Save_Data_For_Training.py

import pandas as pd
from sentence_transformers import SentenceTransformer, util
from sympy.printing.pytorch import torch


def assign_labels_to_sentences(headlines, sentences):
    labeled_data = []
    model = SentenceTransformer('distiluse-base-multilingual-cased-v2')
    model = model.to(torch.device("cpu"))  # Für Raspberry Pi 5

    headlines_embedding = model.encode(headlines, convert_to_tensor=True)
    sentences_embedding = model.encode(sentences, convert_to_tensor=True)

    similarity_matrix = util.cos_sim(sentences_embedding, headlines_embedding)

    for i, sentence in enumerate(sentences):
        best_match_idx = torch.argmax(similarity_matrix[i]).item()
        best_score = similarity_matrix[i][best_match_idx].item()

        if best_score > 0.3:
            print(f"Beste Überschrift: {headlines[best_match_idx]}, Text: {sentence}, Score: {best_score:.3f}")
            labeled_data.append((sentence, headlines[best_match_idx]))
        else:
            labeled_data.append((sentence, None))

    return labeled_data

def save_texts_with_labels_to_csv(answer_and_label_as_dataframe):
    try:
        df = pd.DataFrame(answer_and_label_as_dataframe, columns=["Text", "Label"])
        df.to_csv('training/training_elias.csv', index=False)
        return True
    except Exception as e:
        print(f"Error while saving CSV: {e}")
        return False