There are participants kaggle.com?

Interesting experience:

just to start

— do team

— what algorithms are typically used (if it does not know how :) )


Thank you.
October 8th 19 at 00:11
3 answers
October 8th 19 at 00:13
— To start, really difficult. A very long time veshti to the task.
— In any case needed if you are a genius. It is very easy to get stuck on some algorithm. Need a strong mathematician + coder at least.
— The algorithm strongly depends on the task. But, basically, in addition to the standard set of datamining is not a bad thing to know computational math (gradient methods, boundary value problem, transora task, etc.) and standard solutions to standard problems.
Gut, thanks for the reply. Ask a couple of questions:
— does it make sense to look for teammates there? if there is an easy way to find a ready-made team to join?
— at the expense of the algorithms. I asked because according to the statistics of previous contests people love decision and tries different combinations of classifiers. Ie similar approaches vlob not working. Interestingly, this is a General trend or again depends on the task.

I am working to do the classification of texts. It would be interesting to try a few approaches. - Gustave.Ha commented on October 8th 19 at 00:16
I have very little experience, I will Express only their value judgments.
Best team score among the "friends". Ie those you know who do not experience difficulties in communication, with whom you can meet to discuss of the algorithm for a beer. The Internet is definitely convenient, but meeting physically at one point for some reason always adds intelligence. My choice and advice — definitely work in his team.
Standard solutions work. But slow and extremely inefficient. But the task is often 90% — these are the preparatory stuff. Only the last 10% — a place for imagination and intellect. The lion's share of work will have to be done is purely mechanical. The easiest way to "solve vlob" to understand that there is a solution, and then to find workarounds to speed up to improve is to add precision.

If you are associated with semantic analysis, then your knowledge is clearly much more deeply than I do. You definitely should try on different celengan. - sigrid_Sporer16 commented on October 8th 19 at 00:19
At the expense of the team agree. But "their" and next to that is luxury :). I think I will try on a simple contest, and see what happens.

At the expense of semantics — it's bad you know. I'm more on the side of statistics, it's easier. Honestly tried to Wade through homepages.inf.ed.ac.uk/jbos/comsem/ but did not understand how from the text to induce something formal, like logics of the first order. For small texts and grammar goes, but on toy examples the case comes. So stats and ML seem to be more productive. - Gustave.Ha commented on October 8th 19 at 00:22
October 8th 19 at 00:15
At the expense of the team agree. But "their" and next to that is luxury :). I think I will try on a simple contest, and see what happens.

At the expense of semantics — it's bad you know. I'm more on the side of statistics, it's easier. Honestly tried to Wade through homepages.inf.ed.ac.uk/jbos/comsem/, but did not understand how from the text to induce something formal, like logics of the first order. For small texts and grammar goes, but on toy examples the case comes. So stats and ML seem to be more productive.
October 8th 19 at 00:17
Here he began to write a detailed answer to the question "how easy is it to start?"
habrahabr.ru/post/165001/
Thank you. Though, I don't understand why you went from sql. But maybe this is a reason. For the link to Vowpal Wabbit, recommended it to me recently.

By the way, I usually work with sklearn (scikit-learn, a python library). - Gustave.Ha commented on October 8th 19 at 00:20

Find more questions by tags Data miningAlgorithmsMachine learning