Hi there 
,
After some investigation, it turns out that pyarrow has an issue with numpy.dtype values (which df.dtypes returns).
The issue has been filed and hopefully will be taken care of soon.
In the meantime, a custom exception with a more useful error message will be thrown.
In some cases, a possible workaround is to convert DataFrame cells to strings with df.astype(str).
For example:
df = pd.DataFrame([["foo", "bar"]])
foo = df.dtypes.astype(str)
st.write(foo)
I hope this helps ease some of the burden.
Please let us know if that doesn’t work for you, and we’ll try to come up with a better solution.
And thanks for using Streamlit 