app.py

from sentence_transformers import SentenceTransformer
from fastapi import FastAPI
import pickle
import pandas as pd
from pydantic import BaseModel
from fastapi.middleware.cors import CORSMiddleware
import torch

corpus = pickle.load(open("./corpus/all_embeddings.pickle", "rb"))
label_encoder = pickle.load(open("./corpus/label_encoder.pickle", "rb"))
model = SentenceTransformer("sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2")
df = pd.DataFrame(data={"label": pickle.load(open("./corpus/y_all.pickle", "rb"))})

app = FastAPI()

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

class Disease(BaseModel):
    id: int
    name: str
    score: float

@app.get("/")
def greet_json():
    return {"Hello": "World!"}

# @app.post("/")
# async def greet_post():
#     return {"Hello": "Post World!"}

@app.post("/", response_model=list[Disease])
async def predict(query: str):
    query_embedding = model.encode(query).astype('float')
    similarity_vectors = model.similarity(query_embedding, corpus)[0]
    print("Similarity Vector Shape: ", similarity_vectors.shape)
    scores, indicies = torch.topk(similarity_vectors, k=len(corpus))
    print("Scores Shape: ", scores.shape)
    print("Indicies Shape: ", indicies.shape)
    id_ = df.iloc[indicies]
    id_ = id_.drop_duplicates("label")
    scores = scores[id_.index]
    diseases = label_encoder.inverse_transform(id_.label.values)
    id_ = id_.label.values
    diseases = [dict({"id": value[0], "name": value[1], "score" : value[2]}) for value in zip(id_, diseases, scores)]
    return diseases