Facet wrapping

[nicolaskruchten]

No description provided.

[kmader]

Definitely interested in this feature, also having different scales for different facets. I didn't immediately see how to do this.

[nicolaskruchten]

Indeed, we don't have a free-scales feature at the moment, but we can look at adding it when we add facet-wrapping.

[sakethsaxena]

Hey!
I recently started using plotly express, just wondering if there is any momentum on this feature yet?

I was trying to generate a facetted scatter plot using a single dimension/column which has about 80-90 unique entries using facet_col in px.scatter() but it throws a value error.

[joelostblom]

I am also very interested in having the equivalent of seaborn's col_wrap parameter. In the meantime, is there any workaround for achieving this behavior with plotly express? I regularly need to plot > 10 subplots and don't have another variable to use for row facetting.

[chriddyp]

Could you introduce a new column("row_number") into your data frame that would be used to break apart into rows?

[joelostblom]

@chriddyp Thanks for the suggestion, that would work around the problem for my own visualization needs.

One of the reasons I am interested in plotly express is for teaching people new to programmatic data visualization. For this purpose it would be great if there is a wrapping parameter, since that would be easier to understand.

[nicolaskruchten]

@chriddyp's workaround is the best way forward right now, and we're very keen to add this to our library... Pull requests are welcome, of course, if someone wants to give it a shot before we get around to it :)

The API I would aim for is basically the same as Seaborn's, and the modifications would be in https://github.com/plotly/plotly.py/blob/master/packages/python/plotly/plotly/express/_core.py (basically wherever facet_row or facet_col is referenced we'd need an if block to handle the wrapping case).

We could also think about synthesizing a "row number" into the data frame higher up in the pipeline and reusing the existing logic.

[joelostblom]

Hmm, maybe I misunderstood the proposed workaround initially... I just tried adding row numbers to simulate col_wrap=3 and ended up with the plot below:

Initially, I thought that empty columns would be removed and wrapped around like in this mockup:

The current plotly express behavior makes more sense than what I was thinking, but I am suspecting this means I misunderstood what you meant when you proposed the workaround @chriddyp and @nicolaskruchten. In this example, how would you add row numbers to the "Sun, Sat, Thur, and Fri" observations to achieve the desired wrap around behavior?

[nicolaskruchten]

Yeah... This is going to be a pretty rough workaround until we implement this for real: you'd have to provide your own row and column, and on top of that you wouldn't get per-subplot titles, you'd be stuck with the per-column and per-row "row_number=1" type labelling, unless you added your own annotations.

[nicolaskruchten]

OK, I've got a working version of PX with a new facet_num_cols kwarg that results in this (and is incompatible with facet_row and marginal*... should be in the next version of plotly! Thanks for the nudge :)

[joelostblom]

Beautiful! Thanks for your responsiveness!