app.py

import streamlit as st
from transformers import AutoTokenizer, AutoModelForCausalLM, AutoModel
import os
import pandas as pd
import numpy as np
from transformers import pipeline
from sklearn.metrics.pairwise import cosine_similarity, manhattan_distances, euclidean_distances



@st.cache
def load_model():
    tokenizer = AutoTokenizer.from_pretrained("stanford-crfm/pubmedgpt")
    model = AutoModel.from_pretrained("stanford-crfm/pubmedgpt")
    return tokenizer, model

tokenizer, model = load_model()
pipe = pipeline('feature-extraction', model=model, tokenizer=tokenizer)
def get_embedding(desc):
    return np.squeeze(pipe(desc)).mean(axis=0)


st.set_page_config(
    page_title="Clinical Trials Best Match [Eye Diseases]",
    page_icon="🧑‍💻",
    layout="wide",
)

# Constants
embs = []

# Heading
st.title('Clinical Trials Search')


# Gene File, 128 dim embeddings
data = np.load("data.npy")


@st.cache(allow_output_mutation=True)
def get_sim(emb_desc, data):
    ids = []
    scores = []
    for i in data:
        score = cosine_similarity(emb_desc, i['data'])
        ids.append(i['ids'])
        scores.append(score)
    df = pd.DataFrame(data={"url": ids, "scores": scores}).sort_values(by='scores')

    return df

st.subheader("🖮 Enter your clinical trial study description")
text = st.text_area('Example')

with st.spinner():
    emb = get_embedding(text)


st.subheader("💻 Hit Search")

if st.button("Compute"):
    with st.spinner('Searching...'):
        df = get_sim(emb, data=data)
        st.dataframe(df)