app.py

"""
Prompt A/B Tester - Streamlit Web Application
Aplikacja do testowania promptów OpenAI GPT z metodologią A/B
"""

import streamlit as st
import time
from datetime import datetime
from config import ConfigManager
from api_handler import APIHandler
from test_runner import TestRunner

# Page configuration
st.set_page_config(
    page_title="Prompt A/B Tester",
    page_icon="🔬",
    layout="wide",
    initial_sidebar_state="expanded"
)

# Custom CSS for better text visibility in disabled text areas
st.markdown("""
    <style>
    /* Improve visibility of disabled text areas */
    .stTextArea textarea[disabled] {
        color: #FFFFFF !important;
        opacity: 1 !important;
        -webkit-text-fill-color: #FFFFFF !important;
    }

    /* For light theme */
    [data-theme="light"] .stTextArea textarea[disabled] {
        color: #000000 !important;
        -webkit-text-fill-color: #000000 !important;
    }
    </style>
    """, unsafe_allow_html=True)

# Initialize session state
if 'initialized' not in st.session_state:
    st.session_state.initialized = True
    st.session_state.responses = []
    st.session_state.test_running = False
    st.session_state.logs = []
    st.session_state.results = None
    st.session_state.rated_responses = []

# Initialize components
config = ConfigManager()
api_handler = APIHandler(config.get_api_key())
test_runner = TestRunner(api_handler)

# Helper functions
def log_message(message):
    """Add message to logs"""
    timestamp = datetime.now().strftime("%H:%M:%S")
    st.session_state.logs.append(f"[{timestamp}] {message}")

def clear_logs():
    """Clear all logs"""
    st.session_state.logs = []

def update_progress_callback(current, total):
    """Progress callback for test runner"""
    # Progress is handled by st.progress in Streamlit
    pass

# Title
st.title("🔬 Prompt A/B Tester")
st.markdown("**Testuj i porównuj wydajność dwóch wersji promptów OpenAI GPT**")

# Sidebar - Configuration
with st.sidebar:
    st.header("⚙️ Konfiguracja")

    # API Key
    st.subheader("🔑 Klucz API OpenAI")

    # Check if API key is in secrets (Hugging Face deployment)
    api_key_from_secrets = None
    try:
        if hasattr(st, 'secrets') and 'OPENAI_API_KEY' in st.secrets:
            api_key_from_secrets = st.secrets['OPENAI_API_KEY']
            st.success("✅ Klucz API załadowany z Secrets")
    except:
        pass

    if not api_key_from_secrets:
        api_key_input = st.text_input(
            "Klucz API",
            type="password",
            value=config.get("api_key", ""),
            help="Wprowadź swój klucz API OpenAI"
        )

        if st.button("💾 Zapisz i Waliduj API Key"):
            if api_key_input:
                api_handler.set_api_key(api_key_input)
                config.set_api_key(api_key_input)
                success, message = api_handler.validate_api_key()
                if success:
                    st.success(f"✅ {message}")
                else:
                    st.error(f"❌ {message}")
            else:
                st.warning("⚠️ Klucz API nie może być pusty")

    st.divider()

    # Model and Parameters
    st.subheader("🤖 Model i Parametry")

    # Refresh models
    if st.button("🔄 Odśwież modele"):
        models = api_handler.get_available_models()
        st.success(f"Znaleziono {len(models)} modeli")

    models = api_handler.get_available_models()
    model = st.selectbox(
        "Model",
        options=models,
        index=models.index(config.get("model", "gpt-4o")) if config.get("model", "gpt-4o") in models else 0,
        help="Wybierz model OpenAI do testowania. Aplikacja automatycznie dostosowuje parametry API dla wszystkich modeli."
    )

    temperature = st.slider(
        "Temperature",
        min_value=0.0,
        max_value=2.0,
        value=config.get("temperature", 0.1),
        step=0.1,
        help="0.0 = deterministyczne, 2.0 = kreatywne"
    )

    max_tokens = st.number_input(
        "Max Tokens",
        min_value=100,
        max_value=4000,
        value=config.get("max_tokens", 2000),
        step=100
    )

    num_responses = st.number_input(
        "Liczba odpowiedzi (na prompt)",
        min_value=1,
        max_value=50,
        value=config.get("num_responses", 5),
        step=1,
        help="Ile odpowiedzi wygenerować dla każdego promptu"
    )

    # Save settings
    if st.button("💾 Zapisz ustawienia"):
        settings = {
            "model": model,
            "temperature": temperature,
            "max_tokens": max_tokens,
            "num_responses": num_responses
        }
        config.update_settings(settings)
        st.success("✅ Ustawienia zapisane")

    # Cost estimation
    if st.button("💰 Szacuj koszt"):
        total_responses = num_responses * 2
        cost = api_handler.estimate_cost(model, total_responses)
        st.info(f"**Szacunkowy koszt:**\n\n"
                f"Model: {model}\n\n"
                f"Odpowiedzi: {total_responses}\n\n"
                f"Koszt: ${cost:.4f} USD")

# Main tabs
tab1, tab2, tab3, tab4 = st.tabs(["🚀 Test", "⭐ Ocenianie", "📊 Wyniki", "📋 Logi"])

# Tab 1: Test
with tab1:
    st.header("🚀 Przeprowadź Test A/B")

    # Input method selection
    input_method = st.radio(
        "Metoda wprowadzania promptów:",
        options=["📁 Upload plików", "✏️ Wklej tekst"],
        horizontal=True
    )

    prompt_a = None
    prompt_b = None

    if input_method == "📁 Upload plików":
        col1, col2 = st.columns(2)

        with col1:
            st.subheader("Prompt A")
            file_a = st.file_uploader(
                "Wybierz plik dla Promptu A",
                type=['md', 'txt'],
                key="file_a"
            )
            if file_a:
                prompt_a = file_a.read().decode('utf-8')
                st.success(f"✅ Wczytano {len(prompt_a)} znaków")

        with col2:
            st.subheader("Prompt B")
            file_b = st.file_uploader(
                "Wybierz plik dla Promptu B",
                type=['md', 'txt'],
                key="file_b"
            )
            if file_b:
                prompt_b = file_b.read().decode('utf-8')
                st.success(f"✅ Wczytano {len(prompt_b)} znaków")

    else:  # Wklej tekst
        col1, col2 = st.columns(2)

        with col1:
            st.subheader("Prompt A")
            prompt_a = st.text_area(
                "Wklej treść Promptu A",
                height=300,
                key="text_a"
            )

        with col2:
            st.subheader("Prompt B")
            prompt_b = st.text_area(
                "Wklej treść Promptu B",
                height=300,
                key="text_b"
            )

    # Test controls
    st.divider()

    col1, col2, col3 = st.columns([1, 1, 3])

    with col1:
        run_button = st.button(
            "🚀 Uruchom Test",
            type="primary",
            disabled=st.session_state.test_running,
            use_container_width=True
        )

    with col2:
        if st.button("🗑️ Wyczyść", use_container_width=True):
            st.session_state.responses = []
            st.session_state.results = None
            st.session_state.rated_responses = []
            clear_logs()
            st.rerun()

    # Run test
    if run_button:
        # Validation
        if not config.get_api_key():
            st.error("❌ Najpierw ustaw klucz API w panelu bocznym")
        elif not prompt_a or not prompt_b:
            st.error("❌ Oba prompty muszą być wypełnione")
        else:
            clear_logs()
            log_message("🚀 Rozpoczynanie testu A/B")
            log_message(f"Model: {model}, Temperature: {temperature}, Max tokens: {max_tokens}")
            log_message(f"Liczba odpowiedzi na prompt: {num_responses}")

            # Progress tracking
            progress_bar = st.progress(0)
            status_text = st.empty()

            total_iterations = num_responses * 2
            current = 0

            st.session_state.test_running = True

            # Generate responses for Prompt A
            log_message("🔄 Generowanie odpowiedzi dla PROMPTU A...")
            for i in range(num_responses):
                current += 1
                progress_bar.progress(current / total_iterations)
                status_text.text(f"Generowanie: {current}/{total_iterations} (A-{i+1})")

                response = api_handler.generate_response(
                    prompt_a, model, temperature, max_tokens
                )

                st.session_state.responses.append({
                    'Option': 'A',
                    'Response_ID': i + 1,
                    'Response': response,
                    'Score': None
                })

                if response.startswith("ERROR"):
                    log_message(f"   A-{i+1}/{num_responses}... ❌ {response}")
                else:
                    log_message(f"   A-{i+1}/{num_responses}... ✅ ({len(response)} znaków)")

                time.sleep(0.5)

            # Generate responses for Prompt B
            log_message("\n🔄 Generowanie odpowiedzi dla PROMPTU B...")
            for i in range(num_responses):
                current += 1
                progress_bar.progress(current / total_iterations)
                status_text.text(f"Generowanie: {current}/{total_iterations} (B-{i+1})")

                response = api_handler.generate_response(
                    prompt_b, model, temperature, max_tokens
                )

                st.session_state.responses.append({
                    'Option': 'B',
                    'Response_ID': i + 1,
                    'Response': response,
                    'Score': None
                })

                if response.startswith("ERROR"):
                    log_message(f"   B-{i+1}/{num_responses}... ❌ {response}")
                else:
                    log_message(f"   B-{i+1}/{num_responses}... ✅ ({len(response)} znaków)")

                time.sleep(0.5)

            progress_bar.progress(1.0)
            status_text.text("✅ Zakończono!")
            log_message(f"\n✅ GENEROWANIE ZAKOŃCZONE - wygenerowano {len(st.session_state.responses)} odpowiedzi")

            st.session_state.test_running = False
            st.success("🎉 Test zakończony! Przejdź do zakładki '⭐ Ocenianie'")

# Tab 2: Rating
with tab2:
    st.header("⭐ Oceń Odpowiedzi")

    if not st.session_state.responses:
        st.info("ℹ️ Najpierw uruchom test w zakładce '🚀 Test'")
    else:
        st.markdown("**Oceń każdą odpowiedź w skali 1-5:**")
        st.markdown("1 = Bardzo słaba | 2 = Słaba | 3 = Średnia | 4 = Dobra | 5 = Bardzo dobra")
        st.divider()

        # Display responses for rating
        for idx, resp in enumerate(st.session_state.responses):
            with st.container():
                col1, col2 = st.columns([4, 1])

                with col1:
                    option_label = f"{resp['Option']}-{resp['Response_ID']}"
                    st.subheader(f"Option: {option_label}")
                    st.markdown("**Odpowiedź:**")
                    st.text_area(
                        "Odpowiedź:",
                        value=resp['Response'],
                        height=150,
                        key=f"response_display_{idx}",
                        disabled=True,
                        label_visibility="collapsed"
                    )

                with col2:
                    st.markdown("**Ocena**")
                    score = st.number_input(
                        "Ocena (1-5)",
                        min_value=1,
                        max_value=5,
                        value=resp.get('Score', 3),
                        step=1,
                        key=f"score_{idx}",
                        label_visibility="collapsed"
                    )
                    st.session_state.responses[idx]['Score'] = score

                st.divider()

        # Calculate results button
        if st.button("📊 Oblicz Wyniki", type="primary", use_container_width=True):
            st.session_state.rated_responses = st.session_state.responses.copy()
            st.session_state.results = test_runner.calculate_results(st.session_state.rated_responses)
            st.success("✅ Wyniki obliczone! Przejdź do zakładki '📊 Wyniki'")
            st.rerun()

# Tab 3: Results
with tab3:
    st.header("📊 Wyniki Testu A/B")

    if not st.session_state.results:
        st.info("ℹ️ Najpierw oceń odpowiedzi i kliknij 'Oblicz Wyniki' w zakładce '⭐ Ocenianie'")
    else:
        results = st.session_state.results

        # Results table
        st.subheader("Podsumowanie")

        col1, col2, col3, col4, col5 = st.columns(5)

        with col1:
            st.metric("Option", "A")
        with col2:
            st.metric("Count", results['A']['count'])
        with col3:
            st.metric("Score", f"{results['A']['score']:.2f}")
        with col4:
            st.metric("Min", results['A']['min'])
        with col5:
            st.metric("Max", results['A']['max'])

        col1, col2, col3, col4, col5 = st.columns(5)

        with col1:
            st.metric("Option", "B")
        with col2:
            st.metric("Count", results['B']['count'])
        with col3:
            st.metric("Score", f"{results['B']['score']:.2f}")
        with col4:
            st.metric("Min", results['B']['min'])
        with col5:
            st.metric("Max", results['B']['max'])

        st.divider()

        # Winner
        if results['A']['score'] > results['B']['score']:
            winner = 'A'
            diff = results['A']['score'] - results['B']['score']
            st.success(f"🏆 **ZWYCIĘZCA: Prompt {winner}**\n\nPrzewaga: +{diff:.2f} punktu")
        elif results['B']['score'] > results['A']['score']:
            winner = 'B'
            diff = results['B']['score'] - results['A']['score']
            st.success(f"🏆 **ZWYCIĘZCA: Prompt {winner}**\n\nPrzewaga: +{diff:.2f} punktu")
        else:
            st.info("🤝 **REMIS**\n\nObie opcje uzyskały identyczny wynik")

        st.divider()

        # Export section
        st.subheader("📥 Eksport Wyników")
        st.markdown("Pobierz wyniki w preferowanym formacie:")

        settings = {
            "model": config.get("model"),
            "temperature": config.get("temperature"),
            "max_tokens": config.get("max_tokens"),
            "num_responses": config.get("num_responses")
        }

        timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")

        # Create 3 columns for export buttons
        col1, col2, col3 = st.columns(3)

        with col1:
            # CSV Export
            csv_buffer = test_runner.export_to_csv(
                st.session_state.rated_responses,
                results,
                settings
            )
            st.download_button(
                label="📄 CSV",
                data=csv_buffer,
                file_name=f"prompt_ab_test_{timestamp}.csv",
                mime="text/csv",
                use_container_width=True,
                help="Pobierz w formacie CSV"
            )

            # JSON Export
            json_buffer = test_runner.export_to_json(
                st.session_state.rated_responses,
                results,
                settings
            )
            st.download_button(
                label="🔧 JSON",
                data=json_buffer,
                file_name=f"prompt_ab_test_{timestamp}.json",
                mime="application/json",
                use_container_width=True,
                help="Pobierz w formacie JSON"
            )

        with col2:
            # Excel Export
            excel_buffer = test_runner.export_to_excel(
                st.session_state.rated_responses,
                results,
                settings
            )
            st.download_button(
                label="📊 Excel",
                data=excel_buffer,
                file_name=f"prompt_ab_test_{timestamp}.xlsx",
                mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
                use_container_width=True,
                help="Pobierz w formacie Excel"
            )

            # TXT Export
            txt_buffer = test_runner.export_to_txt(
                st.session_state.rated_responses,
                results,
                settings
            )
            st.download_button(
                label="📝 TXT",
                data=txt_buffer,
                file_name=f"prompt_ab_test_{timestamp}.txt",
                mime="text/plain",
                use_container_width=True,
                help="Pobierz w formacie TXT"
            )

        with col3:
            # Markdown Export
            markdown_buffer = test_runner.export_to_markdown(
                st.session_state.rated_responses,
                results,
                settings
            )
            st.download_button(
                label="📋 Markdown",
                data=markdown_buffer,
                file_name=f"prompt_ab_test_{timestamp}.md",
                mime="text/markdown",
                use_container_width=True,
                help="Pobierz w formacie Markdown"
            )

            # Word Export
            word_buffer = test_runner.export_to_word(
                st.session_state.rated_responses,
                results,
                settings
            )
            st.download_button(
                label="📄 Word",
                data=word_buffer,
                file_name=f"prompt_ab_test_{timestamp}.docx",
                mime="application/vnd.openxmlformats-officedocument.wordprocessingml.document",
                use_container_width=True,
                help="Pobierz w formacie Word"
            )

# Tab 4: Logs
with tab4:
    st.header("📋 Logi Testów")

    if st.button("🗑️ Wyczyść logi"):
        clear_logs()
        st.rerun()

    if st.session_state.logs:
        log_text = "\n".join(st.session_state.logs)
        st.text_area(
            "Logi:",
            value=log_text,
            height=500,
            disabled=True,
            label_visibility="collapsed"
        )
    else:
        st.info("ℹ️ Brak logów do wyświetlenia")

# Footer
st.divider()
st.markdown(
    """
    <div style='text-align: center; color: gray; font-size: 0.9em;'>
    Prompt A/B Tester v2.0 (Streamlit) | Created by Heuristica.pl - Marek Staniszewski
    </div>
    """,
    unsafe_allow_html=True
)