deteksi_content.py

import os
import toml
#from dotenv import load_dotenv

#load_dotenv() 

# Load secrets from TOML file
#os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = ".streamlit/secrets.toml"

""" google_creds = os.getenv('GOOGLE_APPLICATION_CREDENTIALS_JSON')
if google_creds:
    with open('credentials.json', 'w') as f:
        f.write(google_creds)
    os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = "credentials.json" """

import streamlit as st
from datetime import datetime
import pandas as pd
from lime.lime_text import LimeTextExplainer
from test import predict_hoax, predict_proba_for_lime
import streamlit.components.v1 as components
from load_model import load_model
from st_aggrid import AgGrid, GridOptionsBuilder, GridUpdateMode
from styles import COMMON_CSS
from google.cloud import storage
from io import StringIO
import pytz



def save_corrections_to_gcs(bucket_name, file_name, correction_data):
    try:
        client = storage.Client()  # Uses the credentials set by the environment variable
        bucket = client.bucket("dashboardindohoax-bucket")
        blob = bucket.blob("koreksi_pengguna.csv")

        # Check if the blob (file) exists
        if blob.exists():
            # Download existing CSV from GCS
            existing_data = blob.download_as_string().decode('utf-8')
            existing_df = pd.read_csv(StringIO(existing_data))
        else:
            # Create a new DataFrame if the file does not exist
            existing_df = pd.DataFrame(columns=['Timestamp', 'Title', 'Content', 'Prediction', 'Correction'])

        # Append the new data to the existing data
        new_data_df = pd.DataFrame(correction_data)
        updated_df = pd.concat([existing_df, new_data_df], ignore_index=True)

        # Convert the DataFrame back to CSV and upload
        updated_csv_data = updated_df.to_csv(index=False)
        blob.upload_from_string(updated_csv_data, content_type='text/csv')
    except Exception as e:
        # If GCS fails, save locally as fallback
        local_file = "koreksi_pengguna.csv"
        if os.path.exists(local_file):
            existing_df = pd.read_csv(local_file)
        else:
            existing_df = pd.DataFrame(columns=['Timestamp', 'Title', 'Content', 'Prediction', 'Correction'])

        new_data_df = pd.DataFrame(correction_data)
        updated_df = pd.concat([existing_df, new_data_df], ignore_index=True)
        updated_df.to_csv(local_file, index=False)
        st.warning(f"GCS save failed ({e}), saved locally to {local_file}.")

def show_deteksi_konten():
    st.markdown(COMMON_CSS, unsafe_allow_html=True)

    if 'correction' not in st.session_state:
        st.session_state.correction = None
    if 'detection_result' not in st.session_state:
        st.session_state.detection_result = None
    if 'lime_explanation' not in st.session_state:
        st.session_state.lime_explanation = None
    if 'headline' not in st.session_state:
        st.session_state.headline = ""
    if 'content' not in st.session_state:
        st.session_state.content = ""
    if 'is_correct' not in st.session_state:
        st.session_state.is_correct = None

    # Dropdown for selecting a model
    st.markdown("<h6 style='font-size: 14px; margin-bottom: 0;'>Pilih Model</h6>", unsafe_allow_html=True)
    selected_model = st.selectbox(
        "",
        [
            "cahya/bert-base-indonesian-522M",
            "indobenchmark/indobert-base-p2",
            "indolem/indobert-base-uncased",
            "mdhugol/indonesia-bert-sentiment-classification"
        ],
        key="model_selector_content"
    )

    # Load the selected model
    tokenizer, model = load_model(selected_model)

    st.markdown("<h6 style='font-size: 14px; margin-bottom: 0;'>Masukkan Judul Berita :</h6>", unsafe_allow_html=True)
    st.session_state.headline = st.text_input("", value=st.session_state.headline)

    st.markdown("<h6 style='font-size: 14px; margin-bottom: 0;'>Masukkan Konten Berita :</h6>", unsafe_allow_html=True)
    st.session_state.content = st.text_area("", value=st.session_state.content)

    # Detection button
    if st.button("Deteksi", key="detect_content"):
        st.session_state.detection_result = predict_hoax(st.session_state.headline, st.session_state.content)
        st.success(f"Prediksi: {st.session_state.detection_result}")

        # Prepare the text for LIME
        lime_texts = [f"{st.session_state.headline} [SEP] {st.session_state.content}"]

        # Add a spinner and progress bar to indicate processing
        with st.spinner("Sedang memproses LIME, harap tunggu..."):
            # Explain the prediction
            explainer = LimeTextExplainer(class_names=['NON-HOAX', 'HOAX'])
            explanation = explainer.explain_instance(lime_texts[0], predict_proba_for_lime, num_features=10, num_samples=500)

            # Save the LIME explanation in session state
            st.session_state.lime_explanation = explanation.as_html()

    # Display the detection result and LIME explanation if available
    if st.session_state.lime_explanation:
        lime_html = st.session_state.lime_explanation

        # Inject CSS for font size adjustment
        lime_html = f"""
        <style>
        .lime-text-explanation, .lime-highlight, .lime-classification, 
        .lime-text-explanation * {{
            font-size: 14px !important;
        }}
        </style>
        <div class="lime-text-explanation">
            {lime_html}
        </div>
        """
        components.html(lime_html, height=200, scrolling=True)

    # Display a radio button asking if the detection result is correct
    if st.session_state.detection_result is not None:
        st.markdown("<h6 style='font-size: 16px; margin-bottom: -150px;'>Apakah hasil deteksi sudah benar?</h6>", unsafe_allow_html=True)
        st.session_state.is_correct = st.radio("", ("Ya", "Tidak"))

        if st.session_state.is_correct == "Ya":
            st.success("Deteksi sudah benar.")
        else:
            # Determine the correction based on the prediction
            st.session_state.correction = "HOAX" if st.session_state.detection_result == "NON-HOAX" else "NON-HOAX"

            wib = pytz.timezone('Asia/Jakarta')
            # Display the correction DataFrame
            correction_data = [{
                'Title': st.session_state.headline,
                'Content': st.session_state.content,
                'Prediction': st.session_state.detection_result,
                'Correction': st.session_state.correction,
                'Timestamp': datetime.now(wib).strftime("%Y-%m-%d %H:%M:%S")
            }]

            # Save button
            if st.button("Simpan"):
                # Save the correction data to GCS
                save_corrections_to_gcs("dashboardindohoax-bucket", "koreksi_pengguna.csv", correction_data)
                
                # Create a formatted string with CSS for alignment and multi-line content handling
                formatted_text = f"""
                <div style='font-size: 14px;'>
                    <p style='margin: 0;'><span style='display: inline-block; width: 120px; font-weight: bold;'>Title</span> : <span style='white-space: pre-wrap;'>{st.session_state.headline}</span></p>
                    <p style='margin: 0;'><span style='display: inline-block; width: 120px; font-weight: bold;'>Content</span> : <span style='white-space: pre-wrap;'>{st.session_state.content}</span></p>
                    <p style='margin: 0;'><span style='display: inline-block; width: 120px; font-weight: bold;'>Prediction</span> : {st.session_state.detection_result}</p>
                    <p style='margin: 0;'><span style='display: inline-block; width: 120px; font-weight: bold;'>Correction</span> : {st.session_state.correction}</p>
                </div>
                """
                
                # Display the correction as text
                st.markdown(formatted_text, unsafe_allow_html=True)
                st.success("Koreksi telah disimpan.")