Matplotlib plots are blurry when using a large figsize

For example:

import streamlit as st
import numpy as np
import pandas as pd

np.random.seed(0)
df = pd.DataFrame(np.random.normal(1, 1, size=100))
df.plot(figsize=(25, 5))
st.pyplot()

Results in this unreadable image:

We currently clamp st.pyplot images to Streamlit’s MAXIMUM_CONTENT_WIDTH; images bigger than this get downscaled.

See pyplot.marshall():

image_proto.marshall_images(
        image, None, -2, new_element_proto.imgs, False, channels="RGB", format="PNG"
    )

That -2 value is a magic number that causes marshall_images to do the clamping/downscaling. We could consider either passing a value of 0 (which would not do any downscaling), or we could warn the user if their image is much bigger than MAXIMUM_CONTENT_WIDTH.

Also, we should definitely turn these magic width values into constants:

ORIGINAL_WIDTH = 0
CLAMP_TO_MAX_CONTENT_WIDTH = -1

(or whatever.)

This issue was originally raised in the Streamlit forums, here: https://discuss.streamlit.io/t/matplotlib-plots-are-blurry/1224

[tvst]

Since this was posted 1.5 years ago (😱), we added some arguments to st.image to give developers more control. Taking that work as inspiration, I think the best solution here would be to give developers the same level of control for st.pyplot as we provide for st.image. In fact, we should support the exact same resolution-related parameters for pyplot as for image.

This means st.pyplot would take the following parameters:

• width: None or int
• use_column_width: ‘auto’ | ‘always’ | ‘never’. And also accept None | True | False for historical reasons.
• output_format: ‘JPEG’ | ‘PNG’ | ‘auto’. Where ‘auto’ is the same as ‘PNG’ in this case. And we could also add support for ‘SVG’, since Matplotlib supports that. NOTE: I realize this one is not related to the bug reported here. But might as well add it to the mix, since there are good usecases for this

In addition, we should explicitly document the fact that you can pass dpi=[some number] to st.pyplot() to change the pixel density. (This is automatically supported today because st.plot() forwards its kwargs to Matplotlib’s Figure.savefig() (see docs)

IMPORTANT: When doing this we should make sure to not change the default behavior of st.pyplot!

[kurt-rhee]

I’ve found that in my plots I can change the resolution by increasing the figsize, but the fig always spans the entire width of the page. Is there a way to turn off the auto scaling for pyplot?