utils.py

import pandas as pd
import arxiv
import requests
from pinecone import Pinecone, ServerlessSpec
import logging
import os
import asyncio
from dotenv import load_dotenv

load_dotenv(".env")

script_dir = os.path.dirname(os.path.abspath(__file__))
os.chdir(script_dir)


def get_zotero_ids(api_key, library_id, tag):

    base_url = "https://api.zotero.org"
    suffix = "/users/" + library_id + "/items?tag=" + tag

    header = {"Authorization": "Bearer " + api_key}
    request = requests.get(base_url + suffix, headers=header)

    return [data["data"]["archiveID"].replace("arXiv:", "") for data in request.json()]


def get_arxiv_papers(ids=None, category=None, comment=None):

    logging.getLogger("arxiv").setLevel(logging.WARNING)

    client = arxiv.Client()

    if category is None:
        search = arxiv.Search(
            id_list=ids,
            max_results=len(ids),
        )
    else:
        if comment is None:
            custom_query = f"cat:{category}"
        else:
            custom_query = f"cat:{category} AND co:{comment}"

        search = arxiv.Search(
            query=custom_query,
            max_results=15,
            sort_by=arxiv.SortCriterion.SubmittedDate,
        )
    if ids is None and category is None:
        raise ValueError("not a valid query")

    df = pd.DataFrame(
        {
            "Title": [result.title for result in client.results(search)],
            "Abstract": [
                result.summary.replace("\n", " ") for result in client.results(search)
            ],
            "Date": [
                result.published.date().strftime("%Y-%m-%d")
                for result in client.results(search)
            ],
            "id": [result.entry_id for result in client.results(search)],
        }
    )

    if ids:
        df.to_csv("arxiv-scrape.csv", index=False)
    return df


def get_hf_embeddings(api_key, df):

    title_abs = [
        title + "[SEP]" + abstract
        for title, abstract in zip(df["Title"], df["Abstract"])
    ]

    API_URL = "https://api-inference.huggingface.co/models/malteos/scincl"
    headers = {"Authorization": f"Bearer {api_key}"}

    response = requests.post(
        API_URL, headers=headers, json={"inputs": title_abs, "wait_for_model": False}
    )
    print(str(response.status_code) + "This part needs an update, causing KeyError 0")
    if response.status_code == 503:
        response = asyncio.run(
            asyncio.to_thread(
                requests.post,
                API_URL,
                headers=headers,
                json={"inputs": title_abs, "wait_for_model": True},
            )
        )
        # response = requests.post(
        #     API_URL, headers=headers, json={"inputs": title_abs, "wait_for_model": True}
        # )

    embeddings = response.json()

    return embeddings, len(embeddings[0])


def upload_to_pinecone(api_key, index, namespace, embeddings, dim, df):
    input = [
        {"id": df["id"][i], "values": embeddings[i]} for i in range(len(embeddings))
    ]

    pc = Pinecone(api_key=api_key)
    if index in pc.list_indexes().names():
        while True:
            logging.warning(f"Index name : {index} already exists.")
            return f"Index name : {index} already exists"

    pc.create_index(
        name=index,
        dimension=dim,
        metric="cosine",
        spec=ServerlessSpec(cloud="aws", region="us-east-1"),
    )

    index = pc.Index(index)
    return index.upsert(vectors=input, namespace=namespace)


def main():
    script_dir = os.path.dirname(os.path.abspath(__file__))
    os.chdir(script_dir)
    logging.basicConfig(
        filename="logs/logfile.log",
        level=logging.INFO,
        format="%(asctime)s - %(levelname)s - %(message)s",
    )
    logging.getLogger("arxiv").setLevel(logging.WARNING)
    logging.info("Project Initialization Script Started (Serverless)")

    ids = get_zotero_ids(
        os.getenv("ZOTERO_API_KEY"),
        os.getenv("ZOTERO_LIBRARY_ID"),
        os.getenv("ZOTERO_TAG"),
    )
    print(ids)

    df = get_arxiv_papers(ids=ids)

    embeddings, dim = get_hf_embeddings(os.getenv("HF_API_KEY"), df)

    feedback = upload_to_pinecone(
        api_key=os.getenv("PINECONE_API_KEY"),
        index=os.getenv("INDEX_NAME"),
        namespace=os.getenv("NAMESPACE_NAME"),
        embeddings=embeddings,
        dim=dim,
        df=df,
    )

    logging.info(feedback)
    if feedback is dict:
        return f"Retrieved {len(ids)} papers from Zotero. Successfully upserted {feedback['upserted_count']} embeddings in {os.getenv('NAMESPACE_NAME')} namespace."
    else:
        return feedback


def get_new_papers(df):
    df_main = pd.read_csv("arxiv-scrape.csv")
    df.reset_index(inplace=True)
    df.drop(columns=["index"], inplace=True)
    union_df = df.merge(df_main, how="left", indicator=True)
    df = union_df[union_df["_merge"] == "left_only"].drop(columns=["_merge"])
    if df.empty:
        return "No New Papers Found"
    else:
        df_main = pd.concat([df_main, df], ignore_index=True)
        df_main.drop_duplicates(inplace=True)
        df_main.to_csv("arxiv-scrape.csv", index=False)
        return df


def recommend_papers(api_key, index, namespace, embeddings, df, threshold):

    pc = Pinecone(api_key=api_key)
    if index in pc.list_indexes().names():
        index = pc.Index(index)
    else:
        raise ValueError(f"{index} doesnt exist. Project isnt initialized properly")

    results = []
    score_threshold = threshold
    for i, embedding in enumerate(embeddings):
        query = embedding
        result = index.query(
            namespace=namespace, vector=query, top_k=3, include_values=False
        )
        sum_score = sum(match["score"] for match in result["matches"])
        if sum_score > score_threshold:
            results.append(
                f"Paper-URL : [{df['id'][i]}]({df['id'][i]}) with score: {sum_score / 3} <br />"
            )

    if results:
        return "\n".join(results)
    else:
        return "No Interesting Paper"


def recs(threshold):
    logging.info("Weekly Script Started (Serverless)")

    df = get_arxiv_papers(
        category=os.getenv("ARXIV_CATEGORY_NAME"),
        comment=os.getenv("ARXIV_COMMENT_QUERY"),
    )

    df = get_new_papers(df)

    if not isinstance(df, pd.DataFrame):
        return df

    embeddings, _ = get_hf_embeddings(os.getenv("HF_API_KEY"), df)

    results = recommend_papers(
        os.getenv("PINECONE_API_KEY"),
        os.getenv("INDEX_NAME"),
        os.getenv("NAMESPACE_NAME"),
        embeddings,
        df,
        threshold,
    )

    return results


if __name__ == "__main__":
    choice = int(input("1. Initialize\n2. Recommend Papers\n"))
    if choice == 1:
        print(main())
    elif choice == 2:
        threshold = float(input("Enter Similarity Threshold"))
        print(recs(threshold))
    else:
        raise ValueError("Invalid Input")