[mlpack] GSoC 2014 : Introduction and Interests

[Anand Soni]

Hi Ryan and Marcus,

I have submitted the proposal in detail on the GSoC 2014 portal. I
have tried to take care of all issues we discussed here. Kindly go
through it and let me know if more details are required or if
something has been missed/wrongly put. I have added a few references
at the end too. Also, I have mentioned in the proposal that we had a
talk about the project so that you can recognize me easily.
(PS : There are some formatting issues in the submitted version which
I couldn't correct. Let me know if that is a problem). Looking forward
to working with you this Summer.

Hope to hear from you soon.

Regards.
Anand Soni

On Thu, Mar 13, 2014 at 9:01 PM, Anand Soni <anand.92.soni at gmail.com> wrote:
> Hi Ryan,
> I will definitely keep the code as flexible as possible considering possible
> future additions.
>
> And regarding the new classifier, the algorithm is fixed i.e the back
> propagation algorithm. We can improve on the efficiency of implementation
> which I will keep in mind if I go for implementing it. I welcome any other
> queries from your side.
>
> Thanks a lot.
> Regards,
> Anand
> ________________________________
> From: Ryan Curtin
> Sent: 13-03-2014 20:06
> To: Anand Soni
> Cc: Marcus Edel; mlpack at cc.gatech.edu
> Subject: Re: [mlpack] GSoC 2014 : Introduction and Interests
>
> On Thu, Mar 13, 2014 at 03:35:55PM +0530, Anand Soni wrote:
>> Hi Marcus,
>>
>> I have decided on the metrics to be implemented. Mostly, I will be
>> implementing and testing with the following metrics :
>>
>> a) Accuracy
>> b) F-measure
>> c) Lift
>> d) Precision
>> e) Recall
>> f) Matthews correlation coefficient (turned about to be a good
>> suggestion from you)
>> g) Mean Squared error
>> h) Mean predictive information (very similar to cross entropy, but
>> more easily interpreted)
>>
>> So, there are 8 metrics in total. The last two are probabilistic
>> measures. I will talk about these metrics in detail in the proposal.
>> Please tell me if you want something to be added or removed.
>
> I would like to know if your code will be flexible enough to support
> additional metrics that may be added in the future.  If it's not, that
> is something to consider.
>
>> Also, does mlpack already has a classifier based on feed-forward
>> neural networks and the backpropagation algorithm? If not and if time
>> permits, I would like to add it as a classifier for mlpack.
>
> No, it does not; if you are interested in adding this, let's make sure
> it is faster than competing implementations, either through careful
> implementation, or by implementing a more efficient algorithm.
>
> --
> Ryan Curtin    | "Here... there... what's the difference?"
> ryan at ratml.org |   - Prisoner



-- 
Anand Soni | Junior Undergraduate | Department of Computer Science &
Engineering | IIT Bombay | India