Hi,
thanks for the answer...but i was looking a more in-depth explanations.
i knew the difference between description.py and descriptionTemplate, i was
trying to understand why the syntax between the two is different.
Same for n and w... i was looking for something like: "ok, if you increase
n you will get this effect, while ....etc." and a rule of thumb to choose
their values.
i have already watched the swarming video. I have tried to run it over the
"sum of two feature" example i described in my first mail but i still get
the same ugly results...so my other doubts still hold.
for what concerning 4) i was wondering if there is a simpler way to do it
than the for cycle ...
last but not least, have you tried to put None as predicted field? it is
working for you?
thanks again
Alessandro
p.s. i'll copy my first email here for quick reference:
1) Is there a tutorial or a reasoned example (i.e. highlighting the main
implementation/options choices.) where i can start from? for now I have
started reading the examples and going up in the "hierarchy", trying to
understand the source code documentation...but it's almost a reverse
engineering approach :). I have found the slides of the oscon lecture...is
there also a video?
2) I have found the description_template.py but it seems to me that has a
quite different syntax respect to the others description files (e.g. the
encoders). am i missing something?
3) in the scalar_1 example:
a) the randomData feature is ignored, isn't it?
b) how the "n" and "w" are chosen? is there a rule?
c) why is the SP disabled?
d) i have tried to change this example to classify the patterns
according to the sum of two features, i.e.
class, field1, field2, randomData
float,float,float,float
,,,
3,1,2, something
7,3,4, something
7,1,6, something
1,0,1, something
5,1,4, something
....
.....
to do this i've just added an encoder for field2 and randomData...but i
am unable to get a decent result. where i am wrong?
4) is there a way to save the model produced and use it again both to make
inferences and to do some more train over new data?
As what concerning the latter i have hacked the OpfRunExperiment.py
file with a for cycle
model=runExperiment(sys.argv[1:])
for c in range (0,9):
runExperiment(sys.argv[1:],model)
but i am quite sure that there is a cleaner and smarter way of doing
this
5) if i put some pattern with None as predicted field, as suggested here (
https://github.com/numenta/nupic/wiki/Spatial-Classification), the model is
actually modified and None is threated as a category/value... and the
output is "None".
On Mon, Sep 9, 2013 at 5:59 PM, Fergal Byrne <[email protected]>wrote:
> Hi Alessandro,
>
> I'll give it a try.
>
> 2. and 3. The description.py file for a certain model describes the
> parameters to use for a run of the CLA. descriptionTemplate.py (in
> examples/opf/experiments/template/base) is a template file for creating an
> individual description.py.
>
> The n and w, as well as which fields are used or ignored, are all choices
> made when editing (or generating) the description.py file for the model.
>
> Take a look at the video on swarming which has just been referenced on
> this list. This shows how a swarming algorithm searches for the best set of
> parameters to put in to the description.py, and you can also hand-code
> those choices.
>
> 4. Yes, you can dump the CLA state at the end of a run, and load it up
> again for future use.
>
>
>
>
> On Mon, Sep 9, 2013 at 4:16 PM, Alessandro Bardelli <
> [email protected]> wrote:
>
>> is there anyone willing/able to answer the others questions?
>> thanks
>>
>> Alessandro
>>
>>
>> On Thu, Sep 5, 2013 at 1:50 AM, Matthew Taylor <[email protected]> wrote:
>>
>>> Hi Alessandro. Thanks for commenting. I can answer one of your questions.
>>>
>>> On Wed, Sep 4, 2013 at 12:16 AM, Alessandro Bardelli <
>>> [email protected]> wrote:
>>>
>>>> 1) Is there a tutorial or a reasoned example (i.e. highlighting the
>>>> main implementation/options choices.) where i can start from? for now I
>>>> have started reading the examples and going up in the "hierarchy", trying
>>>> to understand the source code documentation...but it's almost a reverse
>>>> engineering approach :). I have found the slides of the osconlecture...is
>>>> there also a video?
>>>>
>>>
>>> There will be a video, but OSCON is slow to produce it. Last I heard, it
>>> will be available sometime in November. In the meantime, there are some
>>> videos on our YouTube channel that might interest you:
>>>
>>> http://www.youtube.com/channel/UC8-ttzWLgXZOGuhUyrPlUuA
>>>
>>> ---------
>>> Matt Taylor
>>> OS Community Flag-Bearer
>>> Numenta
>>>
>>> _______________________________________________
>>> 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
>>
>>
>
>
> --
>
> Fergal Byrne
>
> ExamSupport/StudyHub [email protected] http://www.examsupport.ie
> Dublin in Bits [email protected] http://www.inbits.com +353 83
> 4214179
> Formerly of Adnet [email protected] http://www.adnet.ie
>
> _______________________________________________
> 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
