unidata / python-training Goto Github PK

When a PR is merged, it should trigger nikola deploy_github and publish the new pages to the gh-pages branch. It is not currently doing that, as the changes in #4 are not in the gh-pages version of index.rst.

There are still 3 PRs that need porting over here from python-gallery:

• GLM + ABI
• Hurricane Tracker and SPC Tracker
• Radar and Sfc obs

Starting notes for improvement:

• Using generators to refactor loops: http://pastebin.com/9qgPQxV3

This would have caught the minor bug we found sooner. Need to add the conda-forge/label/rc channel somehow and make it an "allow_failure" build on Travis.

One of the exercises (involving making a power function), has the solution in the cell. This probably escaped the check because it doesn't have the %load in the cell.

Create a declarative version of the surface data plotting notebook.

Obvious thing to do is the station plot. Could also throw in some gridding and contouring to make a nice composite plot. Bonus points to also include a satellite image. Probably most easily done with the upper air data.

In reviewing some support tickets for separate issues, I discovered that multiple notebooks across the gallery and workshop

pages/gallery/500hPa_Vorticity_Advection.ipynb
pages/gallery/850hPa_Temperature_Advection.ipynb
pages/gallery/Ageostrophic_Wind_Example.ipynb
pages/gallery/declarative_500_hPa.ipynb
pages/gallery/HILO_Symbol_Plot.ipynb
pages/gallery/miller_composite.ipynb
pages/gallery/MSLP_temp_winds.ipynb
pages/gallery/Smoothing_Contours.ipynb
pages/gallery/xarray_500hPa_map.ipynb
pages/workshop/MetPy_Advanced/QG Analysis.ipynb
pages/workshop/MetPy_Case_Study/MetPy Case Study.ipynb

attempt to access data or guide users with outdated ncei.noaa.gov/thredds/<access>/grid/<product> data access URLs for GFS/NAM/NARR products. These can now generally be found at ncdc.noaa.gov/thredds/<access>/<product> (NCDC data access) which was correctly described in some of these same notebooks. If this was already known or expected, let me know; I didn't see anything on this elsewhere!

• We should probably get back the actual grid point instead of just what we requested.
• Where's the reference for the time in the return?

@dopplershift commented on Mon Sep 12 2016

It would be nice to do a cool inset axes like this:

@ahaberlie commented on Tue Sep 13 2016

For ideas on how to approach this, cartopy has an official gallery item that does something similar:

http://scitools.org.uk/cartopy/docs/latest/examples/effects_of_the_ellipse.html

The query on the surface obs notebook takes many minutes to get from THREDDS and is not cached. We've discussed trying to get the METAR products from THREDDS test and use the METAR parser. Worth a shot - certainly a good real world test!

We should only do things one way:

• Consistent imports
• use of subplots vs. add_subplots vs. axes()
• Use Matplotlib OO interface rather than pyplot where possible

• Why are we calling parse_cf if we’re going to hard-code the data projection to PlateCarree—should use `cartopy_crs.
• Why index and squeeze? Should try to use sel, otherwise there’s really no point to our putting this in an xarray

Would be good to probably migrate the nbviewer notebooks over to this repo (or at least have the hyperlink have the correct repo name). Not sure those will stay alive if the old repos are mothballed.

flake8-nb exists, so it should be possible to flake8 check the notebooks.

We need an automated check to see if we are calling deprecated functions - probably limited to warnings from MetPy and siphon for sanity?

We should look at having easy access to running the notebooks in the cloud using something like MyBinder, or even with our own on something like Jetstream.

Create a satellite plotting notebook using the new declarative interface.

We've had several systems quarantine the miniconda install - a note to turn off the virus checker for install would be good. Avast seems to be the worst about doing this.

Pandas notebook has an extra import statement that should be removed.

SciTools (the UK Met office group) has some interesting ideas in their course materials. Things I thought were interesting:

• Exercise breakdown into parts in different cells
• Exercises seem different from ours, more of a jump from what's taught
• NumPy broadcasting image
• Numpy exercises in general

While in general I think import metpy.plots as mpplots, then thinks like mpplots.ImagePlot() are "better" for certain values of "better", it flies in the face of the GEMPAK replacement spirit. Importing all the things you need as from metpy.plots import ImagePlot, MapPanel, PanelContainer, ... is also verbose. After discussion, we decided in training materials targeted towards GEMPAK users to go with from metpy.plots.declarative import *. Yes, I don't love it, but it's the most user forward solution here.

Could hit a few sources like:

• GFS
• HRRR
• NDFD
• National model blend

Good example that uses NCSS but is simple on the plotting. There's a variety of ways to do multi-panel. Bonus points for using widgets to allow changing plot.

Update the call to the now-removed dewpoint_rh with dewpoint_from_relative_humidity.

Update and cleanup the Skew-T notebook before AMS.

From Unidata/python-workshop#418: It would be good to develop some holistic examples where we put everything we've taught into an exercise or two at the end of a workshop (see AMS 2018 case study). Ideally, this would incorporate remote access of multiple datasets with different visualizations (likely of a specific case(s), i.e. SkewT, satellite/radar, surface & upper air plots, and model data).

Environment was renamed unidata - need to replace instances of training in the install docs.

Almost had this for a winter storm. Just not sure what lines are of particular interest to make a good spaghetti plot.

There are still solution buttons in the Case Study notebook

In Millersville Unidata/python-workshop#143 we had a request for upper-air maps with standard dewpoint depression, etc. While we don't currently have a good data source for that, it should be on the list.

Would be great to hit the radar server and download data for some recent, interesting case. Could use OPeNDAP to get part of the level2 data, maybe also download NIDS for things like tornado detections.

For all the weird ways jupyter and conda envs together get weird behavior:

• no jupyter in env
• mac: Library/Jupyter stale configs

Need to take advantage of better siphon API.

For instance, in the 250 hPa example, this:

cat = TDSCatalog('http://thredds.ucar.edu/thredds/catalog/grib/'
                 'NCEP/GFS/Global_0p5deg/latest.xml')
best_ds = list(cat.datasets.values())[0]
ncss = NCSS(best_ds.access_urls['NetcdfSubset'])

should be

cat = TDSCatalog('http://thredds.ucar.edu/thredds/catalog/grib/'
                 'NCEP/GFS/Global_0p5deg/catalog.xml')
latest_ds = cat.latest
ncss = latest_ds.subset()

Or could even elide the last two lines.

Go through the siphon notebook and do any needed cleanup/updating before AMS.

This is probably trivial using xarray.

The distractors here are not very good and it's easy to get to the one correct answer. Needs some thought to improve.

Let's name the notebooks consistently. I don't love using spaces, but that's me being an old man yelling at clouds. After conversation with @dopplershift we decided to go with that, so the repo notebook names need cleaned up.

Skew-T notebook improvements:

• Remove the list around zero in the isotherm plotting example

This exercise should be removed as decided at the Winnipeg workshop.

We've got a mix of non green background exercise boxes (which do work in jupyter lab now) and no background boxes. Which shall we choose to standardize on?

At some point in the last two weeks, Travis moved this repo to .com from .org, which broke our encryption. Tests for PRs work but will not deploy correctly currently.

Add pre-release testing for mpl et al.

In pythonic data analysis be sure to include using the dictionary get with to have default value and a dictionary entry for different than defaults.

To report a non-security related issue, please provide:

• the version of the software with which you are encountering an issue
• environmental information (i.e. Operating System, compiler info, java version, python version, etc.)
• a description of the issue with the steps needed to reproduce it

If you have a general question about the software, please view our Suggested Support Process.

The pandas notebook is too exercise heavy. It results in a choppy class flow and long periods of no work for the more advanced students.

Right now, when adding a new notebook to the gallery, we also need to:

• Capture a thumbnail
• Add to the main gallery page

We really should be able to automate extracting an image from the rendered notebook (Nikola handles making proper thumbnails for us). Probably also need to figure out a way to organize the gallery--maybe folders are enough?

Add an example overlaying a GLM field on top of an IR image.

Right now new notebooks have to be added to the outline. We've also talked in the past about developing some system to generate a zip file, and maybe an agenda site, for each workshop. This becomes more important now that we're in this unified repository.

Work on the model data plotting notebooks and use declarative plotting where possible.

It'd be nice to get the notebook/solution testing to run on windows!