Hi,

On Wed, May 11, 2016 at 4:20 AM, 박진만 <[email protected]> wrote:

> Thank you all. I think that you two are talking about the same version.
>
>
> I'm not familiar with coding C++, so I prefer to use python language.
>
> Then can I use the fastest TP if I include temporal_memory.py like this:
>
>       from nupic.research.temporal_memory import TemporalMemory
>
not exactly. Either you work with a more high-level OPF and in the
description.py you will use `"implementation": "cpp"` (or "tm_cpp", to be
exactly sure search the examples or the code).

Or you work with the TM,SP,.. classes directly, then you'll need to import
`from nupic.bindings.... import ...`

All in all, the API is the same, so you can write your code using the Py
versions and then switch to the C++ wrappers. I'd say this is a recommended
devel approach.

>
> Or Should I use C++ language to use the fastest TP?
>
My unconfirmed experiment suggests using C++ directly, see the PR for "low
level anomaly example" in nupic.core is even significantly faster then
using the C++/swig wrappers for python, but you've said your not very
familiar with CPP so the wrappers should be a good choice.

> (Can you tell me how to include the fastest TP version?)
>
>
> Also, can you tell me the fastest SP and CLA version and how to include
> them?
>
As above, use the "cpp" implementation, I think it's a default in OPF
description.py experiments.

Cheers,


>
> Thank you for answering my questions so kindly.
>
>
>
>
> 2016-05-11 0:27 GMT+09:00 Matthew Taylor <[email protected]>:
>
>> Marcus has been making some really great performance improvements to
>> the C++ TM, and claims it is now faster than the fastest version of
>> the old TP algorithms.
>>
>> If you want speed, I believe your best bet is to use NuPIC and the
>> tm_cpp implementation.
>> ---------
>> Matt Taylor
>> OS Community Flag-Bearer
>> Numenta
>>
>>
>> On Tue, May 10, 2016 at 6:57 AM, cogmission (David Ray)
>> <[email protected]> wrote:
>> > That would be here: HTM.java ! :-)
>> >
>> > Just being funny, but quite possibly true!
>> >
>> > I think the temporal_memory.py runs the new C++ TM underneath, and so
>> for
>> > Python - it might be the fastest because the algorithm has just been
>> updated
>> > to process things more efficiently... But the Numenta engineers can
>> confirm
>> > this.
>> >
>> > Cheers,
>> > David
>> >
>> > On Tue, May 10, 2016 at 7:36 AM, 박진만 <[email protected]> wrote:
>> >>
>> >> Hello Nupic, I saw that there are many versions of SP, TP and CLA in
>> nupic
>> >> library.
>> >>
>> >> For example, there are many TP versions; TP.py, TP10X2.py,
>> >> temporal_memory.py.. etc.
>> >>
>> >> What I want to know is that which one is the fastest one among those TP
>> >> implementations.
>> >>
>> >> I want to use the fastest one because I want to deal with a big amount
>> of
>> >> data.
>> >>
>> >> Can you tell me the most useful and the fastest SP, TP, and CLA
>> >> implementation codes?
>> >>
>> >> thank you.
>> >>
>> >>
>> >
>> >
>> >
>> > --
>> > With kind regards,
>> >
>> > David Ray
>> > Java Solutions Architect
>> >
>> > Cortical.io
>> > Sponsor of:  HTM.java
>> >
>> > [email protected]
>> > http://cortical.io
>>
>>
>


-- 
Marek Otahal :o)

Reply via email to