update photosynthesis package to use xarray and analyze entire induction curve protocols

[dschneiderch]

This pull request implements the new data model for psII measurements and associated analyze_* functions.
We are using a new class to store multiple data arrays. Each data array has 4 dimensions (x, y, frame_label, measurement). Each data array must have a predefined name: darkadapted, lightadapted, rgb, chlorophyll, anthocyanin

[x] PSII_data class to store multiple xarray data arrays
[x] read_cropreporter to import data from cropreporter instrument
[] read_walz to import data from walz imaging pam
[x] analyze_yii to analyze photosynthetic efficiency for any number of measurements
[x] analyze_npq to analyze nonphotochemical quenching for any number of measurements
[x] tests
[x] documentation

Type of update
Is this a:

• New feature or feature enhancement
• Work in progress

Associated issues
closes #636

Additional context
Add any other context about the problem here.

[dschneiderch]

here's an issue from 2015(!) Don't squeeze DataArray before plotting

[nfahlgren]

Using FacetGrid directly seems to work for both one and multiple measurements:

gd = FacetGrid(data=fvfm, col="measurement", col_wrap=1)
gd.map_dataarray(xr.plot.imshow, "x", "y")

We could put this directly in the YII and NPQ functions for now and not use the _debug module.

[dschneiderch]

@nfahlgren next level wizardry!
another idea I had overnight is to check whether dims x and y are present and if so call pcolormesh(). i actually think they do something like this under the hood but then squeeze the dims too early. I think this would cover most use cases.

[dschneiderch]

also, for hte col_wrap arg I suggest col_wrap=int(np.ceil(fvfm.measurement.size/3)) where the number is the max # of columns

[dschneiderch]

this will all fail if the col arg is removed from this test because then _show_dataarray() returns a histogram as a tuple which does not have self.fig.savefig()

should we just always limit to pcolormesh()? what does a line plot return from xarray?

def test_plantcv_print_image_dataarray():
    da = psii_cropreporter('darkadapted').squeeze('measurement', drop=True)
    try:
        #this fails without col or row as an arg because xarray will create a matplotlib histogram that comes back as a tuple and then print_image will try to use self.fig.savefig which doesn't exist for a tuple
        pcv.print_image(img=da, col='frame_label', filename='/dev/null') 
    except RuntimeError:
        assert False
    # Assert that the image was plotted without error
    assert True

[dschneiderch]

@nfahlgren I think I may have broken the debug= functionality again. sometimes I see the plots even when I have debug=None or debug='print'. clearly I don't understand something.

however, in terms of plotting data array data, I think it will be best to limit to facetgrid functionality using pcolormesh(). it greatly simplifies print_image and plot_image from what I can tell.
the plot output from xarray is different based on the number of dimensions:
1D: da.plot() as a list len 1 containing matplotlib.lines.Line2D bc xarray makes a line plot
3+D: da.plot() comes back as a tuple bc xarray makes a histogram unless col and/or row is supplied for facetting
2D: da.plot() will make an image. facetted images if extra dims are specified as kwargs col or row. in practice I found it's easy to lose track of this and provide too many or not enough dims so I explicitly call pcolormesh() with some sanity checks/error messages in a new private module _show_dataarray . I also check for x and y because I think thats the main use case

[nfahlgren]

I agree, limiting to pcolormesh makes sense for our usecase

[nfahlgren]

Alright, I think everything is wrapped up! Thanks for the huge effort on this @dschneiderch and @HaleySchuhl!