As many of us have already begun to play with some Kaggle datasets of competitions currently active I thougth it would be good to know @Jeremy 's general guidelines for competing in those during the course.
First part of the question would be when and where to share the code we are using. My understanding is that all code that we share on this forum (including “standard” lesson’s notebooks if applied to active comps) should be mentioned at some point in Kaggle forum. (This “standard” part probably just as a mention to github fastai repository that is public anyway).
So… would this understanding be ok? Will rule 1 be rule: sharing here ===> sharing in kaggle forum?
Second part of the question, I guess its discarded to make a huge “kaggle team” for fastai because of the limitations of submissions it would imply. (Being able to submit different ideas fluently is essential to kaggle comps). “Problem is”…
well, with the tools we are learning now anyone here could actually “accidentally” win.
So, are we supposed to actually compete? Or in the winning scenario we would just assume to be labelled as “cheaters” by kaggle? I don’t care as much about not claiming the prize but I wouldn’t like to have this “cheater” label over me…
So, I guess the general question for Jeremy is: what do you want us to do, and how do you expect us to do it? For me it is clear we are representing fast.ai in Kaggle during this month so maybe some guidelines about how to do it will be useful.
Hello @miguel_perez, It’s funny, but I asked myself exactly the same questions, I want to participate in competitions, but I also want fast.ai to get credit for this fantastic library, it would be great to have a framework.
@miguel_perez, I appreciate your paying attention to these issues and posting these questions. I see some of these same questions being raised in other threads and I think it’s good to have a “reference” thread directly addressing this topic so the guidelines are clear for all. I think we all want to do the right thing and be good ambassadors for fast.ai!
For running comps, you shouldn’t share code here, but instead share thru kaggle kernel or kaggle forum and link to here - or form a team and share privately.
Yes, compete as best you can! The tools here are publicly available, so there’s no problem using them. If you use the pretrained weights files in
fastai/models/weights, you will need to link to them in a kaggle forum post (generally speaking - check the rules of the comp you’re entering).
Got it! This is going to be fun!
@jeremy: I see a “lock” icon in the upper left corner of the screen, next to “Part 1 v2”.
Could that block access to the link posted on Kaggle forums, for non-registered participants of Part 1 v2 ?
@miguel_perez: great points you raised.
Sorry - I thought I had a solution to this dilemma and after writing an overly long post went to recheck the kaggle rules…Turns out I missed some details that are crucial:
No private sharing outside teams
Privately sharing code or data outside of teams is not permitted. It’s okay to share code if made available to all participants on the forums. (guess they are referring to their forums, that’s where all the conversation should take place)
Team mergers are allowed and can be performed by the team leader. In order to merge, the combined team must have a total submission count less than or equal to the maximum allowed as of the merge date. The maximum allowed is the number of submissions per day multiplied by the number of days the competition has been running.
Yes - so what I’m suggesting is you post over there, and link to it from here; not the other way around
I’ve a habit of of re-watching videos and trying to understand the lecture. After that I head towards assignment. Eventually never able to start with kaggle. Guys any recommendations?
I hear you. I have had the same problems. May be try the other way around - Start with a Kaggle Competition. Then re-watch portions of the video as needed to complete a basic submission. Repeat & Iterate
@vikbehal @ramesh , I begin with assignment/lesson notebook, then try to use that in a Kaggle comp that same week. It helps to “generalize” the knowledge