copy_to_readonly_numpy_array needlessly copies pandas series objects

[RZachLamberty]

the function _plotly_utils.basevalidators.copy_to_readonly_numpy_array performs a full copy of pd.Series objects which contain existing np.ndarray data as the values attribute. we could utilize the values attribute to dramatically speed up trace generation, especially for large dataframes.

environment: plotly version 3.1.0. macos high sierra 10.13.6. plotly installation via conda

working example:

import plotly
import _plotly_utils.basevalidators
import numpy

print('plotly version: {}'.format(plotly.__version__))

df = pd.DataFrame({'x': np.random.randint(0, 100, 1000000)})

# using `ipython` time magic
print('\ncoercing series')
%time v1 = _plotly_utils.basevalidators.copy_to_readonly_numpy_array(df.x)

print('\naccessing np values directly')
%time v2 = _plotly_utils.basevalidators.copy_to_readonly_numpy_array(df.x.values)

example output:

plotly version: 3.1.0

coercing series
CPU times: user 987 ms, sys: 35.5 ms, total: 1.02 s
Wall time: 854 ms

accessing np values directly
CPU times: user 1.45 ms, sys: 17 µs, total: 1.46 ms
Wall time: 1.49 ms

so a performance difference of approx 1000x

[RZachLamberty]

I may work on a pull request for this today, but I believe this can be directly addressed by adding an extra elseif level checking if the type of v is pd.Series and if the type of v.values is np.ndarray (it could be pd.Categorical, e.g.)

[jonmmease]

Awesome, thanks for digging into this. A PR would be very welcome!

[jonmmease]

Done in #1149