Thanks everyone.

So, we decided to have following improvements;

1. /s/ml/machineLearner
2. Instead of specifying storage-location, we should get the model-name
from the user.
3. feature elements should be wrapped from a features element.
4. prediction element currently replaces an existing element in the
incoming message, we can support the append as well.

On Tue, Mar 24, 2015 at 11:56 PM, Chanaka Fernando <[email protected]>
wrote:

> Hi Manorama,
>
> I guess that the names are not finalized yet. The name "ml" does not make
> much impression on the users. Please give a valid name like
> "MachineLearner" and that would make life easier for ESB developers. All
> our mediators has self explanatory names.
>
> Thanks,
> Chanaka
>
> On Tue, Mar 24, 2015 at 4:09 PM, Supun Sethunga <[email protected]> wrote:
>
>> Hi Maninda,
>>
>> With the initial release of ML, the model would not update on the fly. If
>> we need to update the model, it has to be done in ML by training a model
>> with the same configurations, but with the new data. In future releases we
>> will be looking at this for possible improvements.
>>
>> Here the term "domain" refers to the definition of the data which use to
>> train the model. In the case you mentioned, If the model was trained using
>> the data collected from all incoming requests; then domain is "all sales".
>> and can be used to handle all incoming requests. But if the model was
>> trained using the data collected from the incoming requests of, say
>> Seller_A, then it cannot be used to predict/handle incoming requests of all
>> sellers, which is a different domain.
>> When we say "domain", it also includes the Schema (set of
>> features/variables) of the dataset on which the prediction is going to be
>> done. Handling multiple schema is tricky. If the data from a different
>> domain doesn't contain some of the features (variables in the dataset) used
>> to train the model, then that model cannot be used to predict on that
>> particular data. So we have to work with this limitation, hence all the
>> data has to be on the same domain (i.e. should have the same schema).
>>
>> Regards,
>> Supun
>>
>> On Tue, Mar 24, 2015 at 3:10 PM, Maninda Edirisooriya <[email protected]>
>> wrote:
>>
>>> Does this update the model time to time or use the same modal throughout
>>> the ESB up time? It is better if we can update the model while the ESB is
>>> up and running.
>>> There may be some use cases related to multiple domains right? So I
>>> think would be more generic and useful if we can apply the mediator not
>>> based on the domain but with a custom argument expression. For example if
>>> the target service is SalesService and the operation is getTotalSalesCount
>>> we want to use the modal to analyze all incoming requests from all domains.
>>> WDYT?
>>>
>>> Thanks.
>>>
>>>
>>> *Maninda Edirisooriya*
>>> Senior Software Engineer
>>>
>>> *WSO2, Inc.*lean.enterprise.middleware.
>>>
>>> *Blog* : http://maninda.blogspot.com/
>>> *E-mail* : [email protected]
>>> *Skype* : @manindae
>>> *Twitter* : @maninda
>>>
>>> On Tue, Mar 24, 2015 at 12:00 PM, Manorama Perera <[email protected]>
>>> wrote:
>>>
>>>> Hi,
>>>>
>>>> ESB mediator for WSO2 ML enables making use of the models generated by
>>>> WSO2 Machine Leaner, to predict values in message mediation in ESB.
>>>>
>>>>    - Machine Learner feature needs to be installed in ESB prior to
>>>>    configure the ML mediator.
>>>>    - Mediator will extract the data from the incoming messages and
>>>>    send to the ML model as features.
>>>>    - The mediator use the Machine Learner features to predict the
>>>>    output value for every message passing through the mediator.
>>>>
>>>> *Proposed Mediator Configuration*
>>>>
>>>> <ml>
>>>>     <model storage-location="string"/>
>>>>     <feature name="string" expression="xpath"/>+
>>>>     <prediction expression="xpath"/>
>>>> </ml>
>>>>
>>>> *model*
>>>> storage-location : Storage location of the generated model.
>>>>
>>>> *feature*
>>>> name : Feature name according to the generated model.
>>>> expression : XPath expression used to determine the feature value.
>>>>
>>>> *prediction*
>>>> expression : XPath expression to determine the location to insert the
>>>> predicted value.
>>>>
>>>> The generated models can be applied to predict only the data coming
>>>> from the same domain, as the data used to train the model. So the mediator
>>>> should be configured only to mediate messages which carry data coming from
>>>> the same domain, as the data used to train the model.
>>>>
>>>> Any suggestion are appreciated.
>>>>
>>>> Thanks.
>>>>
>>>> --
>>>> Manorama Perera
>>>> Software Engineer
>>>> WSO2, Inc.;  http://wso2.com/
>>>> Mobile : +94716436216
>>>>
>>>> _______________________________________________
>>>> Architecture mailing list
>>>> [email protected]
>>>> https://mail.wso2.org/cgi-bin/mailman/listinfo/architecture
>>>>
>>>>
>>>
>>> _______________________________________________
>>> Architecture mailing list
>>> [email protected]
>>> https://mail.wso2.org/cgi-bin/mailman/listinfo/architecture
>>>
>>>
>>
>>
>> --
>> *Supun Sethunga*
>> Software Engineer
>> WSO2, Inc.
>> http://wso2.com/
>> lean | enterprise | middleware
>> Mobile : +94 716546324
>>
>> _______________________________________________
>> Architecture mailing list
>> [email protected]
>> https://mail.wso2.org/cgi-bin/mailman/listinfo/architecture
>>
>>
>
>
> --
> --
> Chanaka Fernando
> Technical Lead
> WSO2, Inc.; http://wso2.com
> lean.enterprise.middleware
>
> mobile: +94 773337238
> Blog : http://soatutorials.blogspot.com
> LinkedIn:http://www.linkedin.com/pub/chanaka-fernando/19/a20/5b0
> Twitter:https://twitter.com/chanakaudaya
> Wordpress:http://chanakaudaya.wordpress.com
>
>
>
>
> _______________________________________________
> Architecture mailing list
> [email protected]
> https://mail.wso2.org/cgi-bin/mailman/listinfo/architecture
>
>


-- 

Thanks & regards,
Nirmal

Senior Software Engineer- Platform Technologies Team, WSO2 Inc.
Mobile: +94715779733
Blog: http://nirmalfdo.blogspot.com/
_______________________________________________
Architecture mailing list
[email protected]
https://mail.wso2.org/cgi-bin/mailman/listinfo/architecture

Reply via email to