Please find answers inline:
The questions I have:
1) Are we going to implement one extension for every algorithms or an new
extension for each learning algorithms.
One extension for all algorithms (please looks at
https://github.com/wso2-extensions/siddhi-execution-math for an example)
2) Are we going to use the same extension for training and predicting by
adding a new parameter to select if it is a training or a prediction :
For every algorithm, we will have two methods namely: fit (for training)
and predict, similar to Scikit-Learn API
Regarding the proposal :
1) Do I have to mention about the performance and accuracy evaluation
techniques, benchmark data sets that will be used, or is it enough to just
focus on the high level design of the proposed system.
When it comes to performance, Please run your model against well-known
datasets and present your result.
Thanks,
Upul
On Sat, Mar 25, 2017 at 10:26 AM, Nadheesh Jihan <nadhee...@gmail.com>
wrote:
> Hi,
>
> I sent the link to my prototype for the project proposal-24. I was
> expecting some feedback from you. Your feedback will be valuable for me to
> understand the requirements and produce a better proposal.
>
> Link to the prototype - https://github.com/Nadheesh/
> siddhi-execution-streaming-ml
>
> As I said earlier, I'm very interested in this project, since it is
> aligned with my passion and skill set. Therefore, I want to do my best for
> this project. I hope you don't mind me in asking some further questions
> since I want to further improve my prototype.
>
> In my prototype, I use a Stream Processor. So far, I have implemented a
> single functions to train and predict using *Perceptrons*.
>
> *sml:pml( name_of_algorithm, label, feature1, feature2, feature3)*
>
> *name_of_algorithm* - Name of training algorithms. (For now only
> Perceptron)
> *label* - The attribute name which contains the label of data (For
> supervise algorithms)
> *feature1, feature2...* - Attribute names of features
>
> The extension update the *Perceptron* model per each event. I tried to
> keep my algorithm very simple since we need to perform the training process
> with a small latency(real-time). And just to illustrate the prediction
> procedure, this stream processor returns a prediction for every given
> training data event as well
> Since this is a prototype I did not implement both versions.To implement
> the prediction procedure we can use the same extension or an another
> extension. However, I think we will need to persist the models in order to
> implement the prediction separately.
>
> The questions I have:
>
> 1) Are we going to implement one extension for every algorithms or an new
> extension for each learning algorithms.
>
> 2) Are we going to use the same extension for training and predicting by
> adding a new parameter to select if it is a training or a prediction
>
> Regarding the proposal :
>
> 1) Do I have to mention about the performance and accuracy evaluation
> techniques, benchmark data sets that will be used, or is it enough to just
> focus on the high level design of the proposed system.
>
> Thank you.
>
> --
> Best regards,
> *Nadheesh Jihan*
> Undergraduate | Department of Computer Science and Engineering
> Faculty of Engineering
> University of Moratuwa
>
--
Upul Bandara,
Associate Technical Lead, WSO2, Inc.,
Mob: +94 715 468 345.
_______________________________________________
Dev mailing list
Dev@wso2.org
http://wso2.org/cgi-bin/mailman/listinfo/dev
