Added handling for zero data and zero range

[jbednar]

Previously, Datashader would return ZeroDivision or other errors in all of these cases:

• No datapoints (common when interactively zooming into a small region, so should be valid)
• A single datapoint (which caused the x and y ranges to be infinitely small)
• Multiple datapoints sharing a single x or y value (which caused either the x or y ranges to be infinitely small)

The case with two distinct points has always worked, but the rest give errors:

import pandas as pd, datashader as ds, datashader.transfer_functions as tf
cvs = ds.Canvas(plot_width=5, plot_height=5)
df = pd.DataFrame(dict(x=[1,2], y=[-2,-5]))
print(str(cvs.points(df,'x','y').values))

[[0 0 0 0 1]
 [0 0 0 0 0]
 [0 0 0 0 0]
 [0 0 0 0 0]
 [1 0 0 0 0]]

No data points:

df = pd.DataFrame(dict(x=[], y=[]))
print(str(cvs.points(df,'x','y').values))

ValueError: No non-NaN x coordinates found.

A single data point:

df = pd.DataFrame(dict(x=[1], y=[-2]))
print(str(cvs.points(df,'x','y').values))

ZeroDivisionError: float division by zero

Two points, but only 1 x value:

df = pd.DataFrame(dict(x=[1,1], y=[-2,5]))
print(str(cvs.points(df,'x','y').values))

ZeroDivisionError: float division by zero

Two points, but only 1 y value:

df = pd.DataFrame(dict(x=[1,2], y=[-2,-2]))
print(str(cvs.points(df,'x','y').values))

ZeroDivisionError: float division by zero

Float array with one data point:

df = pd.DataFrame(dict(x=[1], y=[2]))
print(str(cvs.points(df,'x','y', agg=ds.sum('x')).values))

ZeroDivisionError: float division by zero

[jbednar]

After this PR is applied, all these cases are considered legal:

import pandas as pd, datashader as ds, datashader.transfer_functions as tf
cvs = ds.Canvas(plot_width=5, plot_height=5)
df = pd.DataFrame(dict(x=[1,2], y=[-2,-5]))
print(str(cvs.points(df,'x','y').values))

[[0 0 0 0 1]
 [0 0 0 0 0]
 [0 0 0 0 0]
 [0 0 0 0 0]
 [1 0 0 0 0]]

No data points:

df = pd.DataFrame(dict(x=[], y=[]))
print(str(cvs.points(df,'x','y').values))

No x values; defaulting to range -0.5,0.5
No y values; defaulting to range -0.5,0.5
[[0 0 0 0 0]
 [0 0 0 0 0]
 [0 0 0 0 0]
 [0 0 0 0 0]
 [0 0 0 0 0]]

A single data point:

df = pd.DataFrame(dict(x=[1], y=[-2]))
print(str(cvs.points(df,'x','y').values))

No x range; defaulting to x-0.5,x+0.5
No y range; defaulting to y-0.5,y+0.5
[[0 0 0 0 0]
 [0 0 0 0 0]
 [0 0 1 0 0]
 [0 0 0 0 0]
 [0 0 0 0 0]]

Two points, but only 1 x value:

df = pd.DataFrame(dict(x=[1,1], y=[-2,5]))
print(str(cvs.points(df,'x','y').values))

No x range; defaulting to x-0.5,x+0.5
[[0 0 1 0 0]
 [0 0 0 0 0]
 [0 0 0 0 0]
 [0 0 0 0 0]
 [0 0 1 0 0]]

Two points, but only 1 y value:

df = pd.DataFrame(dict(x=[1,2], y=[-2,-2]))
print(str(cvs.points(df,'x','y').values))

No y range; defaulting to y-0.5,y+0.5
[[0 0 0 0 0]
 [0 0 0 0 0]
 [1 0 0 0 1]
 [0 0 0 0 0]
 [0 0 0 0 0]]

Float array with one data point:

df = pd.DataFrame(dict(x=[1], y=[2]))
print(str(cvs.points(df,'x','y', agg=ds.sum('x')).values))

No x range; defaulting to x-0.5,x+0.5
No y range; defaulting to y-0.5,y+0.5
[[nan nan nan nan nan]
 [nan nan nan nan nan]
 [nan nan  1. nan nan]
 [nan nan nan nan nan]
 [nan nan nan nan nan]]

For now I've left in the "No ..." print statements, to show that it's working, but as none of these are meant to be error conditions, I'll remove the printing in the next commit.