`to_json` does not handle color as `np.nan` properly

[ChiQiao]

When creating a scatter plot with colors including np.nan the figure shows fine.

import json
import numpy as np
import plotly.graph_objects as go

go.Figure(go.Scatter(x=[0, 0], y=[0, 1], marker={"color": [0, np.nan]}))

However, if I convert the figure to JSON and load it back, there will be an error.

fig = go.Figure(go.Scatter(x=[0, 0], y=[0, 1], marker={"color": [0, np.nan]}))
go.Figure(**json.loads(fig.to_json()))

Error message:

ValueError: 
    Invalid element(s) received for the 'color' property of scatter.marker
        Invalid elements include: [None]

    The 'color' property is a color and may be specified as:
      - A hex string (e.g. '#ff0000')
      - An rgb/rgba string (e.g. 'rgb(255,0,0)')
      - An hsl/hsla string (e.g. 'hsl(0,100%,50%)')
      - An hsv/hsva string (e.g. 'hsv(0,100%,100%)')
      - A named CSS color:
            aliceblue, antiquewhite, aqua, aquamarine, azure,
            beige, bisque, black, blanchedalmond, blue,
            blueviolet, brown, burlywood, cadetblue,
            chartreuse, chocolate, coral, cornflowerblue,
            cornsilk, crimson, cyan, darkblue, darkcyan,
            darkgoldenrod, darkgray, darkgrey, darkgreen,
            darkkhaki, darkmagenta, darkolivegreen, darkorange,
            darkorchid, darkred, darksalmon, darkseagreen,
            darkslateblue, darkslategray, darkslategrey,
            darkturquoise, darkviolet, deeppink, deepskyblue,
            dimgray, dimgrey, dodgerblue, firebrick,
            floralwhite, forestgreen, fuchsia, gainsboro,
            ghostwhite, gold, goldenrod, gray, grey, green,
            greenyellow, honeydew, hotpink, indianred, indigo,
            ivory, khaki, lavender, lavenderblush, lawngreen,
            lemonchiffon, lightblue, lightcoral, lightcyan,
            lightgoldenrodyellow, lightgray, lightgrey,
            lightgreen, lightpink, lightsalmon, lightseagreen,
            lightskyblue, lightslategray, lightslategrey,
            lightsteelblue, lightyellow, lime, limegreen,
            linen, magenta, maroon, mediumaquamarine,
            mediumblue, mediumorchid, mediumpurple,
            mediumseagreen, mediumslateblue, mediumspringgreen,
            mediumturquoise, mediumvioletred, midnightblue,
            mintcream, mistyrose, moccasin, navajowhite, navy,
            oldlace, olive, olivedrab, orange, orangered,
            orchid, palegoldenrod, palegreen, paleturquoise,
            palevioletred, papayawhip, peachpuff, peru, pink,
            plum, powderblue, purple, red, rosybrown,
            royalblue, rebeccapurple, saddlebrown, salmon,
            sandybrown, seagreen, seashell, sienna, silver,
            skyblue, slateblue, slategray, slategrey, snow,
            springgreen, steelblue, tan, teal, thistle, tomato,
            turquoise, violet, wheat, white, whitesmoke,
            yellow, yellowgreen
      - A number that will be interpreted as a color
        according to scatter.marker.colorscale
      - A list or array of any of the above

Interestingly, the to_plotly_json method works fine:

fig = go.Figure(go.Scatter(x=[0, 0], y=[0, 1], marker={"color": [0, np.nan]}))
go.Figure(**json.loads(json.dumps(fig.to_plotly_json())))

Questions: Is this the expected behavior for the to_json method, and why it behaves differently from the to_plotly_json method?

[alexcjohnson]

Oh that's an interesting one. If you just json.dumps([0, np.nan]) you get '[0, NaN]' which is invalid JSON: in a browser JSON.parse('[0, NaN]') throws an error. The actual stringification we do fixes this incorrect behavior of the base json package and gives [0,null], which is valid JSON and is handled correctly by plotly.js (generally a null will just be ignored and we'll do whatever the default would have been). But when we bring 'null' back into Python we don't know that it came from np.nan, it could have also come from None so that's what we turn it into.

The problem is that np.nan is a number (ironic, given that NaN stands for not a number, but type(np.nan) == float) whereas None is not.

Anyone see a problem changing the number validator to always accept None? That's the only way I see to get around this.