Subplots with plotly express

[cyonghui81]

Can i do subplots, where each subplot is a time series chart using plotly express ?

[nicolaskruchten]

Yes, you can use facet_row and facet_col on any 2d-cartesian plot, including those with date axes.

[briangottfried]

@nicolaskruchten, is it possible to make subplots where you combine two charts created with plotly express?
Something like this:

[nicolaskruchten]

@briangottfried there's no easy way to compose like this at the moment but we're working on it! See this issue: plotly/plotly.py#2647

[floriangehrig]

Is there any update on this yet?

[nicolaskruchten]

There's no news on composing PX charts like this, but Plotly.py v4 came out this week, and it features a new and much-easier-to-use subplots system! Check it out here https://plot.ly/python/subplots/

Also, as of this week, Plotly Express is now part of Plotly.py (accessible as plotly.express) and is interoperable with the make_subplots function described in the link above.

[AliceMarb]

Is it possible to use make_subplots for maps (scatter geos) in plotly express? I currently have a plotly express map with a slider for dates, but would rather have multiple subplots of the different dates.

[nicolaskruchten]

You can use make_subplots on maps but Plotly Express doesn't support facet_row or facet_col on map-like traces just yet. Coming soon though!

See this issue: plotly/plotly.py#2646

[AliceMarb]

Wait sorry could you clarify for me: If I have a plotly express map like:
fig = px.scatter_geo(df, lat="lat", lon="lon", color="count", hover_name="text", hover_data=None,
size="count", animation_frame="dec", projection="natural earth", size_max = 50)

how do I put a few of them using make_subplots?
I got an error when I tried to.

Thanks for your help so far!

[nicolaskruchten]

Ah sorry: today you can make a figure with a single map with px.scatter_geo or a figure with multiple maps with go.make_subplots but you can't use them together.

In the future, you'll be able to say px.scatter_geo(facet_row=, facet_col=) but that's not supported yet. See this issue: plotly/plotly.py#2646

[Maxpyt]

There's no news on composing PX charts like this, but Plotly.py v4 came out this week, and it features a new and much-easier-to-use subplots system! Check it out here https://plot.ly/python/subplots/

Also, as of this week, Plotly Express is now part of Plotly.py (accessible as plotly.express) and is interoperable with the make_subplots function described in the link above.

That sounds great! Could you explain how this interoperability works?
When using

fig = make_subplots(rows=2, cols=2)
fig.add_trace(plotly.express.scatter(...),row=1,col=1)

it still throws value errors.
Thanks!

[nicolaskruchten]

Yes, the interoperability actually goes the other way: facetted figures created with px can be edited with fig.add_trace(row=1, col=1).

px.scatter() returns a full figure rather than a trace, so you cannot pass its output to add_trace() on a figure created externally to px.

As I said, we're working on a solution for this kind of thing but we're not there yet :)

See also Composition and Overlaying issues

[kozmers]

Without concerning the right color bar, the scatters can be realized by adding the fist trace of every figure to subplots: add_trace(fig['data'][0]).

fig1 = px.scatter(color='colname1')
fig2 = px.scatter(color='colname2')
trace1 = fig1['data'][0]
trace2 = fig2['data'][0]

fig = make_subplots(rows=2, cols=1, shared_xaxes=False)
fig.add_trace(trace1, row=1, col=1)
fig.add_trace(trace2, row=2, col=1)

[vanshika97]

Yes, you can use facet_row and facet_col on any 2d-cartesian plot, including those with date axes.

How to create subplots in px.bar_polar() where facet_row and facet_col parameters don't exist?

[nicolaskruchten]

@vanshika97 we don't yet support facetting for non-cartesian (i.e. polar) subplots. See this tracking issue: plotly/plotly.py#2646

[mattou78400]

Hey all,

It's been a while, I was wondering whether any fix/update has been found for that?

Thanks!

[xhuyvn]

Hi all,

Does can I make subplot with plotly express sunburst? It's been a while, but topic did not update solution.
Thanks.

[nicolaskruchten]

@xhuyvn at the moment we still only support faceting on 2d cartesian functions like scatter, bar, line etc... see this tracking issue: plotly/plotly.py#2646

[bharathngowda]

How to make n number of subplots in plotly?
I can achieve the same in Matplotlib as seen in the below example.

In the below example as the length of list 'x' changes the subplots are plotted. We don't define the row and column number for each subplot. Its plotted automatically in the order of the list elements. We define only total rows and cols for the figure only.

import matplotlib.pyplot as plt

x=['Plotly','Express','Dash','Python']

fig,ax=plt.subplots(nrows=int(len(x)/2), ncols=2,figsize=(15,7))

for i,j in zip(ax.flat,x):
i.bar([x for x in range(5)],[x for x in range(0,100,20)])
i.set_title(j)
fig.tight_layout()
plt.show()

Image as below

[set92]

Without concerning the right color bar, the scatters can be realized by adding the fist trace of every figure to subplots: add_trace(fig['data'][0]).

fig1 = px.scatter(color='colname1')
fig2 = px.scatter(color='colname2')
trace1 = fig1['data'][0]
trace2 = fig2['data'][0]

fig = make_subplots(rows=2, cols=1, shared_xaxes=False)
fig.add_trace(trace1, row=1, col=1)
fig.add_trace(trace2, row=2, col=1)

With his help I wrote a method to automate plotting boxplot and violin, for histogram it should be easy but there is some bug.

def facetting_data(df, n_cols=3, to_plot='box'):
    numeric_cols = df.select_dtypes('number').columns
    n_rows = -(-len(numeric_cols) // n_cols)  # math.ceil in a fast way, without import
    row_pos, col_pos = 1, 0
    fig = make_subplots(rows=n_rows, cols=n_cols)

    for col in numeric_cols:
        # trace extracted from the fig
        trace = getattr(px, to_plot)(df[col])["data"][0]
        # auto selecting a position of the grid
        if col_pos == n_cols: row_pos += 1
        col_pos = col_pos + 1 if (col_pos < n_cols) else 1
        # adding trace to the grid
        fig.add_trace(trace, row=row_pos, col=col_pos)
    return fig

fig = facetting_data(data, to_plot='violin')
# final tweaks
fig.update_layout(width=1000, height=800, title='Violin per feature', title_x=0.5)
fig.show()

EDIT: For histogram you can use fig = facetting_data(data.select_dtypes(np.float64), to_plot='histogram'), at least with my dataset was failing because a couple int columns (Happiness Rank and Year). Not sure why, but with them it draw the first trace and after that the rest of the traces were almost static.

[nicolaskruchten]

@set92 it looks like this facetting_data function does the same thing as the built-in facet_row and facet_col arguments of functions like px.box() and px.histogram() and px.violin() to me, no?

In general, folks, I promise I'll post an update on this issue when Plotly Express supports faceting for non-cartesian subplots. We're not working on it in the next couple of months, but we would happily accept a pull request if someone wants to work on it... The main bit of the code that needs modifying is here: https://github.com/plotly/plotly.py/blob/fb84bc52b9f4c424e93402ebfbe7b23502a601f4/packages/python/plotly/plotly/express/_core.py#L2047

[nicolaskruchten]

@bharathngowda please check out the facet_col_wrap option documented here: https://plotly.com/python/facet-plots/#wrapping-column-facets

[set92]

@nicolaskruchten I think they don't. For what I saw facet_row/facet_col is to provide a new dimension to divide the data no? The facetting_data is much simpler, I made it because I was trying to do an histogram/boxplot.. per feature. If I do it without arguments it will plot all the boxplots in the same graph, which makes hard to see some statistics right away.

For example, the first image is with a direct call, and the second is using the make_subplots function to create a individual graph per feature.

Plotly call	facetting_data function

Reproducible example with scatter+regression plot, since is a little bit different than previous function:
Dataset is world happiness (42kb), same as in kaggle, but resized.

import plotly.express as px
df = pd.read_csv('World_Happiness_2015_2017_.csv')
px.scatter(df, x=df.select_dtypes(np.float64).columns, y='Happiness Score', trendline='ols')

And with the function, that updated to work with 2 traces per graph is:

def facetting_scatter_plot(df, n_cols=3, y='Happiness Score'):
    numeric_cols = df.select_dtypes('number').columns
    n_rows = -(-len(numeric_cols) // n_cols)  # math.ceil in a fast way, without import
    row_pos, col_pos = 1, 0
    fig = make_subplots(rows=n_rows, cols=n_cols, subplot_titles=numeric_cols)

    for col in numeric_cols:
        # trace extracted from the fig
        trace = px.scatter(df, x=col, y=y, trendline='ols', trendline_color_override='#DC143C')["data"]
        # auto selecting a position of the grid
        if col_pos == n_cols: row_pos += 1
        col_pos = col_pos + 1 if (col_pos < n_cols) else 1
        # adding trace to the grid
        fig.add_trace(trace[0], row=row_pos, col=col_pos)
        fig.add_trace(trace[1], row=row_pos, col=col_pos)
    return fig

fig = facetting_scatter_plot(data.select_dtypes(np.float64))
# final tweaks
fig.update_layout(width=1000, height=800, title='% of variation in Happiness Score explained by each feature', title_x=0.5)
fig.show()

And it returns:

Probably it doesn't make much sense since you can zoom in, resize... but with boxplot happens the same as with histogram, each feature is too different from the rest so they fit better in individual graphs, and if I do a call per feature they would take a larger space.

[nicolaskruchten]

You can do almost the same thing with

px.scatter(df, x=df.select_dtypes(np.float64).columns, y='Happiness Score', trendline='ols',
          facet_col="variable", facet_col_wrap=3).update_xaxes(matches=None)

[nicolaskruchten]

This issue is kind of branching off into various directions so I've created a few different issues to track different parts of it, as well as worked to clarify our documentation (links below) so I'm going to close it and lock it now.

Tracking issues:

• Non-cartesian faceting: enable non-cartesian facets in Plotly Express: polar, ternary, maps etc plotly.py#2646
• Figure composition (side by side, vertical): enable plotly express figure composition plotly.py#2647
• Figure overlay (on top of): PX figure overlay / layering plotly.py#2648

Documentation:

• Plotly Express facets
• Plotly Express marginal plots
• Arbitrary subplots
• Mixed subplots
• Mixing trace types in a single plot
• The Figure Data Structure, where all this stuff is configured