Iron out which minimum dependency versions, exactly, are needed

[fedarko]

Current dependencies (for which this is a concern):

• altair
• biom-format
• click
• pandas
• pytest (not necessarily required, but I think this is a good idea to require anyway so the user can verify their installation is working if desired)
  • update: this, pytest-cov, flake8, and black can be saved as dev requirements in setup.py
• scikit-bio
• numpy

To get context about how each of these libraries are used within rankratioviz, you can navigate to the root directory of the repo and run grep -ri "alt\." * -A 2, where alt is just the name of the module as imported (e.g. skbio for scikit-bio, or pd for pandas). The -A option can be configured to show context after matching occurrences, which helps us figure out what not only what module functions are called but what arguments of these functions are used. This isn't a perfect overview of things -- e.g. if a "from [module] import [thing]" command is used this won't catch that, or if you declare an object of type [module.thing] then this won't show all the ways that object is used -- but it's a solid start.

(also, note that that search command might fail if you somehow invoke a module on one line and then invoke its function on the next line -- is that even doable in python? -- but we're following flake8 on this project so i don't think that'll happen here)

[fedarko]

Altair

• Vega-Lite output version
• Chart, X, Y, Color, Scale, value should all be available
• Should be at least v3.1.0, I think.

biom-format

• Interoperability with all biom files that could possibly be used here? By specifying the [hdf5] extra functionality, I think that covers it.

click

• as long as it works this doesn't really matter I guess

pandas

• Series, (Sparse)DataFrame, Index, merge, read_csv, isna, .testing.assert_frame_equal, .errors.ParserError, to_numeric, creation of a DF from scipy sparse matrix data
• biom's minimum version of pandas is >= 0.20.0 fyi
• looks like 0.20.1 should be a reasonable minimum

pytest

• There was a breaking bug introduced in version 4.1 -- see here. Currently this is set to use pytest<4.1, but eventually we should change that when possible.
• ok, now set to >= 4.2.

scikit-bio

• OrdinationResults.read()
• OrdinationResults features and samples properties
• looks like some biplot/ordinationresults functionality was added in 0.5.3, which corroborates why DEICODE requires at least this version. So it's probs ok to stick with this.

numpy

• right now, just need .NaN and .arange. these are present in v1.3 of numpy, but pandas needs v1.12.0 at least. So let's use that.

[fedarko]

also, update readme when this is sorted out

[fedarko]

We don't directly use numpy, but it's needed in order for biom to be installed at all. And it looks like including it in the install_requires field of setup() (even before biom-format) doesn't suffice.

I imagine most users will probably already have numpy installed, but if we can update the readme or any sort of installation instructions about this.

Yeah, so we do need numpy for a few things (in particular, the np.NaN value and np.arange(). But it looks like both of these things should be available as far back as numpy version 1.3 (not that I expect anyone to actually use a version as old as that).

[fedarko]

I forgot to mention it in the message, but this is closed by commit fedarko@e94c09b.