That's great.  I was not aware that you could have empty fields in a data
point.  I do have plenty of data actually.  I have more than 4 parameters
so that is not a problem.  I was just using 4 as an example.

I will try this method and see if there is a performance difference from my
way of offsetting to the next data point.  Thanks a lot.

Ozgur


On Tue, Jul 8, 2014 at 9:49 PM, Matthew Taylor <[email protected]> wrote:

> I would probably split up the x and y values into different fields.
> Assuming each value has some time associated with it:
>
> timestamp,x,y
> timestamp,float,float
> T,,
> 12/30/10 09:00,x,
> 12/30/10 10:00,x,
> 12/30/10 11:00,x
> 12/30/11 00:00,,y
>
> It's okay to send in a row with no data for a field, so you could pass the
> x values at each time they actually occur, and the summarized y value at
> the end of the day. The only thing I'm worried about is you having enough
> data. Four data points a day is not a lot, and it will take a long time for
> NuPIC to learn the patterns unless you have a lot of historical data to
> pass it.
>
>
> ---------
> Matt Taylor
> OS Community Flag-Bearer
> Numenta
>
>
> On Tue, Jul 8, 2014 at 5:41 AM, R. Özgür Aksu <[email protected]> wrote:
>
>> Yes sorry for not being more specific. Each line is a single data point
>> in time. For example, you can think of the input values (xn) being readings
>> for a day in the morning, and the y output being the final overall
>> evaluation of the day in the evening. A concrete example can be seismic
>> readings and I would want Nupic to alert me if there is an earthquake soon
>> at that location. Wouldn't that be incredible?
>>  On 7 Jul 2014 17:59, "Matthew Taylor" <[email protected]> wrote:
>>
>>> Ozgur,
>>>
>>> I'm a little confused by your question, so hopefully you can
>>> clarify... Does "x1, x2, x3, x4, y" represent one row of data you want
>>> to push into NuPIC for a point in time? Is there a timestamp
>>> associated with this row? Or is "x1, x2, x3, x4, y" 5 different values
>>> for 5 different points in time?
>>>
>>> Thanks,
>>> ---------
>>> Matt Taylor
>>> OS Community Flag-Bearer
>>> Numenta
>>>
>>>
>>> On Fri, Jul 4, 2014 at 10:35 AM, R. Özgür Aksu <[email protected]>
>>> wrote:
>>> > Hi,
>>> >
>>> > I would like to test Nupic again on a time series but I'm not sure of
>>> the
>>> > best way to set this up.  I have a training set with parameters and
>>> target
>>> > output:
>>> > x1, x2, x3, x4, y.
>>> >
>>> > But what I really want to do is given x1-4, predict the 'y'
>>> enumeration.  I
>>> > have tried to do this before by offsetting y like this:
>>> > xa1, xa2, xa3, xa4, 0.
>>> > xb1, xb2, xb3, xb4, ya.
>>> > xc1, xc2, xc3, xc4, yb.
>>> >
>>> > And predicting the next y value.
>>> >
>>> > Is there a better approach?  Is this a correct way to set this up?
>>> >
>>> > Thanks.
>>> > Ozgur
>>> >
>>> >
>>> > _______________________________________________
>>> > nupic mailing list
>>> > [email protected]
>>> > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>>> >
>>>
>>> _______________________________________________
>>> nupic mailing list
>>> [email protected]
>>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>>>
>>
>> _______________________________________________
>> nupic mailing list
>> [email protected]
>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>>
>>
>
> _______________________________________________
> nupic mailing list
> [email protected]
> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>
>
_______________________________________________
nupic mailing list
[email protected]
http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org

Reply via email to