You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@mahout.apache.org by Matt Mitchell <go...@gmail.com> on 2012/12/27 16:26:58 UTC

time based price predictions

I'm looking for a way to predict prices based on day of the week, and
possibly even buckets of days like weekends, holidays etc.. I have the data
(item-id, price, date) ... Does anyone have any advice for how to get
started with this using Mahout?

Thanks,
Matt

Re: time based price predictions

Posted by Manuel Blechschmidt <Ma...@gmx.de>.
Hi Matt,
you can model the problem for price prediction as a time series problem. Currently Mahout has no direct support yet for time series like e.g. R http://stat.ethz.ch/R-manual/R-patched/library/stats/html/ts.html

Like Ted already said you have a kind of regression problem. Currently there are not a lot of people doing directly regression with mahout. Most of the classification implementation are also suitable for regression. Nevertheless if you do not have a Ph.D. in statistics it will be tough to transfer the classification algorithms to regression algorithms. I would currently say that a good start for you are ARIMA models:

http://en.wikipedia.org/wiki/Autoregressive_integrated_moving_average

As far as I know there is no ARIMA implementation in Mahout.

If you want to include events e.g. modeling the price drop of a certain product when the successor is introduced it becomes tricky. You can check out project Black Swan http://www.blackswanevents.org/ They try to correlate events like changing governments with e.g. tax increase.

Here is also an answer from me directly for time series analysis:
http://mail-archives.apache.org/mod_mbox/mahout-user/201112.mbox/%3C93573C87-ACB4-4351-ABE8-EB141AE592F2@gmx.de%3E

To sum it up: There is currently no good support for time series analysis in Mahout. Get you hands dirty and implement it or use R.

Hope that helps
    Manuel

On 27.12.2012, at 16:26, Matt Mitchell wrote:

> I'm looking for a way to predict prices based on day of the week, and
> possibly even buckets of days like weekends, holidays etc.. I have the data
> (item-id, price, date) ... Does anyone have any advice for how to get
> started with this using Mahout?
> 
> Thanks,
> Matt

-- 
Manuel Blechschmidt
M.Sc. IT Systems Engineering
Dortustr. 57
14467 Potsdam
Mobil: 0173/6322621
Twitter: http://twitter.com/Manuel_B


Re: time based price predictions

Posted by Ted Dunning <te...@gmail.com>.
You have a sort of a regression problem here.

Add features of each item if you can.  Then add day-of-week, weekend or
holiday features.  Fit your regression.

Can you say the size of your data?

On Thu, Dec 27, 2012 at 7:26 AM, Matt Mitchell <go...@gmail.com> wrote:

> I'm looking for a way to predict prices based on day of the week, and
> possibly even buckets of days like weekends, holidays etc.. I have the data
> (item-id, price, date) ... Does anyone have any advice for how to get
> started with this using Mahout?
>
> Thanks,
> Matt
>