Altair fails to render chart out of pandas dataframe on Streamlit

[hugolvc]

1

I have the following code in Python, using Streamlit as framework:

          try:
               native_data = data.copy()
               
               # Create Altair chart with native data
               st.write(f"Debug: Native data type: {type(native_data)}")
               chart = alt.Chart(native_data).mark_bar().encode(
                   x=alt.X(x_col, type='nominal'),
                   y=alt.Y(y_col, type='quantitative')
               ).properties(
                   title=self._label,
                   width=400,
                   height=300
               )
               
           except Exception as chart_creation_error:
               st.write(f"Debug: Chart creation error: {chart_creation_error}")
               return "table_only"

           try:
               container.altair_chart(chart, use_container_width=True)
               return "success"
               
           except Exception as render_error:
               st.write(f"Debug: Chart rendering error: {render_error}")
               st.write(f"Debug: Render error type: {type(render_error)}")
               import traceback
               st.write("Debug: Render error traceback:")
               st.code(traceback.format_exc())
               raise render_error

And this is the debug output:

Debug: Native data type: <class 'pandas.core.frame.DataFrame'>

Debug: Chart rendering error: You passed a <class 'narwhals.stable.v1.DataFrame'> to is_pandas_dataframe.

I can see that the dataframe used to generate the Altair object is a Pandas dataframe but, nevertheless, Altair seems to confuse it with a Narwhals dataframe.

I haven't found a single reference to a similar case, so any help will be much appreciated.

[MarcoGorelli]

Hi @hugolvc

I'd like to fix this, but can't see how to reproduce the issue, as the code isn't runnable. For example, what's container?

If you could make the example reproducible, I'd happily try fixing it

[MarcoGorelli]

I've tried running

import streamlit as st
import altair as alt
import pandas as pd

native_data = pd.DataFrame({'a': [1,2,3], 'b': [4,5,6]})
x_col = 'a'
y_col = 'b'

# Create Altair chart with native data
st.write(f"Debug: Native data type: {type(native_data)}")
chart = alt.Chart(native_data).mark_bar().encode(
   x=alt.X(x_col, type='nominal'),
   y=alt.Y(y_col, type='quantitative')
).properties(
   title='foo',
   width=400,
   height=300
)

st.container().altair_chart(chart, use_container_width=True)

but it renders fine in streamlit

Is there any other library you're using? Could you show the full error traceback please?

[MarcoGorelli]

Based on The StackOverflow question, it looks like you just need to upgrade your version of Altair from 5.4.0 to 5.5.0