search

mlr-archive / mlr-tutorial

The mlr package online tutorial
http://mlr-org.github.io/mlr/
20 stars 11 forks source link

Add a chapter on clustering to the mlr tutorial #71

Closed jakob-r closed 6 years ago

jakob-r commented 7 years ago

As mentioned here we might need a dedicated chapter for clustering.

schiffner commented 7 years ago

Thx @jakob-r for transferring the issue.

The current tutorial structure is in the spirit of mlr in that it emphasizes the unified interface (make a task, make a learner, train it, etc.) instead of individual learning tasks. So, it's good for folks that want to learn mlr in depth, but I agree, when you are in a hurry and/or just interested in how to solve a certain learning task it's tedious to collect the info scattered over multiple pages.

I wouldn't add a whole clustering chapter because it would just repeat the information that is already there in a different order (please correct me if something essential about clustering is missing). Instead, I would add a "quick start" example for clustering. Some background info: We have talked several times about extending the "quick start" section here. We plan to add short examples for all standard use cases. These should be complete (so that folks could copy and edit the code for their application), but without in-depth explanations -- just some hints where to find more information.

@rwarnung, @PhilippPro: Agreed?

jakob-r commented 7 years ago

I agree. After all it's about machine learning methods and they are interchangeable for the most part. There is no classification chapter - why should there be a clustering chapter. Could you (@rwarnung) point which are special clustering features of mlr other than makeClusterTask?

rwarnung commented 7 years ago

@schiffner a quick start example sounds very good for me. Of course with enough abstraction it is not that different from other learners. I just didn't find it 100% clear how to do a cluster analysis using mlr.

@jakob-r you are right with enough efforts one (and I ;) ) is able to do it. I just thought addressing some things directly were good. E.g. I am often asked about the differences between caret and mlr and there are a lot (you know better than I do). For instance, as far as I know built-in stacking is a clear distinction. And if I think again clustering is one too. Thus making them easily available could draw further attention to the package.

pat-s commented 6 years ago

Following up on this, is there still a need for a clustering intro? If yes, who would be a person that has expertise in this? @rwarnung?

jakob-r commented 6 years ago

It might be part of the mlr hackathon at WhyR so let's close it here.

rwarnung commented 6 years ago

@pat-s thank you for coming back to my post. In fact I have played a bit but finally did not do that much clustering. So, unfortunately I don't really have that much expertise ... Thank you all for the work on mlr - I love it ! @jakob-r thank you too!

pat-s commented 6 years ago

It might be part of the mlr hackathon at WhyR so let's close it here.

Ok!