app.py

import gradio as gr
import pandas as pd
import requests
import os
import time
from collections import Counter
from io import StringIO

# =====================================================
# CONFIGURAÇÃO
# =====================================================
MODEL_ID = "dslim/bert-base-NER"
API_URL = f"https://router.huggingface.co/{MODEL_ID}"

HF_API_TOKEN = os.getenv("HF_API_TOKEN")
if not HF_API_TOKEN:
    raise RuntimeError("HF_API_TOKEN não encontrado. Configure em Settings → Secrets.")

HEADERS = {
    "Authorization": f"Bearer {HF_API_TOKEN}",
    "Content-Type": "application/json"
}

# =====================================================
# FUNÇÃO PRINCIPAL
# =====================================================
def analyze_abstracts(csv_text):

    # ---------- Ler CSV ----------
    try:
        df = pd.read_csv(StringIO(csv_text))
    except Exception as e:
        return pd.DataFrame(
            [["Erro ao ler CSV", str(e), ""]],
            columns=["Term", "Frequency", "Hot"]
        )

    if "abstract" not in df.columns:
        return pd.DataFrame(
            [["Erro", "Coluna 'abstract' não encontrada", ""]],
            columns=["Term", "Frequency", "Hot"]
        )

    all_terms = []

    # ---------- Processar abstracts ----------
    for abstract in df["abstract"].dropna():

        payload = {
            "inputs": abstract,
            "parameters": {
                "aggregation_strategy": "simple"
            },
            "options": {
                "wait_for_model": True
            }
        }

        try:
            response = requests.post(API_URL, headers=HEADERS, json=payload, timeout=60)
            result = response.json()

            # Caso modelo ainda esteja carregando
            if isinstance(result, dict) and "error" in result:
                print("Aviso da API:", result)
                continue

            # Caso correto
            if isinstance(result, list):
                for ent in result:
                    term = ent.get("word", "").strip().lower()
                    if len(term) > 2:
                        all_terms.append(term)

        except Exception as e:
            print("Erro na requisição:", e)
            continue

    if not all_terms:
        return pd.DataFrame(
            [["Nenhum termo médico encontrado (verifique logs da API)", "", ""]],
            columns=["Term", "Frequency", "Hot"]
        )

    # ---------- Frequência ----------
    term_counts = Counter(all_terms)
    df_terms = (
        pd.DataFrame(term_counts.items(), columns=["Term", "Frequency"])
        .sort_values(by="Frequency", ascending=False)
        .reset_index(drop=True)
    )

    # ---------- Hot terms ----------
    threshold = df_terms["Frequency"].quantile(0.90)
    df_terms["Hot"] = df_terms["Frequency"] >= threshold

    return df_terms

# =====================================================
# INTERFACE
# =====================================================
interface = gr.Interface(
    fn=analyze_abstracts,
    inputs=gr.Textbox(
        lines=18,
        placeholder="Cole aqui seu CSV com colunas: date,abstract"
    ),
    outputs=gr.Dataframe(),
    title="Medical Hot Terms Analyzer",
    description=(
        "Extrai termos médicos de abstracts usando PubMedBERT (NER) "
        "e identifica hot terms (top 10%)."
    ),
)

if __name__ == "__main__":
    interface.launch()