search

amueller / scipy_2015_sklearn_tutorial

Scikit-Learn tutorial material for Scipy 2015
Creative Commons Zero v1.0 Universal
576 stars 311 forks source link

First unsupervised or first model selection #2

Open amueller opened 9 years ago

amueller commented 9 years ago

What do you think about whether to do first unsupervised learning or first model selection? In 2013 they did first supervised learning, examples, grid-search, and then unsupervised learning. In the outline I proposed, we'd do both supervised and unsupervised learning in the beginning, and then cross-validation and grid-searches etc.

Do you have a preference?

As we have quite a bit of time, I think it would be nice to start the whole thing by explaining the different kinds of learning, that is supervised, unsupervised, (maybe reinforcement) roughly, and then say that sklearn does both supervised and unsupervised and show the API.

amueller commented 9 years ago

FYI I was planning to use some material from my strata tutorial, but I also have to look through the 2013 notebooks more carefully again: https://github.com/amueller/pydata-strata-2015

I like to interleave notebooks and slides, I think they only did notebooks last time.

kastnerkyle commented 9 years ago

I like the general layout. It would be good to interleave and I will have to take a look through your slides - supervised and unsupervised together make sense to me, then we show how you can combine them to build pipelines.

Notebooks are good but I don't know if using them for the whole thing is better than having slides for the core, and like 4-6 good application/example notebooks.

On Thu, Mar 5, 2015 at 2:46 PM, Andreas Mueller notifications@github.com wrote:

FYI I was planning to use some material from my strata tutorial, but I also have to look through the 2013 notebooks more carefully again: https://github.com/amueller/pydata-strata-2015

I like to interleave notebooks and slides, I think they only did notebooks last time.

— Reply to this email directly or view it on GitHub https://github.com/amueller/scipy_2015_sklearn_tutorial/issues/2#issuecomment-77437108 .

amueller commented 9 years ago

I was quite happy with my strata talk, but it was only 1:30, not 7h (!!). I think with that much time, we should definitely include exercises, and I'll have a look what they did last time. Anyhow, for not I want to submit something and get that off my todo list, we can worry about actually creating the content later ;)

kastnerkyle commented 9 years ago

Sounds great to me. It will be fun! Whatever mods you need to make for the submission are OK with me

On Thu, Mar 5, 2015 at 2:57 PM, Andreas Mueller notifications@github.com wrote:

I was quite happy with my strata talk, but it was only 1:30, not 7h (!!). I think with that much time, we should definitely include exercises, and I'll have a look what they did last time. Anyhow, for not I want to submit something and get that off my todo list, we can worry about actually creating the content later ;)

— Reply to this email directly or view it on GitHub https://github.com/amueller/scipy_2015_sklearn_tutorial/issues/2#issuecomment-77439374 .