Hi Obdiah, Chetan,

I'm sorry for the long delay, I'm on a holiday now. I'll try to have a look
at Linguist again.
Did you make any progress so far with the problem?

Cheers, Mark


On Fri, Feb 14, 2014 at 11:11 PM, Chetan Surpur <[email protected]> wrote:

> Hi Obadiah,
>
> Actually, that code was contributed by NuPIC community member Mark (I've
> cc'd him). Maybe he can answer your question.
>
> Thanks,
> Chetan
>
>
> On Thu, Feb 13, 2014 at 10:11 PM, Obadiah Lam <[email protected]> wrote:
>
>>  Hi Chetan,
>>
>>  I'm from QUT in Australia and I've been trying to get your linguist
>> example to work. I've installed NuPIC (on Ubuntu) and all the tests work.
>>
>>  I copied your code to the NUPIC folder (where i downloaded the source
>> code for nupic - should I have placed it in the NTA folder where the
>> binaries were instead?) and followed the README instructions, running
>>
>> python client/linguist.py data/tiny.txt
>>
>>
>>  There was a KeyError (line 122 of linguist.py), and doing
>>
>> print repr(result.inferences)
>> print repr('dictionary')
>>
>>  showed that I should be using 'multiStepBestPredictions' and [1]
>> instead of [0] ('prediction' and [0] did not exist in the dictionaries).
>>
>>  Is this correct? What should I be expecting as the output? Currently
>> the stories don't make any sense (I trained it on sNuPIC Linguist KeyError:
>> 'prediction'ome Seuss text files and it's giving me "you win" over and over
>> again).
>>
>>  Any help is muchly appreciated.
>>
>>  Cheers,
>> Obadiah
>>
>>
>>    *Obadiah Lam *|* Research Assistant* | *Electrical Engineering and
>> Computer Science*
>> Science and Engineering Faculty | Queensland University of Technology
>> Gardens Point, S Block 1107 | 2 George Street, Brisbane, QLD 4000
>> email: [email protected]
>>  <[email protected]>
>> https://wiki.qut.edu.au/display/cyphy/Obadiah+Lam
>>
>
>


-- 
Marek Otahal :o)
_______________________________________________
nupic mailing list
[email protected]
http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org

Reply via email to