NLU comparison does not plot results with more than one run

[TEGELB]

Rasa version: 1.8.1

Python version: 3.7.5

Operating system (windows, osx, ...): Windows

Issue:
When running the comparison of different pipeline configurations with default settings out of a folder, as laid out here, the diagramm is not plotted and an error occurs.
Error occurs when comparing two as well as six configurations.
All files are saved correctly up to that point.

No error occurs when the amount of runs is set to one manually.

Error (including full traceback):

![grafik](https://user-images.githubusercontent.com/56721736/76779152-2ec5b980-67ab-11ea-92fb-73765fbdaefc.png)

Command or request that led to error:

rasa test nlu

[chkoss]

I already traced down that the following happens when running e.g. rasa test nlu --runs 2.

Here, number_of_examples is plotted vs. mean, where mean is an average over runs of what's in the file output:
https://github.com/RasaHQ/rasa/blob/master/rasa/test.py#L219

The problem is, the shapes don't match. Let’s call the value of --runs r and the number of percentages in --percentages p. Then number_of_examples has shape (r*p,), the content of output has shape (r, p) and mean therefore has shape (p,).

This is caused by how number_of_examples is determined in compare_nlu:
https://github.com/RasaHQ/rasa/blob/master/rasa/nlu/test.py#L1413
This function simply appends all the values for each run and percentage into one 1D array of shape (r*p,).

[indam23]

Just ran into this too - it works in 1.7, in case that helps

[m-vdb]

I was able to reproduce the issue using the doc here: https://rasa.com/docs/rasa/user-guide/testing-your-assistant/#comparing-nlu-pipelines and the right configuration files.