Only rerun the function without rerunning the whole script

Using Streamlit
,
1

Hi, I have a function that runs some heavy computation to produce an image,
and i want to display the image with a slider to adjust the brightness.
like this :

def AdjustBrightness(image, plotpath):
    ## -- Brightness adjustments --
    ShiftBy = st.slider('Brightness adjustments')
    ## -- Save image --
    plt.figure(); 
    librosa.display.specshow(image + float(ShiftBy))
    plt.savefig(plotpath)
    ## -- Display image --
    image = Image.open(plotpath)
    st.image(image, caption='Spectrogram & Onset', use_column_width=True)

plotpath = "plot.png"

image = SomeHeavyComputation()

AdjustBrightness(image, plotpath)

However everytime i adjust the slider, the SomeHeavyComputation() is re-ran, which takes up alot of time.
How do i prevent the earlier computations, and only re-run AdjustBrightness() when i adjust the slider?

2

Hi @divinekage -

This is the basis for the caching functionality of Streamlit:

3

Yea I get that. So what I have is that I am displaying 5 pictures. And when I display the 6th picture, which is this to adjust brightness, it will rerun the whole script. And each picture takes 0.5 seconds to load.
So when I change the slider, it’ll take 2.5s to load the past 5 pictures, then follow by another 0.5s for this 6th picture.
Main bottleneck is in the I/O operations (writing reading from file).
So I only want to re-run the 6th picture operation, without rerunning the previous 5 pictures.

4

By caching the load of each picture, you take out the I/O portion, because the cache will read from RAM. Thus 0.5s will become way shorter, hopefully imperceptible.

5

The caching only makes it such that you return values without actually running the function right.
Which means if I do something like

@st.cache()
def AdjustBrightness(image, plotpath):
    ...
    return None

when I run AdjustBrightness(image, plotpath) for the first time it will plot the image and return None
When I run it for the 2nd time. It will not run whats inside. And return None immediately, which is not what I want because the plot will then disappear

6

Cache the loading of the picture from disk. You can cache the brightness as well, but if you cache the loading of the picture it will remove the disk IO

7

If I understand you correctly, i should be getting the binary values from plotpath from cache.
I adjusted my function like this:

@st.cache(suppress_st_warning=True)
def AdjustBrightness(image, plotpath):
    ## -- Brightness adjustments --
    ShiftBy = st.slider('Brightness adjustments')
    ## -- Save image --
    plt.figure(); 
    librosa.display.specshow(image + float(ShiftBy))
    plt.savefig(plotpath)
    ## -- Display image --
    return Image.open(plotpath)

plotpath = "plot.png"
image = SomeHeavyComputation()

AdjustBrightness(image, plotpath) # This runs in 0.00915s, fast because from cache
st.image(image, caption='Spectrogram & Onset', use_column_width=True) # 0.500s

However the bottleneck is still in the st.image() function, which is 0.5s

1 Like