https://github.com/numenta/nupic/issues/440

It is quite easy, but depends when sb actually makes it done. I hope it's
rather soonish.


On Fri, Nov 22, 2013 at 8:52 PM, Fergal Byrne
<[email protected]>wrote:

> Hi Marek,
>
> It's easy enough: map Subutai's code into a dict by naming the components:
>
> {
>   "region": {
>     "name": "ImageLayer1",
>     "type": "SensorRegion",
>     yadayada..
>   "link": {
>     "input": "ImageLayer1",
>     "output": "CLA1",
>     "type": "defaultLink"
> ....
>
> This could be digested by a dict-to-API converter which would
> programmatically run Subutai's Network-building code.
>
>
>
>
> On Fri, Nov 22, 2013 at 6:53 PM, Marek Otahal <[email protected]>wrote:
>
>> hi Fergal,
>>
>> yes, that would be ideal solution. It would require some programming:
>> add id field to each {sensor/sp/tp/class ...} field, define list of
>> inputs, outputs and a dict of linkage.
>>
>> This linear thing could be a first step, as it's almost no-brainer with
>> no visible changes to the code.
>>
>>
>> On Fri, Nov 22, 2013 at 7:06 PM, Fergal Byrne <
>> [email protected]> wrote:
>>
>>> Hi Marek,
>>>
>>> I would generalise that further, by allowing a description.py to specify
>>> the entire Network description, using a dict structure to encode Subutai's
>>> example from the API thread. This would be easy to program and give you the
>>> full power of the API.
>>> —
>>> Sent from Mailbox <https://www.dropbox.com/mailbox> for iPhone
>>>
>>>
>>> On Fri, Nov 22, 2013 at 5:57 PM, Marek Otahal <[email protected]>wrote:
>>>
>>>>        the description.py file tells (in OPF format) the model factory
>>>> how to set up a region.
>>>>
>>>> It consists of the following :
>>>>
>>>> Encoder --> (optional) SP --> (opt) TP --> Classifier
>>>>
>>>>
>>>> The format is something like JSON:
>>>> {
>>>>    {sensor: 'encoder': 'ScalarEncoder' ,...}
>>>>    {SP: ...}
>>>>    {TP: ...}
>>>>    {Classifier: ...}
>>>> }
>>>>
>>>> I think adding support for "linear hierarchy" wouldn't be much pain and
>>>> will be easy gain.
>>>>
>>>> By linear hierarchy I mean stacking multiple SPs and TPs upon each
>>>> other. ( encoder->SP->TP->SP-TP->Classifier).
>>>>
>>>> By non-linear hierarchy I would mean :
>>>>
>>>>   C
>>>>  /   \
>>>> A    B
>>>>
>>>> This linear hierarchy would be easy for users to set up, just write
>>>> more Sp,TP entries in the description.py (it should respect the order
>>>> added).
>>>>
>>>> Advantage would be forming of more stable patterns which come with
>>>> higher hierarchy layers.
>>>>
>>>>
>>>>
>>>> --
>>>> Marek Otahal :o)
>>>>
>>>
>>>
>>> _______________________________________________
>>> nupic mailing list
>>> [email protected]
>>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>>>
>>>
>>
>>
>> --
>> Marek Otahal :o)
>>
>> _______________________________________________
>> nupic mailing list
>> [email protected]
>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
>>
>>
>
>
> --
>
> Fergal Byrne, Brenter IT
>
> <http://www.examsupport.ie>http://inbits.com - Better Living through
> Thoughtful Technology
>
> e:[email protected] t:+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
>
>


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

Reply via email to