Thats useful Matt. So Inferences is just a collection of all predictions.

Makes sense

On Wed, Dec 17, 2014 at 7:55 AM, Matthew Taylor <[email protected]> wrote:
>
> Not sure if you are using OPF or not, but if so every call to model.run()
> returns a result object [1] that contains an "inferences" dictionary with
> each prediction and it's confidence in that prediction. Using this, you can
> git the 2nd, 3rd, etc. predictions. The confidence percentages can be very
> useful, too.
>
> If you're not using the OPF, I can't help yet. I'm still learning how to
> construct with the Network API.
>
> [1]
> https://github.com/numenta/nupic/wiki/Online-Prediction-Framework#model-output-the-modelresult-object
>
> ---------
> Matt Taylor
> OS Community Flag-Bearer
> Numenta
>
> On Wed, Dec 17, 2014 at 12:04 AM, Chandan Maruthi <
> [email protected]> wrote:
>>
>> I finally read Jeffs email to the mailing list on new Ideas for Temporal
>> Pooling. Fascinating indeed.
>> Any update on where the current nupic stands with any of the ideas today?
>>
>> Also, A follow-up question on Temporal Pooling.
>> If you get multiple predictions . For instance : "What is ....", the next
>> work can be "your name", "your age", ""your address" etc.na
>> First how do i get access to the multiple predictions. Secondly If the
>> SDR is combination of mutiple predictions for instance prediction
>> A:"001000" and prediction B is "000100" then how do you convert this
>> combination SDR [001100] to individual predictions that correspond to
>> separate potential inputs.
>>
>> Chandan
>>
>> On Mon, Dec 15, 2014 at 8:22 PM, Chandan Maruthi <
>> [email protected]> wrote:
>>>
>>> Hi
>>> I have a few Temporal Pooler Questions
>>>
>>>    1. Difference btw inference and prediction in the TemporalPooler.
>>>    2. Is there a good description of the TemporalPooler input variables.
>>>    3. I see there are a few new ideas about temporal pooling, what is
>>>    the current direction on the approach  [ref this article
>>>    https://github.com/numenta/nupic/wiki/New-Ideas-About-Temporal-Pooling
>>>    ]
>>>
>>>
>>>
>>> Regards
>>> Chandan Maruthi
>>>
>>

Reply via email to