Sounds like a good plan! But keep an eye on the number of people applying (we will try to make ppl submitting public applications) so that you have a plan B just in case.
Just reviewed your first PR :) H 2016-03-14 21:53 GMT+00:00 Pan Deng <[email protected]>: > Hi Heiko, > > Thanks for your reply! > > I feel like focusing on one project is better than shotgun method for > getting a spot in GSoC. So I guess I'll focusing on applying for the detox > project. I have forked Shogun and am trying to solve some small issues > (mean calculation in linalg), though there're still problems with the codes > for now. Hopefully I will get more familiar with the codes soon. Thanks! > > Best, > Pan Deng > > On Mon, Mar 14, 2016 at 1:43 PM, Heiko Strathmann < > [email protected]> wrote: > >> Hi Pan, >> >> welcome to Shogun! Your skills are very welcome here. This year, we are >> in particular looking for people who are willing to do some proper software >> engineering with us. It is fine if your experience is limited -- we can >> help and guide you -- as long as you are motivated and learn fast :) >> Also make sure to visit our GSoC wiki -- there is lots of info there. >> >> Detox: >> This project is challening -- but it is really important to us. If you >> are interested in large scale framework design, then this is for you. There >> will be much less competition as for other projects as people mostly apply >> to the ML projects. Make sure to send some good initial patches to get some >> communication going between us and you. A good schedule is essential for >> the application -- try to split all major parts into sub-projects. If this >> is planned well (and possible problems investigated early), this project >> will be much more doable. >> >> Fudamental ML: lots of applicants probably. We might take two though. And >> we really want strong applications here. >> >> Approx kernel methods: As for the detox, you are the one of the first >> interested. The project is cool, relatively straight forward (we want to >> add well known algorithms -- and benchmark them). However, the project is >> not among our favourite ones as we want to limit the amount of new >> algorithms added. It will depend a bit on how many applications we receive >> and how strong the are. The background here is simple: you have to >> understand kernel ridge regression and that should do it. >> >> Best >> Heiko >> >> 2016-03-13 21:40 GMT+00:00 Pan Deng <[email protected]>: >> >>> Hi everyone, >>> >>> I am Pan Deng, a Ph.D. student from Memorial Sloan Kettering Cancer >>> Center, New York. I would like to have the chance to join GSoC and Shogun! >>> >>> I have been learning algorithms and large dataset processing in >>> Python/Matlab/R/Java/C++. I am especially interested in C++ programing (a >>> very efficient language but relatively hard to be expertised in) and >>> machine learning (who isn't nowadays!) So Shogun seems to be the best place >>> for me! >>> >>> I have some thoughts/questions about the following ideas, and I would >>> love to get advice from you. >>> >>> *- Shogun Detox* >>> I have no experience on large projects, which seems to be my drawback >>> for this project. However, on the other hand, I would like to learn how to >>> organize large projects, so I will be super patient with the work. Also, I >>> have been following C++ Primer for C++11, so I can help with codes upgrade. >>> >>> I am wondering how proposal will be like for this project. I am guessing >>> I can plan for one subproject per week? >>> >>> *- Fundamental Machine Learning Algorithms/ Large-scale Approximate >>> Kernel Methods* >>> I have had ML course on Coursera, and knowing that this course is too >>> entry, I am taking a Biomedical Machine Learning course offered by our >>> graduate school from late March. This course will be quite practical and we >>> will have several projects to work on. The course will introduce >>> approximate Bayesian computation, Kaplan-Meier estimator, convolutional >>> neural network and so on. I am expecting to be more experienced in ML and >>> help you with the algorithm improvement. >>> >>> Yet realizing that the first project is quite popular, I would like to >>> give a shot for Kernel Methods Implementation. I have learned about SVM/PCA >>> but pretty basic, though I believe I can learn very fast. So I am wondering >>> how much do you expect from the candidate for Kernel Methods project and >>> how likely you'll carry out this project this year? >>> >>> Sorry for making this mail so long and thanks for your reading! >>> >>> Best, >>> Pan Deng >>> >>> >>> >> >
