tools.make_subplots with colspan/rowspan fails when using sharex/sharey

[claria]

Hi,
i am trying to create subplots with colspan/rowspan attribute and shared_xaxes using the code below (Plotly 3.3). As soon as i use the shared_xaxes=True parameter, the code fails with the error message:

Exception: Something went wrong. An axis object for (2,4) subplot cell got deleted.

Example code:

from plotly import tools
import plotly
import plotly.plotly as py
import plotly.graph_objs as go

trace1 = go.Scatter(
    x=[0, 1, 2],
    y=[10, 11, 12]
)
trace2 = go.Scatter(
    x=[2, 3, 4],
    y=[100, 110, 120],
)
trace3 = go.Scatter(
    x=[2, 3, 4],
    y=[100, 110, 120],
)

fig = tools.make_subplots(rows=3, cols=4, specs=[[{'colspan':3}, None, None, None], 
                                                [{'rowspan':2, 'colspan':3}, None, None, {'rowspan':2}], 
                                                [None, None, None, None]],
                          shared_xaxes=True, shared_yaxes=True,
                          vertical_spacing=0.001)

print(fig['layout'])

fig.append_trace(trace1, 1, 1)
fig.append_trace(trace2, 2, 1)
fig.append_trace(trace3, 2, 4)

print(fig['layout'])

fig['layout'].update(height=600, width=600, title='Stacked Subplots with Shared X-Axes')
plotly.offline.iplot(fig, filename='test')

Cheers,
Georg

[empet]

@claria, Remove shared_xaxes=True, because the trace1 and trace 2, that are aligned vertically, do not have the same x range, to share xaxes.
After this change no error is displayed, and you'll get the plot.

[claria]

@empet Aah sorry, i put this minimal example together too quickly. If you change the x axes to overlap, e.g. of all traces to x=[1,2,3], the error remains. And yes, this error is specific to shared_xaxes=True

[empet]

@claria The error persists, because the trace3 is not aligned vertically with no other trace to share xaxis.

[claria]

@empet Nope. Even adding a vertically aligned subplot+trace does not help:

from plotly import tools
import plotly
import plotly.plotly as py
import plotly.graph_objs as go
import plotly.io as pio


trace1 = go.Scatter(
    x=[0, 1, 2],
    y=[10, 11, 12]
)
trace2 = go.Scatter(
    x=[0, 1, 2],
    y=[100, 110, 120],
)
trace3 = go.Scatter(
    x=[0, 1, 2],
    y=[100, 110, 120],
)
trace4 = go.Scatter(
    x=[0, 1, 2],
    y=[100, 110, 120],
)
trace5 = go.Scatter(
    x=[0, 1, 2],
    y=[100, 110, 120],
)

fig = tools.make_subplots(rows=5, cols=4, specs=[[{'colspan':3}, None, None, None], 
                                                 [{'rowspan':2, 'colspan':3}, None, None, {'rowspan':2}], 
                                                 [None, None, None, None],
                                                 [{'rowspan':2, 'colspan':3}, None, None, {'rowspan':2}], 
                                                 [None, None, None, None],
                                                 ],
                          shared_xaxes=True, shared_yaxes=True,
                          vertical_spacing=0.001)

print(fig['layout'])

fig.append_trace(trace1, 1, 1)
fig.append_trace(trace2, 2, 1)
fig.append_trace(trace3, 2, 4)
fig.append_trace(trace4, 4, 1)
fig.append_trace(trace5, 4, 4)

print(fig['layout'])

fig['layout'].update(height=600, width=600, title='Stacked Subplots with Shared X-Axes')
plotly.offline.iplot(fig, filename='test')

[empet]

@claria, This case has the same drawback as the previous example:

I removed shared_xaxes=True to get this plot. The traces 3 and 5 have common x-range, but there is
an empty cell above them. shared_xaxes, respectively yaxes=True work only when on the corresponding column/row we have inserted traces in all cells, and moreover they have the same range.

[claria]

@empet I could accept this as the expected behaviour if shared_xaxes=[[1,1],[2,1],[4,1]] would work so that i could use shared_axes at least for the first column. But this gives the same error message. I would expect this to be a common use case.

[claria]

Hi @empet, in the plot you showed with shared_xaxes=False, the y axes are shared without a problem. Even more strange. I can even add a subplot+trace in the upper right corner and the error with shared_xaxes=True persists.

from plotly import tools
import plotly
import plotly.plotly as py
import plotly.graph_objs as go


trace1 = go.Scatter(
    x=[0, 1, 2],
    y=[10, 11, 12]
)
trace2 = go.Scatter(
    x=[0, 1, 2],
    y=[100, 110, 120],
)
trace3 = go.Scatter(
    x=[0, 1, 2],
    y=[100, 110, 120],
)
trace4 = go.Scatter(
    x=[0, 1, 2],
    y=[100, 110, 120],
)
trace5 = go.Scatter(
    x=[0, 1, 2],
    y=[100, 110, 120],
)
trace6 = go.Scatter(
    x=[0, 1, 2],
    y=[100, 110, 120],
)

fig = tools.make_subplots(rows=5, cols=4, specs=[[{'colspan':3}, None, None, {}], 
                                                 [{'rowspan':2, 'colspan':3}, None, None, {'rowspan':2}], 
                                                 [None, None, None, None],
                                                 [{'rowspan':2, 'colspan':3}, None, None, {'rowspan':2}], 
                                                 [None, None, None, None],
                                                 ],
                          shared_xaxes=True, shared_yaxes=True,
                          vertical_spacing=0.001)

print(fig['layout'])

fig.append_trace(trace1, 1, 1)
fig.append_trace(trace2, 2, 1)
fig.append_trace(trace3, 2, 4)
fig.append_trace(trace4, 1, 4)
fig.append_trace(trace5, 4, 1)
fig.append_trace(trace6, 4, 4)

print(fig['layout'])

fig['layout'].update(height=600, width=600, title='Stacked Subplots with Shared X-Axes')
plotly.offline.iplot(fig, filename='test')

[empet]

@claria, Due to special placement of cells, you should set shared_xaxes, and shared_yaxes as lists of lists, not boolean, as above:

shared_xaxes=[[(1,1), (2,1), (4,1)], [(1,4), (2,4), (4,4)]], 
shared_yaxes=[[(1,1), (1,4)], [(2,1), (2,4)], [(4,1), (4,4)]]

In trace1 I corrected y to ensure the same range as in all traces.
y=[100, 110, 120].

[claria]

@empet, why are the ticks in the first column below the first upper most subplot, in the second column below the lowest one?

Let's summarize the behaviour:

• If you use colspan/rowspan you cannot use shared_{x,y}axes=True , but have to specify the shared cols/rows.
• If you additionally have an empty cell somewhere, shared axes does not work at all.

[empet]

@claria

1. The ticks are put to the plot referenced to xaxis, yaxis, because
   as we can see from the printed Layout:

Layout({
    'xaxis': {'anchor': 'y', 'domain': [0.0, 0.74]},
    'xaxis4': {'anchor': 'free', 'domain': [0.78, 1.0], 'position': 0.0},
    'yaxis': {'anchor': 'x', 'domain': [0.832, 1.0]},
    'yaxis2': {'anchor': 'free', 'domain': [0.416, 0.792], 'position': 0.0},
    'yaxis4': {'anchor': 'free', 'domain': [0.0, 0.376], 'position': 0.0}
})

the cell (1,1) is referenced to the xaxis, and yaxis and both have set the anchor y, respectively x.

With these updates:

fig.layout.xaxis.anchor='free'
fig.layout.yaxis.anchor='free'

the ticks are put to the lowermost cell.
The shared_xaxes, and yaxes given as lists are not treated in detail https://github.com/plotly/plotly.py/blob/master/plotly/tools.py#L1078.
2) shared_xaxes, shared_yaxes given as lists or lists of lists of tuples work in the subplots with empty cells, too.

[claria]

@empet Thank you a lot for your advice. It is very much appreciated. Regarding 2) you are right, shared_axes given as lists work in subplots with empty cells.

Regarding the inconsistency of the axes, adding your fig.layout...='free' lines results in the following plot for me, which is clearly undesired:

[empet]

@claria What is your Plotly version? Here is a Jupyter Notebook https://plot.ly/~empet/14993 running Plotly 3.3.0, and it works with those fig.layout updates.

Please address new questions on Plotly forum https://community.plot.ly/.

[claria]

Ah i checked if the behaviour was the same in Plotly 2.7 so i switched back and forgot to upgrade again. Thanks a lot for the hint!

[jonmmease]

Thanks for working through this @empet. I haven't read through the thread in detail, is there anything to record before closing this issue?