Streamlit app for exploring data & small training runs with Prodigy

Reposting from Twitter: I built a little experimental small app using Streamlit, a new open-source tool for building interaticve ML apps. The app connects to your Prodigy database and lets you load and view the existing datasets. For NER annotations, it merges all spans on the same text and will show you both accepted and rejected spans (in green and red, respectively). You can also use the dropdown to search for specific examples and run quick training experiments :rocket:

It's not supposed to mimick all command-line features – it's mostly a quick experiment to show what's possible.

You can run the app straight from the raw gist by running the following in your Prodigy environment. (Make sure you have the en_core_web_sm model installed – you can change the available models by downloading and editing the Python file.)

pip install streamlit
streamlit run

there is a bug with streamlit on windows(cn)...

err msg:
'gbk' codec can't decode byte 0x85

i edit some code like
st.button("🚀 Start training"):
st.button("Start training"):
will right run