Gantt chart to scatterplots

[cs48a]

Gantt chart now uses scatter plots:

• updating all the functions accordingly
• unit tests updated
• new param: show_hover_fill for hiding the text for the filled area

I updated the Gantt chart to use scatter plots with "toself" fills to draw the recantgles, instead of shapes. This allows hiding and isolating traces.
I tested on all the examples and the results were visually the same as with the shapes.

[jonmmease]

Hi @csabaszan, this sounds like a great update. Thanks for taking it on. Don't worry about the *-plot_ly tests, these have been fixed in the v4_integration branch that will be merged into master shortly. It looks like there is something failing in the *-optional tests related to this PR, though.

[cs48a]

Yeah, first ever pull request... but I'm on it! Thanks for the feedback

[nicolaskruchten]

Thanks for the PR!

Can you tell us a bit more about the motivation for this change? I'd love some more insight into what is causing a problem with the existing figure factory that can't be solved by remaining within the bar paradigm? :)

[nicolaskruchten]

Sorry, I just looked into it a bit more and the current implementation isn't using bar, I just inferred from your description above :)

[cs48a]

@nicolaskruchten I was trying to toogle the traces and when they failed I found #1371 where @jonmmease mentioned that fill "toself" could be a way to go.

Now except for the *-plot_ly tests the tests are successful

[jonmmease]

Hi @csabaszan,

Now that we merged the v4_integration branch and release version 4, we'll need to update this branch to fix the merge conflicts.

Basically, all of the files you modified have moved under the packages/python/plotly directory. I don't think there were any changes in these files other than the move, so you should be able to overwrite the current files in the new locations with your latest versions.

If you run into trouble doing this as a merge resolution with git, feel free to start over and create a new pull request. Thanks!

[nicolaskruchten]

Having looked at this a bit more closely, I would actually recommend using bar traces rather than making rectangles with scatter traces, unless there's a compelling reason not to @jonmmease ?

[jonmmease]

Having looked at this a bit more closely, I would actually recommend using bar traces rather than making rectangles with scatter traces, unless there's a compelling reason not to @jonmmease ?

I honestly don't remember if there was a good reason why I mentioned scatter+fill in #1371 rather than bar. And you're right that bar would be more natural. The one factor that comes to mind is that both the shape and scatter+fill methods give full control over the width of the bars in data coordinates, but it's not clear to me that this would really be needed.

@csabaszan, do you see any reason to use scatter+fill rather than bar? For the bar approach, we would use bars in horizontal mode (https://plot.ly/python/horizontal-bar-charts/#basic-horizontal-bar-chart), specifying a custom base for each bar (https://plot.ly/python/bar-charts/#customizing-individual-bar-base).

[cs48a]

@csabaszan, do you see any reason to use scatter+fill rather than bar? For the bar approach, we would use bars in horizontal mode (https://plot.ly/python/horizontal-bar-charts/#basic-horizontal-bar-chart), specifying a custom base for each bar (https://plot.ly/python/bar-charts/#customizing-individual-bar-base).

@jonmmease I've just looked into bar and yes, I think with horizontal mode and using the base it can be done. I'll try do the switch and update the PR.

[cs48a]

btw what would be the best way to handle timedelta in the x-axis? I can set the type of the axis to date, but I don't know what should x take as an input. I was trying to use it with seconds and milliseconds but it was somewhat strange, because 23-hours is displayed as 1 day.

The code below produces the plot in the image. Resource 2 is shown to be occupied for 1 day, but according to the data it is only 23 hours

from plotly import graph_objs as go
fig = go.Figure()
fig.add_trace(go.Bar(
    y=["resource 1", "resource 2", "resource 3"],
    x=[3600*24*1000, 3600*23*1000, 3600*22*1000],
    base=["2016-01-01 6:00:00", "2016-01-01 6:00:00", "2016-01-01 6:00:00"],
    name='Job1',
    orientation='h',
    text=["a", "b", "c"],
))
fig.layout.barmode='overlay'
fig.layout.xaxis.type = "date"

What really interesting, though, is that if I set x to 0 then a 1-hour long bar is displayed, so could it be that somehow this 1 hour offset is always there?

[jonmmease]

Hi @csabaszan,

Yeah, that's interesting. It looks like there might be something timezone related creeping in when specifying the x value as a numeric value in milliseconds. Here's what I get for your same plot in GMT-4.

In any case, plotly.js doesn't have a timedelta axis type, and I don't think it will be possible to use the date axis type to display times larger that 24 hrs.

[cs48a]

Hi @jonmmease,

Do you think there is some workaround to use bar for Gantt charts? Displaying the length of the period in a readable format is pretty important for a Gantt, imo.

[jonmmease]

Hi @csabaszan,

Since the current API is to specify the start and end of each "bar" as a date, it might be better to stick with your current scatter+fill approach. Maybe this is what I was thinking when I brought up scatter in the first place 🙂

@nicolaskruchten, do you any thoughts on what might be going on with the bar lengths above?

[jonmmease]

Thanks @csabaszan, this looks great! It's really nice to have the legend control. In addition to the inline comments, there's one other change that would be nice to have.

Right now, when you click the legend to hide the bars, the hover points are still visible as small dots. I made some inline comments about making these invisible by setting opacity=0, but even with this change the tooltips will still appear when the bar is hidden. You can make these hover points disappear along with the bar by placing them in the same legendgroup as the associated scatter trace (See https://plot.ly/python/legend/#grouped-legend). To do this, set the legendgroup property of both the filled scatter trace, and the hover scatter trace to be equal to the trace's name.

thanks again!

[nicolaskruchten]

If you put multiple traces in the same legend group, it’s nice to have all but one per group not appear in the legend, that way it looks like just a single legend entry

[jonmmease]

Thanks for the last round of updates @csabaszan. Looks good to me!