feature request: z-ordering parameter for traces

[lucgiffon]

Hello,

it would be very nice to have the possibilty to set the z-ordering priority of traces in plotly. I think the way matplotlib handles it is pretty handy:

• default z-order value for a trace is 0;
• if multiple trace have the same z-order value, matplotlib decides the ordering of traces itself (random or whatever);
• if a trace have a greater z-order than an other, it appears on top of it;
• if a trace have a smaller z-order value than the other, it appears below;

Veeeery useful. Right now I am strugling to have a marker to appear on top of a bunch of line plots and it is very annoying.

[lucgiffon]

By the way, the hack I found to overcome this problem right now is to store the trace I want to add on TOP in a globalvariable (on can think of a list if one have multiple trace to add in a certain ordering) and then, just before rendering the figure, I had the trace to the figure. Then the last added trace appears on top of all other traces.

[huidongchen]

I am encountering the same issue. I would like to add several lines on top of scatter plot generated by px.scatter() .I tried to reorder the traces by doing fig.data = fig.data[::-1] but it only changes the order shown in legend, I still have not been able to add lines on top of points after trying almost all the tricks online, which is a bit frustrating..

it would be super helpful if plotly can support something like zorder

[tritemio]

Another use case for this feature is related to legend ordering. Currently, if you use the order of traces to specify the legend ordering, we cannot specify a z order independently. Since legend order is either None (i.e. insertion order) or alphabetical order, it is impossible to set an arbitrary order of legend items and at the same time set and arbitrary zorder of traces.

[dmalyuta]

Has there been any movement to try to implement this?

[nicolaskruchten]

A few notes on this:

1. This would have to be implemented in Plotly.js rather than in Python
2. Traces are currently drawn in batches by type, so scatter is always drawn on top of bar, etc, and this would be quite complicated to change, so the more straightforward option would be to control the other within types first

[nicolaskruchten]

I should add that fig.data = fig.data[::-1] does work for me:

import plotly.express as px
import plotly.graph_objects as go

fig = px.scatter(x=[1,2,3,4], y=[1,2,3,4], size=[1,2,3,4], color=["a","a","b","b"])
fig.add_trace(go.Scatter(x=[1,2,3,4], y=[1,2,3,4], mode="lines", line_color="black"))
fig.data = fig.data[::-1]
fig.show()

as does reordering fig.data like this:

import plotly.express as px
import plotly.graph_objects as go

fig = px.scatter(x=[1,2,3,4], y=[1,2,3,4], size=[1,2,3,4], color=["a","a","b","b"])
fig.add_trace(go.Scatter(x=[1,2,3,4], y=[1,2,3,4], mode="lines", line_color="black"))
fig.data = (fig.data[1],fig.data[2],fig.data[0])
fig.show()

[JoElfner]

Any news on this? This would be a great feature!

[jaymegordo]

+1, would love to be able to draw scatter traces BEHIND candlesticks for example. When trying to use hover label to inspect individual candles, other lines on top prevent label from showing.

[set92]

I should add that fig.data = fig.data[::-1] does work for me:

import plotly.express as px
import plotly.graph_objects as go

fig = px.scatter(x=[1,2,3,4], y=[1,2,3,4], size=[1,2,3,4], color=["a","a","b","b"])
fig.add_trace(go.Scatter(x=[1,2,3,4], y=[1,2,3,4], mode="lines", line_color="black"))
fig.data = fig.data[::-1]
fig.show()

as does reordering fig.data like this:

import plotly.express as px
import plotly.graph_objects as go

fig = px.scatter(x=[1,2,3,4], y=[1,2,3,4], size=[1,2,3,4], color=["a","a","b","b"])
fig.add_trace(go.Scatter(x=[1,2,3,4], y=[1,2,3,4], mode="lines", line_color="black"))
fig.data = (fig.data[1],fig.data[2],fig.data[0])
fig.show()

Seems this trick doesn't work when there is too much data. Example:

import plotly.express as px
import plotly.graph_objects as go
import numpy as np

df = np.random.randint(0,100,size=(10000, 2))
color_df = np.random.randint(0,4,size=(10000, 1)).T[0]
centers = np.random.randint(0,100,size=(7, 2))

fig = px.scatter(x=df[:, 0], y=df[:, 1], color=color_df)
fig.add_scatter(
    x=centers[:, 0],
    y=centers[:, 1],
    marker=dict(size=20, color="LightSeaGreen"),
    name="Centers",
    mode="markers",
)
fig.data = (fig.data[1], fig.data[0])
fig.show()

I tried removing the optional parameters and use only X,Y but it keeps putting the centers below the data.

[ptim]

Seems this trick doesn't work when there is too much data.

I'm having a similar issue with a Candlestick trace always rendering in the BG, but in my case, I see the the legend re-ordered, but the graph maintains the old z ordering (candles are obscured by the scatter I'm trying to put in the BG).

[benlindsay]

I should add that fig.data = fig.data[::-1] does work for me:

This does allow you to rearrange the order of the traces, but it rearranges it both in the legend and in the plot itself. The layering of the traces still seems to be inextricably linked to the order of the legend, unless there's something I'm missing. Often, I want less important traces to be at the bottom of the legend AND at the bottom layer on the plot. I don't see a way to do this currently. +1 for z-ordering parameter suggestion :)

(BTW, I love plotly and use it all the time. Thanks for maintaining such an awesome library!)

[fleimgruber]

@set92, @ptim just a quick FYI as I got a similar problem (with "too much" data) and got it working following your lead. The hint was in https://github.com/plotly/plotly.py/issues/2345#issuecomment-671461207, i.e. the ordering is done per type. In your example in https://github.com/plotly/plotly.py/issues/2345#issuecomment-778677610 the px.scatter is implicitly converted to a Scattergl but the add_scatter only adds a Scatter object. Now when the graph object for add_trace has the same resulting type (Scattergl), this works for me:

import plotly.express as px
import plotly.graph_objects as go
import numpy as np

df = np.random.randint(0, 100, size=(10000, 2))
color_df = np.random.randint(0, 4, size=(10000, 1)).T[0]
centers = np.random.randint(0, 100, size=(7, 2))

fig = px.scatter(x=df[:, 0], y=df[:, 1], color=color_df)
fig.add_trace(
    go.Scattergl(
        x=centers[:, 0],
        y=centers[:, 1],
        marker=dict(size=20, color="LightSeaGreen"),
        name="Centers",
        mode="markers",
    )
)
fig.show()

[nicolaskruchten]

Please note that the idea of controlling the legend order is tracked in https://github.com/plotly/plotly.py/issues/2345, and is in principle distinct from the drawing order.

[nicolaskruchten]

Update on legend ordering independently of trace ordering in data: this is now possible with the new legendrank attribute in v5.0 :) https://plotly.com/python/legend/#legend-order

[ievgennaida]

@nicolaskruchten Thank you! Works fine even with the subplots.

Is it possible also to specify this for the shapes? I would like to keep sibling shapes in the same z-order.

Is it affecting traces z-order or only legends order in the legends panel?

[nicolaskruchten]

legendrank only impacts the legend, not the z-order, and shapes aren't linked to the legend, so this approach will not impact shapes, no.

[ievgennaida]

@nicolaskruchten Thank you, it's clear now, can it be also mentioned in the example documentation that it's about order, not z-order of the traces?

Basically, my problem was that when you have subplots it's hard to control order of the legends sorted. Ex:

• SubPlot 1 [legend 14, legend 15]
• SubPlot 2 [legend 1, legend 2, legend 3, legend 4]

Expected legend: [legend 1, legend 2, legend 3, legend 4, legend 14, legend 15]

But in reality it was taken grouped from subplots.

This sorting problem is solved by legendrank.

But also sorting collerates with order for me: I am sorting legend by importance and it's expected that legends on top should be rendered top most (same is applicable for the related shapes, but less critical).

Actually to achieve the top most (legend on right) I have reverted order of the traces before appending them and than reverted back by the plotly property 'revert legends order'.

Now I can append in any order to control z and rank legends to keep order.

[maegul]

Related to this, not having a z-order for a ridgeline plot caused me some trouble.

I wanted the bottom traces to be drawn on top. But it seems that both the stacking order for violin plots and the (implicit) z-order are both bound to the fig.data ordering ... it isn't possible without some unintuitive hacking (see my example result at the bottom)

See my comment on the ridgeline plots issue: https://github.com/plotly/plotly_express/issues/25#issuecomment-886375427

See my post in the community forum: https://community.plotly.com/t/ridgeline-joy-plot-without-transparency/55028?u=maegul

[DJM-GM]

px.scatter is implicitly converted to a Scattergl but the add_scatter only adds a Scatter object. Now when the graph object for add_trace has the same resulting type (Scattergl)

And it works for me also.

So pleased that fleimgruber found and offered the work around that, after 'wasting' a day trying to find a solution, I signed up just to say 'Thank You, fleimgruber'.

[MichauGabriel]

For the Scattergl trick, I also had the problem that px.scatter makes either Scattergl or Scatter traces (depending on the number of points to plot?). But one can force webgl in all cases with 'render_mode="webgl"', so that the trick of re-ordering fig.data always works.

[florian6973]

Would be a very nice and useful feature

[vuk119]

Why is it still not implemented after 3 years?!

[janosh]

Why is it still not implemented after 3 years?!

Nobody is being paid to do this. it's not happening unless you or someone else decides to donate their time.

[MikePreston]

Does someone fancy donating their time to this issue? - for the forwarding and advancement of humankind

[frozensimo]

Up! This feature would be ace to have!

[aleferna12]

Life changing feature, would love to see it implemented in the future!

[devamin]

push push

[aymeric-roucher]

Same here, this would be useful!

[beevabeeva]

imperative feature. It should also be possible to alter the order in the rendered plot, by dragging and dropping the key items into the desired order.

[ayjayt]

looks like someone is working on it