from transformers import AutoModel, AutoTokenizer
import streamlit as st


st.header("MeshTagger 🔖")
threshold = st.sidebar.slider("Threshold", value=0.5, min_value=0.0, max_value=1.0)
display_probabilities = st.sidebar.checkbox("Display probabilities")

if "model" not in st.session_state:
    with st.spinner("Loading model and tokenizer..."):
        st.session_state["tokenizer"] = AutoTokenizer.from_pretrained(
            "Wellcome/WellcomeBertMesh"
        )
        st.session_state["model"] = AutoModel.from_pretrained(
            "Wellcome/WellcomeBertMesh", trust_remote_code=True
        )

model = st.session_state["model"]
tokenizer = st.session_state["tokenizer"]

text = st.text_area("", value="This text is about Malaria", height=400)
inputs = tokenizer([text], padding="max_length")
outputs = model(**inputs)[0]

if display_probabilities:
    data = [
        (model.id2label[label_id], label_prob.item())
        for label_id, label_prob in enumerate(outputs)
        if label_prob > threshold
    ]
    st.table(data)
else:
    for label_id, label_prob in enumerate(outputs):
        if label_prob > threshold:
            st.button(model.id2label[label_id])

st.markdown(
    "Examples from The Wellcome Trust are taken from data that are publishes openly at [360 Giving](https://data.threesixtygiving.org/). They are published under a [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/) license."
)