Create a sibling to pyplot.plot that garuntees the creation of a new figure

[ianhi]

Problem

pyplot.plot will only create a new figure if there are no active figures. Thus when I am in a jupyter notebook with the ipympl backend I often write the following snippet:

plt.figure()
plt.plot(...)

There are two issues, one of convenience and one of user confusion:

1. I'd prefer to not have to write plt.figure every time I want a new figure to show up.
2. The current situation can be confusing to new users see %matplotlib ipympl does not allow cell "re-plot" ? ipympl#248
   • In particular this breaks the expectation established by usage of the inline backend that plt.plot in a cell will always create a new figure

My examples are both jupyter centric as that is my primary environment, but it feels as though the proposed solutions would belong in matplotlib proper rather than a backend.

Proposed Solution

Create two new functions in pyplot: plot_figure and plot_subplots that are guaranteed to create a new figure and otherwise act like pyplot.plot. Below are two incomplete implementations that I have been using.

def plot_figure(*args, **kwargs):
    """
    Calls plt.figure() for you, and then acts like plt.plot
    """
    if 'figsize' in kwargs:
        fig = figure(kwargs['figsize'])
        del kwargs['figsize']
    else:
        fig = figure()
    lines = plt.plot(*args, **kwargs)
    return fig, lines

def plot_subplots(*args, **kwargs):
    """
    Calls plt.subplots() for you, and then acts like plt.plot
    """
    if 'figsize' in kwargs:
        fig, ax = plt.subplots(figsize=kwargs['figsize'])
        del kwargs['figsize']
    else:
        fig, ax = plt.subplots()
    lines = plt.plot(*args, **kwargs)
    return fig, ax, lines

Additional context and prior art

Something that would be nice but probably is not technically feasible is a matplotlib figure that is associated with a specific juptyer cell. The proposed solution approximates such a figure per cell behavior. I think this desire is residual from my having "grown up" as it were using the inline backend. There are now features that I want from the interactive backend, but old habits like expecting plt.plot to just create a figure die hard.

[QuLogic]

This special cases plot; what about hist or imshow or anything else? And if you have to learn a second copy of every function, why not use plt.figure? In short, I don't think this is all that much more convenient, unless you only plot single lines.

[timhoffm]

"Create a new figure when using a plotting command" is what the hold(False) machinery was for. That state would be preferable to having separate versions of every function. But hold overall was removed in #3070.

pyplot.plot will only create a new figure if there are no active figures.

That's exactly how pyplot is meant to work: Do everything on the active figure. For convenience, a new figure is created if there is no active figure.

In particular this breaks the expectation established by usage of the inline backend that plt.plot in a cell will always create a new figure

That's a misconception. Every cell creates a new figure. Personally, I find that more intuitive because a cell is a "unit of thought", but I cannot speak for expectations of others.

I'd prefer to not have to write plt.figure every time I want a new figure to show up. In practice, that shouldn't be needed too often if you use cells to structure your code. It's only need if you need multiple figures from a single cell. For certain applications using multple subplots may also be better than using multple figures.

[ianhi]

Every cell creates a new figure. Personally, I find that more intuitive because a cell is a "unit of thought

That is a much better way of putting it. In fact I think that is really what I want, but I assumed it wouldn't be possible. Though given that the inline backend manages to do as much I imagine the the ipympl backend may also be able to. I will close this and potentially open an issue there to discuss how that might be possible to implement.

Thanks!

[QuLogic]

In particular this breaks the expectation established by usage of the inline backend that plt.plot in a cell will always create a new figure

That's a misconception. Every cell creates a new figure. Personally, I find that more intuitive because a cell is a "unit of thought", but I cannot speak for expectations of others.

This is not quite right either. Every cell closes all open figures. This might seem like a meaningless distinction, but if you were to call plt.subplots(); plt.subplots(); plt.subplots(); in a cell, you would get back 3 inline figures. What opens the figure is any plotting call (plot, scatter, etc.), implicitly, and it does so for every cell because the previous cells closed their figures.

[tacaswell]

@ianhi See matplotlib/ipympl#171 for a long discussion of what is going on and why it is a bit subtle to get right.