Rl cleanup

[Rlamboll]

Pull request

Please confirm that this pull request has done the following:

• Tests added
• Documentation added (where applicable)
• Example added (either to an existing notebook or as a new notebook, where applicable)
• Description in CHANGELOG.rst added

Adding to CHANGELOG.rst

Please add a single line in the changelog notes similar to one of the following:

- (`#XX <https://github.com/znicholls/silicone/pull/XX>`_) Added feature which does something
- (`#XX <https://github.com/znicholls/silicone/pull/XX>`_) Fixed bug identified in (`#YY <https://github.com/znicholls/silicone/issues/YY>`_)

[codecov]

Codecov Report

Merging #20 into master will increase coverage by 3.83%.
The diff coverage is 98.11%.

@@            Coverage Diff             @@
##           master      #20      +/-   ##
==========================================
+ Coverage   95.22%   99.05%   +3.83%     
==========================================
  Files           9       10       +1     
  Lines         314      319       +5     
  Branches       52       53       +1     
==========================================
+ Hits          299      316      +17     
+ Misses         11        1      -10     
+ Partials        4        2       -2

Impacted Files	Coverage Δ
...one/database_crunchers/quantile_rolling_windows.py	100% <100%> (+12.5%)	⬆️
src/silicone/cli.py	100% <100%> (ø)	⬆️
src/silicone/utils.py	100% <100%> (ø)	⬆️
src/silicone/stats.py	100% <100%> (ø)
src/silicone/plotting.py	95.23% <90%> (+2.55%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 70f4448...eaeffbc. Read the comment docs.

[znicholls]

Looks good @Rlamboll. There's a few things I'd tweak so I'll make a few PRs so we can discuss!

[znicholls]

lgtm. Only last question. What should happen if your x and y are the same array ie you do the quantile analysis on two identical datasets? You won’t get a linear interpolation back out will you? Rather steps?

[Rlamboll]

It's in steps, yes, as it will only return values found in the y dataset. You can see the answer to this question if you've ever run the plotting.py function (/script that runs this) on CO2 vs CO2 with quantiles on - it's pretty linear and as expected in the center of the distribution for all reasonable quantiles, but at the edges of the data there's less data so quantiles are pulled in the direction of the average value and are notably noisy.

[znicholls]

It's in steps, yes, as it will only return values found in the y dataset. You can see the answer to this question if you've ever run the plotting.py function (/script that runs this) on CO2 vs CO2 with quantiles on - it's pretty linear and as expected in the center of the distribution for all reasonable quantiles, but at the edges of the data there's less data so quantiles are pulled in the direction of the average value and are notably noisy.

Is it worth adding a test for this? Or putting in one of the example notebooks? The reason I ask is that if you use the infiller to infill a timeseries you already have, you won't get back what you put in which can be confusing so it would be good to document why.

[Rlamboll]

If you have equally-spaced datapoints, match them up with the boxes and have very rapidly-decaying weights, the 0.5 quantile will be the original value. You can't guarantee it otherwise, no.

[znicholls]

Ok I think it would be good to explain that in a notebook then. We can either do that in this PR or we can open an issue and then do it in a later PR, I'm happy either way (and conscious of time).

[Rlamboll]

OK, I've added stuff to the notebook to demonstrate the quirks of the windows value and retooled a test to check that you get out what you put in when there's nothing to average over. I've also removed the warning message.

[znicholls]

@Rlamboll I also didn't understand the last part of the notebook. Can you please add some more explanatory plots (or we make an issue and deal with it later).

[znicholls]

@Rlamboll once we've dealt with #24 and the review comments let's merge above. We can leave the notebook for #25

[znicholls]

Looks good @Rlamboll. One suggested extra comment which I think would be helpful otherwise good to go (after rebasing)!