app.py

import streamlit as st
import fitz  # PyMuPDF
from sentence_transformers import SentenceTransformer, util
import faiss
from transformers import pipeline
import os
from pathlib import Path

st.title("Évaluation Stagiaire Data Scientist")

uploaded_file = st.file_uploader("Choisissez un fichier PDF", type="pdf")

def save_uploaded_file(uploaded_file, directory):
    directory = Path(directory)
    directory.mkdir(parents=True, exist_ok=True)
    file_path = directory / uploaded_file.name
    with open(file_path, "wb") as f:
        f.write(uploaded_file.getbuffer())
    return file_path

def extract_text_from_pdf(pdf_path):
    text = ""
    pdf_document = fitz.open(pdf_path)
    for page_num in range(pdf_document.page_count):
        page = pdf_document.load_page(page_num)
        text += page.get_text()
    return text

def index_document(text):
    model = SentenceTransformer('paraphrase-MiniLM-L6-v2')
    documents = [text]
    document_embeddings = model.encode(documents, convert_to_tensor=True)
    index = faiss.IndexFlatL2(document_embeddings.shape[1])
    index.add(document_embeddings.cpu().detach().numpy())
    faiss.write_index(index, 'document_index.faiss')

def get_answer_from_document(question, context):
    qa_pipeline = pipeline('question-answering', model='deepset/roberta-base-squad2')
    result = qa_pipeline(question=question, context=context)
    return result

def generate_questions(text, num_questions=5, num_beams=5):
    question_generation_pipeline = pipeline("text2text-generation", model="valhalla/t5-base-qg-hl")
    input_text = "generate questions: " + text
    questions = question_generation_pipeline(input_text, max_length=512, num_beams=num_beams, num_return_sequences=num_questions)
    return [q['generated_text'] for q in questions]

def evaluate_responses(user_responses, correct_answers):
    model = SentenceTransformer('paraphrase-MiniLM-L6-v2')
    user_embeddings = model.encode(user_responses, convert_to_tensor=True)
    correct_embeddings = model.encode(correct_answers, convert_to_tensor=True)
    scores = []
    for user_emb, correct_emb in zip(user_embeddings, correct_embeddings):
        score = util.pytorch_cos_sim(user_emb, correct_emb)
        scores.append(score.item())
    return scores

def generate_training_plan(scores, threshold=0.7):
    plan = []
    for idx, score in enumerate(scores):
        if score < threshold:
            plan.append(f"Revoir la section correspondant à la question {idx+1}")
        else:
            plan.append(f"Passer à l'étape suivante après la question {idx+1}")
    return plan

if uploaded_file is not None:
    file_path = save_uploaded_file(uploaded_file, "uploaded_documents")
    st.write(f"Fichier téléchargé et sauvegardé sous : {file_path}")

    document_text = extract_text_from_pdf(file_path)
    st.write("Texte extrait du document PDF:")
    st.write(document_text[:1000])  # Affiche les 1000 premiers caractères du texte extrait

    index_document(document_text)

    st.subheader("Questions générées")
    questions = generate_questions(document_text, num_questions=5)
    for idx, question in enumerate(questions):
        st.write(f"Question {idx+1}: {question}")

    st.subheader("Évaluer les réponses de l'utilisateur")
    user_responses = [st.text_input(f"Réponse de l'utilisateur {idx+1}") for idx in range(5)]
    if st.button("Évaluer"):
        correct_answers = ["La réponse correcte 1", "La réponse correcte 2", "La réponse correcte 3", "La réponse correcte 4", "La réponse correcte 5"]
        scores = evaluate_responses(user_responses, correct_answers)
        for idx, score in enumerate(scores):
            st.write(f"Question {idx+1}: Score {score:.2f}")

        st.subheader("Plan de formation personnalisé")
        training_plan = generate_training_plan(scores)
        for step in training_plan:
            st.write(step)