Capturing matplotlib figures in Jupyter notebooks

[z0u]

I'm using Modal in a Jupyter notebook, as shown in the guide. It works, but I'm having trouble with the cell output.

@app.function()
def train(epochs):
    from time import sleep
    from IPython.display import clear_output
    import matplotlib.pyplot as plt

    # Iteratively plot a simple function to demonstrate
    xs = []
    ys = []
    for i in range(epochs):
        clear_output(wait=True)
        xs.append(i)
        ys.append(i**2)
        plt.plot(xs, ys)
        plt.show()
        sleep(1)

with modal.enable_output(), app.run():
    #train.local(5)
    train.remote(5)

The plot displays when running locally, but not remotely. I'd like to be able to display plots during training so I can see metrics directly in my notebook (updated once per epoch).

Is there any way to get rich content to display from within a remote function? Or do I need to pass data back some other way? Or, do I need to bite the bullet and emit metrics to a remote service?

[mwaskom]

Hi, the logging that goes back to the local client only includes the stdout/stderr pipes; we don't capture matplotlib figures. You'll need to return from the Function to show the figure. Here's an example that works: