Add multiseries support to graph_prediction_vs_actual_over_time

[MichaelFu512]

Pull Request Description

Closes #4285

A small snippet of what graph_prediction_vs_actual_over_time looks like for multiseries:

[codecov]

Codecov Report

Merging #4284 (92e1b64) into main (53bd61b) will increase coverage by 0.1%.
The diff coverage is 100.0%.

@@           Coverage Diff           @@
##            main   #4284     +/-   ##
=======================================
+ Coverage   99.7%   99.7%   +0.1%     
=======================================
  Files        355     355             
  Lines      39096   39154     +58     
=======================================
+ Hits       38976   39034     +58     
  Misses       120     120             

Files Changed	Coverage Δ
evalml/model_understanding/visualizations.py	100.0% <100.0%> (ø)
...s/model_understanding_tests/test_visualizations.py	100.0% <100.0%> (ø)

[jeremyliweishih]

One small testing change but LGTM otherwise.

[jeremyliweishih]

is there any autosizing we can take advantage of? Also not sure if we need the single_series case: can it just match what we already had before for time series?

[MichaelFu512]

Is there any autosizing we can take advantage of?

• For python plotly, it doesn't look like there's any autosizing besides doing what I did in line 490 (though correct me if I'm wrong). If I left the parameters blank for the multiseries graph (i.e. don't define x and y) and let it use the default values the graph looks really squished.

• For single series though I can probably leave the parameters blank since it is only one graph.

Also not sure if we need the single_series case: can it just match what we already had before for time series?

• The issue is that for single_series case I want the data that corresponds to that single series.
  • In the before case for time series, y is taken from data["target"] whereas for single series I want to take it from curr_df["target"] where curr_df = data[data["series_id"] == single_series]
  • If I wanted to do that with the before code, I'd probably have to add more if statements and such so it made more sense to me to have it included with the multiseries case.
  • Also I'd like for the single series plot to have a different title than the time series plot.

[jeremyliweishih]

cool - thanks for the explanation. Works for me!

[jeremyliweishih]

LGTM

[eccabay]

Thanks! Just a docstring comment