display association rules plots in streamlit

[batmanscode]

issue #884

Changes

Added display_format

def plot_model(model, plot="2d", scale=1, display_format=None):
.
.
.
    if display_format=='streamlit'
        st.write(fig)
    else:
        fig.show()

Notes

• Didn't include import streamlit as st since this will only work inside streamlit anyway
• Didn't use st.plotly_chart since st.write can plot many types of charts and won't have to be changed if plotly is dropped later

Update

• Added display_format to anywhere plot_model can be called 😃
• Error handling

Additional:

In nlp.py where fig.iplot() is used, I had to change asFigure = save_param to asFigure = True because streamlit needs a plotly object to interpret (https://discuss.streamlit.io/t/cufflinks-in-streamlit/2232/2)

.
.
.
if display_format=='streamlit':
    fig = df.iplot(asFigure=True) # plotly obj needs to be returned for streamlit to interpret
    st.write(fig)
else:
    df.iplot()

[Yard1]

Hey @batmanscode, thanks for your contribution! I am not familiar with streamlit, but can it be applied to other plots too - for example, in regression or clustering?

[batmanscode]

Hi @Yard1, yes absolutely! Adding a conditional st.write(fig) anywhere there is a fig.show() will work.

I would like to add display_format='streamlit' to the plots in other modules as well but it seems for some the change will have to be in ..internal/tabular and that will take a little more time for me to go through.

And as a side note, I do recommend checking out streamlit when you have the time 😃

[batmanscode]

Okay so maybe ..internal/tabular doesn't really need it since there is an option to save the plot in plot_model() so a streamlit app can just display the saved image.

[Yard1]

I think it would be a good idea to have a unified API. Do you think you could add that? If you have trouble with modifying internal/tabular, I am happy to help.

[batmanscode]

Yeah I think you're right. Makes more sense to have it uniform.

Yes, I should be able to! Thank you, I will let you know if I run into any problems 😃

[batmanscode]

Hi @Yard1, my last commit is failing some tests but I'm not sure why. Can you please have a look when you have some time?

[batmanscode]

Btw, for plots like on this line:

Why is it df3 = df2.iplot()? Based on some tutorials I saw, I thought it should be df2.iplot() or iplot(df3). Or have I missed something? 😅

[batmanscode]

Btw, for plots like on this line:

Why is it df3 = df2.iplot()? Based on some tutorials I saw, I thought it should be df2.iplot() or iplot(df3). Or have I missed something? 😅

I wonder if this could be related to #319

[Yard1]

@batmanscode Apologies, I must have missed this. I'll take a look as soon as I am able.

[batmanscode]

No problem at all @Yard1, I just figured you must have been busy.

Alright, thank you 😃

[Yard1]

@batmanscode Could you merge master into this branch, so the tests can pass? Thanks.

[Yard1]

@batmanscode So the issue was that evaluate_models needs to be updated any time there's a new parameter added to plot_models. One more thing before I merge this, though - could the default value for display_format be "matplotlib", and could you make it throw an exception if neither "matplotlib" nor "streamlit" is passed as the display_format parameter? It's easier to understand than just None. Thanks.

[batmanscode]

Oh right, thank you for the fix and explanation 😃 will keep that in mind for any future changes.

Okay sure, that does sound clearer. I will go through and add this exception. And I think you mean "plotly" 😆

I do have a concern about naming the default plots:
Right now plot_model uses plotly for every module and in this case plotly being default for display_format makes sense but, if in a future release, some modules use something other than plotly, it might get a little weird.

Or do you think that it's unlikely and plot_model will always use one plotting library in every module?

For example:
If the nlp module uses some other library for plotting, then its plot_model would have display_format=newlib or streamlit while all the other modules would have display_format=plotly or streamlit which would mean all the display_format options won't work universally.

In this case, another option could be to change display_format to something like special_formatting=None or streamlit. This way each module can have its own default plotting library without causing any confusion.

[Yard1]

For example:
If the nlp module uses some other library for plotting, then its plot_model would have display_format=newlib or streamlit while all the other modules would have display_format=plotly or streamlit which would mean all the display_format options won't work universally.

In this case, another option could be to change display_format to something like special_formatting=None or streamlit. This way each module can have its own default plotting library without causing any confusion.

Fair point. Let's keep None for now, then. Just please make sure that a user cannot pass an invalid value.

[batmanscode]

Alright will do. I think the if/else statements need to be changed as well. Right now things will work fine even if an invalid value is passed since it will just execute fig.show() if anything other than 'streamlit' is given.

Right now it's:

if display_format == 'streamlit':
    st.write(fig)
else:
    fig.show()

But something like this is probably better:

if display_format == 'streamlit':
    st.write(fig)
if disploy_format == None:
    fig.show()

[Yard1]

@batmanscode It's fine to leave it as is, considering we'll have value checks beforehand.

[batmanscode]

Ah yes, thanks!

[batmanscode]

All done @Yard1. Sorry, forgot to comment that I was finished.

[Yard1]

Just one thing, please change this in all strings

[Yard1]

One last thing - can you run black on all modified files? Thanks.

[batmanscode]

One last thing - can you run black on all modified files? Thanks.

Done 👍🏻

[Yard1]

Thanks, looks great.

[Yard1]

@batmanscode Hey, are you on PyCaret slack? Could you join if you haven't already and DM me there? I have no experience with streamlit but would like to expand the support for it to yellowbrick visualizers and I could use your help. Thanks.

[batmanscode]

Thank you @Yard1, I've learnt a lot doing this 😃

Sure, I can join and contact you there, happy to help. I actually haven't used slack in a few years lol.

[nataliaasaa]

Hey! Does anyone knows a way to show evaluate_model results on streamlit? Thanksss

[batmanscode]

Hey! Does anyone knows a way to show evaluate_model results on streamlit? Thanksss

Hi, evaluate_model() creates a little UI in a notebook using ipywidgets (I think) so this will not work in a streamlit app, at least right now

You can create your own evaluate_model() like function/script to work in streamlit using plot_model() (which is what evaluate_model() uses internally)

Something like this should work 😊:

"""
# Evaluate Model
"""

plots = st.selectbox('select plot', ("auc", "threshold")) # docs will have the other plots, not all are supported

plot_model(model, plot=plots, display_format="streamlit")