Good stuff Michael!  Nice that it's all there in Rmoses.  I have time on 
this planned for the weekend and will definitely see how I go getting some 
parsing done in Rmoses.  I have slack and am subscribed to the moses 
channel, so i will post there for some guidance if hit issues. ;-)

On Tuesday, 11 February 2020 14:43:52 UTC+11, Michael Duncan wrote:
>
> lance, apropos to your comment about not finding documentation on 
> eval-table, there is a combo parser for Rmoses because i didn't know it 
> existed when i was writing the code :P
> feel free to ask questions about it here or the moses channel in OpenCog 
> slack if you do slack.
>
> On Monday, February 10, 2020 at 5:14:58 PM UTC+8, Lance White wrote:
>>
>> Thanks Michael, I am not fluent in R but can hack my way through for 
>> sure!  Really great to have this option. I had downloaded it, but did not 
>> realise it can evaluate too.  Will definitely spin it up and explore 
>> functionality.
>>
>> On Monday, 10 February 2020 14:53:06 UTC+11, Michael Duncan wrote:
>>>
>>> if you are familiar with R you can use the wrapper code i wrote for 
>>> binary classification problems:  https://github.com/mjsduncan/Rmoses
>>>
>>> the documentation is crappy and it needs to be rewritten in the 
>>> tidyverse idiom but it handles producing and scoring combos on training and 
>>> testing partitions and producing feature counts from model ensembles.
>>>
>>> it includes a combo parser so you can try the combos on new samples.
>>>
>>> On Monday, February 10, 2020 at 5:30:19 AM UTC+8, linas wrote:
>>>>
>>>>
>>>>
>>>> On Sun, Feb 9, 2020 at 4:26 AM Lance White <[email protected]> wrote:
>>>>
>>>>> Hi to All,
>>>>>
>>>>> A really basic question about Moses.  So I can run the examples and 
>>>>> test files no problems.  But how do I use the output combo program?
>>>>>
>>>>> moses -H it -i disjunction.csv
>>>>>
>>>>> 0 or($1 $2 $3) 
>>>>>
>>>>> -1 true 
>>>>>
>>>>> -1 or($1 $2) 
>>>>>
>>>>>
>>>>> Let's say I want to use the following on a data set how do I feed it 
>>>>> values and get a result output by moses?
>>>>>
>>>>>
>>>>>  or($1 $2 $3) 
>>>>>
>>>>
>>>> Well, if you have three values, each being either true or false, well, 
>>>> just pipe those values through the logical or function, and you're done.
>>>>
>>>> Yes, this sounds silly, doesn't it?  Every user of moses has built some 
>>>> large, complex system around it to feed it tables of data and then to 
>>>> process other data using these outputs. Unfortunately, all of these are 
>>>> proprietary systems, but it occurs to me that maybe one of these could be 
>>>> open-sourced. I'll have to ask. The one I know best can take structured 
>>>> and 
>>>> unstructured data from a variety of sources, filter and process these into 
>>>> training-data sets, collect up a number of best-fit moses colbo results, 
>>>> average them together into an ensemble, and then output that as a 
>>>> data-processing pipeline: you feed it tables (or individual lines from 
>>>> tables) and it makes predictions based on that input.  Practical 
>>>> experience 
>>>> shows that it is more-or-less comparable to "decision forests" (a decision 
>>>> forest being an ensemble of decision trees; a decision tree being 
>>>> kind-of-like a combo program but obtained from different algorithms).  
>>>> Both 
>>>> moses and decision-forests max out at a certain level of accuracy, beyond 
>>>> with it takes an ungodly amount of training time to improve on.  There was 
>>>> a nascent effort to apply deep learning techniques, but it foundered on a 
>>>> lack of funding.
>>>>
>>>> I'm not sure I answered your question, but that's what I've got.
>>>>
>>>> -- Linas
>>>>
>>>> -- 
>>>> cassette tapes - analog TV - film cameras - you
>>>>
>>>

-- 
You received this message because you are subscribed to the Google Groups 
"opencog" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/opencog/4fd7d097-d1e6-4273-815a-79f1559f7fc8%40googlegroups.com.

Reply via email to