with codebase, boost test coverage, remove excessive dependencies, organize API of JMotif and "mavenize" the build. I hope to release 1.0 _stable_ version. We have some plans of bringing in Sequitur and n-Grams language models, but there is no one working on high-throughput end - i.e. iSAX. If you would be able to make it running by connecting code back to JMotif, I think you will be better with us in your experimentation than with old codebase.

[vaske maskinsen]

Hi Pavel,

thank you very much for providing this clear and helpful information. I'm working in so many fields right now that I don't want to promise too much, but I will try my best to make iSAX work with the new codebase again. I have to deal with millions of huge series, so iSAX seems to be helpful. Sadly I am a C++ coder and working for only one year with java now, so my code may not always be high quality. If it is working for me (could take some time or could be never...), I will of course contribute the code. I could also provide some additional code (SAXification on hadoop, SAX compression etc), but I first have to check if that's ok for my university since it is part of my phd work. I guess at least if I publish the results first it should be ok...
Again, thank you Pavel, also for the GREAT jMotif framework! :)