better (and working) tutorials about matminer HOT 9 CLOSED

Note - I have created what should be a working Jupyter notebook for the initial intro.

Someone should go through the remaining tutorials and make sure they work ...

from matminer.

Picking this up, if no one objects. Also, I kind of prefer the pymatgen format for showing the notebooks, which is to have the exported html linked in the documentation and a link to the nbviewer page for interactive versions. I'm not sure it'll play nicely with plotly, but if it does, I'll switch it over.

from matminer.

@montoyjh , when plotting in your tutorial, will you consider using figrecipes plotly wrappers? @ardunn and I are improving its defaults and adding 1-2 new plotly plots here and would appreciate your feedback if you end up using any of them. We may be able to submit a PR within the next few days if you can wait with plotting as there will be some re-structuring.

from matminer.

Sure, that sounds great. The plotly functionality seems pretty great for looking at outliers or in areas of descriptor space you're particularly interested in.

from matminer.

I'm also having a bit of trouble with accessing 2 of the tutorials via git-lfs, specifically MPDS_UO_bondlengths.ipynb and the one on PbTe bandgaps. I can get them from the gists linked in the docs, if need be, but just wanted to mention it.

from matminer.

@montoyjh We had lots of problems with git-lfs and tried our best to nuke git-lfs and all traces. Instead of git-lfs we wanted to do all the tutorials via gist - i.e., so they don't occupy space in the main repo, avoid the problems of git-lfs, but are still easy to access/modify/update/comment on. Was there some remnants of git-lfs left over?

from matminer.

Ah, great.

As far as gist, can I propose an alternative solution? If you leave the notebooks in unexecuted form, they take up significantly less space (all of the notebooks take up around 100-200K, as opposed to 20-40M). If done this way, they can still be included in the repo. When building the docs, you can run jupyter nbconvert --to html --execute which will execute the cells and export to html, which then would be copied to the _static folder and linked in docs.

Pros:

• Don't have to maintain external gists (i. e. one place for notebooks where they get fed into the docs directly)
• I can write a unit-test for the notebooks if they're on the main repo, to ensure the examples are working.

Cons:

• the exported HTML in the docs are still heavy (seems like roughly 50% of what the executed notebooks are). Perhaps there's a way to not include these in the repo but still in the docs?

from matminer.

from matminer.

The tutorials have been fixed and moved to matminer_examples, I'm considering this closed; additional thoughts should go to the matminer_examples repo

from matminer.