`mpl_to_plotly` does not preserve axis labels (bar plots are useless)

[krassowski]

While mpl_to_plotly is little known and receives little love, this bug is pretty easy to fix. Would you be open to a PR?

Details

import matplotlib.pyplot as plt
from plotly.tools import mpl_to_plotly
from plotly.offline import plot

In matplotlib this produces a barplot with labels:

labels = ['a', 'b', 'c']
values = [1, 2, 3]

f = plt.figure(figsize=(6, 4))
plt.bar(labels, values)
plt.tight_layout()

But conversion to plotly looses the labels:

plotly_fig = mpl_to_plotly(f)
plot(plotly_fig)

A minimal fix would be to modify prep_ticks by appending:

    if axis_dict.get("type") == "date":
        return axis_dict
    
    vals = []
    texts = []
    for tick in axis.majorTicks:
        vals.append(tick.get_loc())
        texts.append(tick.label1.get_text())
    
    if texts:
        axis_dict = {}
        axis_dict['tickmode'] = 'array'
        axis_dict['tickvals'] = vals
        axis_dict['ticktext'] = texts
    
    return axis_dict

which produces:

prep_ticks is defined in:

plotly.py/plotly/matplotlylib/mpltools.py

Lines 428 to 514 in c54a2bd

	def prep_ticks(ax, index, ax_type, props):
	"""Prepare axis obj belonging to axes obj.
	positional arguments:
	ax - the mpl axes instance
	index - the index of the axis in `props`
	ax_type - 'x' or 'y' (for now)
	props - an mplexporter poperties dictionary
	"""
	axis_dict = dict()
	if ax_type == "x":
	axis = ax.get_xaxis()
	elif ax_type == "y":
	axis = ax.get_yaxis()
	else:
	return dict() # whoops!
	scale = props["axes"][index]["scale"]
	if scale == "linear":
	# get tick location information
	try:
	tickvalues = props["axes"][index]["tickvalues"]
	tick0 = tickvalues[0]
	dticks = [
	round(tickvalues[i] - tickvalues[i - 1], 12)
	for i in range(1, len(tickvalues) - 1)
	]
	if all([dticks[i] == dticks[i - 1] for i in range(1, len(dticks) - 1)]):
	dtick = tickvalues[1] - tickvalues[0]
	else:
	warnings.warn(
	"'linear' {0}-axis tick spacing not even, "
	"ignoring mpl tick formatting.".format(ax_type)
	)
	raise TypeError
	except (IndexError, TypeError):
	axis_dict["nticks"] = props["axes"][index]["nticks"]
	else:
	axis_dict["tick0"] = tick0
	axis_dict["dtick"] = dtick
	axis_dict["tickmode"] = None
	elif scale == "log":
	try:
	axis_dict["tick0"] = props["axes"][index]["tickvalues"][0]
	axis_dict["dtick"] = (
	props["axes"][index]["tickvalues"][1]
	- props["axes"][index]["tickvalues"][0]
	)
	axis_dict["tickmode"] = None
	except (IndexError, TypeError):
	axis_dict = dict(nticks=props["axes"][index]["nticks"])
	base = axis.get_transform().base
	if base == 10:
	if ax_type == "x":
	axis_dict["range"] = [
	math.log10(props["xlim"][0]),
	math.log10(props["xlim"][1]),
	]
	elif ax_type == "y":
	axis_dict["range"] = [
	math.log10(props["ylim"][0]),
	math.log10(props["ylim"][1]),
	]
	else:
	axis_dict = dict(range=None, type="linear")
	warnings.warn(
	"Converted non-base10 {0}-axis log scale to 'linear'" "".format(ax_type)
	)
	else:
	return dict()
	# get tick label formatting information
	formatter = axis.get_major_formatter().__class__.__name__
	if ax_type == "x" and "DateFormatter" in formatter:
	axis_dict["type"] = "date"
	try:
	axis_dict["tick0"] = mpl_dates_to_datestrings(axis_dict["tick0"], formatter)
	except KeyError:
	pass
	finally:
	axis_dict.pop("dtick", None)
	axis_dict.pop("tickmode", None)
	axis_dict["range"] = mpl_dates_to_datestrings(props["xlim"], formatter)
	if formatter == "LogFormatterMathtext":
	axis_dict["exponentformat"] = "e"
	return axis_dict

[gvwilson]

thanks very much @krassowski - yes, we would be very happy to get a PR. thanks - @gvwilson

[willtryagain]

Hi @krassowski,

Thanks for looking into this issue. I noticed you have a fix ready for this and wanted to check in on its status. Are you still actively working on it or do you have plans to submit it for review soon?

I'm asking because this bug is an upstream dependency blocker for me, where wandb relies on Plotly internally.

If you're no longer able to work on it, would you be open to me submitting a separate PR for a fix?

Let me know what works best. Thanks!

[krassowski]

@willtryagain absolutely please submit sorry I forgot to open the PR earlier!