New plot generated when plot is created within function.

[melomcr]

Dear All,
I'm trying to update a single plot (or re-generate the same plot) depending on a value given by an ipywidget, but I get new plots when the ggplot call is inside a function call.
In my particular case, my "ggplot" call is inside a function called by an "observe" method attached to a widget. When the following example code is ran within a jupyter notebook, new plots are generated and appended to the notebook output every time the widget changes values:

from plotnine import *
from plotnine.data import *
from ipywidgets import widgets
from IPython.display import display

%matplotlib notebook

playSlider = widgets.IntSlider(min=0, max=5)

def createPlot(*args):
    p = ggplot(mtcars, aes(x='hp', y='wt', color='mpg')) + geom_point() +facet_wrap("~cyl") + theme_bw()
    p.draw()

playSlider.observe(createPlot, 'value')

display(playSlider)

Using %matplotlib notebook or %matplotlib inline makes no difference, new plot are appended to the cell's output.

I did find the "_draw_using_figure" and "_create_figure" methods in the code, but my attempts to force new plots to be created in the same matplotlib figure were unsuccessful.

Is this a problem in the implementation or just me missing the correct syntax?

[melomcr]

[SOLVED]
Using some code form another question, and some more reading, I understood the correct syntax to update the same image.
However, iterating over datasets I ended up "accumulating" points on the plot, instead of switching between points. By adding a little loop that removed artists, that problem was solved. I am attaching the code here for future reference, if someone is trying something similar in jupyter notebooks.

%matplotlib notebook

playString  = widgets.Play(
    value=4,
    min=4,
    max=8,
    step=2,
    description="Iterate over the string optimization",
)

playStringSlider = widgets.IntSlider(min=4, max=8, step=2)
widgets.jslink((playString,'value'),(playStringSlider,'value'))

stringCtl = widgets.HBox([playString, playStringSlider])

fig = None
axs = None
plot = None

def plotString(*args):
    
    global fig, axs, plot
    
    newValue = float(args[0]['new'])
    
    p = p9.ggplot() + p9.geom_point(mtcars.loc[mtcars.cyl == newValue],\
                                    p9.aes(x='wt', y='hp', color='mpg')) + \
                   p9.geom_path(mtcars.loc[mtcars.cyl == newValue],\
                                   p9.aes(x='wt', y='hp', color='mpg')) + \
                   p9.theme_bw() + p9.xlim([1,6]) + p9.ylim([0,400])
            
    if fig is None:
        fig, plot = p.draw(return_ggplot=True)
        axs = plot.axs
    else:
        p = copy(p)
        for artist in plt.gca().lines +\
                        plt.gca().collections +\
                        plt.gca().artists + plt.gca().patches + plt.gca().texts:
            artist.remove()
        p._draw_using_figure(fig, axs)
    
playStringSlider.observe(plotString, 'value')

display(stringCtl)

playStringSlider.value = 6

I would be glad to receive feedback on this approach.

[has2k1]

@melomcr, thanks for posting the solution. I am yet to make sense of it, but I think some form of this need to go into the documentation. Maybe even make it easier by providing the wrapper code if possible.

[melomcr]

I would be glad to help. I am a heavy user of ggplot in R, and your implementation for python made my day. I am switching my notebooks to use it, but as of now I have a lot of one-liners and notebook cells in R just to do ggplots. I can't stand matplotlib...

Would you @has2k1 be interested in a notebook using this example, and maybe a couple other widgets? I can make a pull request for the plotnine-examples folder.

[has2k1]

That is great. More examples may help create a helper function/class in plotnine; need to minimise the necessity of hidden methods like _draw_using_figure in user code.

[khaeru]

Could maybe draw take an optional kwarg figure?

If it exists, first scrub the artists as in melomcr's comment, and then _draw_using_figure. If not, then _draw on a fresh figure/axes.

(Also, _draw_using_figure and _draw are quite similar, and seem like they could be merged without becoming too complex.)

[has2k1]

@khaeru, Yes they are quite similar and could do with a refactoring.