import pandas as pd
import jiwer
import plotly.express as px
import plotly.graph_objects as go
from constants import CSV_FILE, TABLE_HEADERS
# Ses işleme için (Gradio otomatik halletse de importlar iyi olur)
import os

# --- YARDIMCI FONKSİYONLAR (Madalya, Badge, Link) ---
def get_medal(rank):
    if rank == 1: return "<span class='medal'>🥇</span>"
    if rank == 2: return "<span class='medal'>🥈</span>"
    if rank == 3: return "<span class='medal'>🥉</span>"
    return f"<span class='rank-num'>{rank}</span>"

def get_badge(value, high_is_good=False, is_rtf=False):
    try:
        val = float(value)
        if is_rtf: # RTF için farklı eşikler (Düşük iyi)
            if val < 0.05: css = "badge-excellent"
            elif val < 0.10: css = "badge-good"
            elif val < 0.20: css = "badge-average"
            else: css = "badge-poor"
            return f"<span class='badge {css}'>{val:.2f}x</span>"
        else: # WER/CER için (Düşük iyi)
            if val < 10.0: css = "badge-excellent"
            elif val < 20.0: css = "badge-good"
            elif val < 50.0: css = "badge-average"
            else: css = "badge-poor"
            return f"<span class='badge {css}'>{val:.2f}%</span>"
    except: return str(value)

def make_clickable_model(model_name, link):
    if pd.isna(link) or link == "" or str(link).lower() == "nan": return model_name
    return f"<a href='{link}' target='_blank'>{model_name}</a>"

def parse_params_to_num(param_str):
    """'1.55B', '769M' gibi stringleri sayıya çevirir."""
    try:
        if pd.isna(param_str): return 0
        param_str = str(param_str).upper()
        if param_str.endswith('B'): return float(param_str[:-1]) * 1e9
        if param_str.endswith('M'): return float(param_str[:-1]) * 1e6
        if param_str.endswith('K'): return float(param_str[:-1]) * 1e3
        return float(param_str)
    except: return 0

# --- VERİ YÜKLEME VE İŞLEME ---
def load_raw_data():
    try:
        df = pd.read_csv(CSV_FILE)
        # Sayısal sütunları temizle
        for col in df.columns:
            if "WER" in col or "CER" in col or "RTF" in col:
                df[col] = pd.to_numeric(df[col], errors='coerce')
        
        # Parametreleri sayıya çevir (Grafik için)
        df['Params_Num'] = df['Params'].apply(parse_params_to_num)
        
        # Ortalama WER'e göre sırala
        if "Average_WER" in df.columns:
            df = df.sort_values(by="Average_WER", ascending=True).reset_index(drop=True)
        return df
    except Exception as e:
        print(f"Data loading error: {e}")
        return pd.DataFrame()

# --- GELİŞMİŞ FİLTRELEME ---
def filter_data(search_query, model_types, max_params):
    df = load_raw_data()
    if df.empty: return df

    # 1. Metin Arama
    if search_query:
        df = df[df.apply(lambda row: row.astype(str).str.contains(search_query, case=False).any(), axis=1)]
    
    # 2. Model Tipi Filtresi
    if model_types:
        df = df[df['Model_Type'].isin(model_types)]
    
    # 3. Parametre Boyutu Filtresi (Slider değeri Milyon cinsinden)
    if max_params < 2000: # Slider maksimumda değilse filtrele
        max_params_val = max_params * 1e6
        df = df[df['Params_Num'] <= max_params_val]
        
    return df.reset_index(drop=True)

# --- HTML TABLO OLUŞTURMA ---
def dataframe_to_html(df_filtered):
    if df_filtered.empty: return "<div style='text-align:center; padding:20px;'>No models found matching criteria.</div>"
    
    # Orijinal linkleri almak için ham veriyi oku
    df_raw = pd.read_csv(CSV_FILE)
    
    html = "<table class='leaderboard-table'><thead><tr>"
    for header in TABLE_HEADERS: html += f"<th>{header}</th>"
    html += "</tr></thead><tbody>"
    
    for index, row in df_filtered.iterrows():
        rank = index + 1
        medal = get_medal(rank)
        model_name = row['Model']
        try:
            link = df_raw.loc[df_raw['Model'] == model_name, 'Model_Link'].values[0]
            model_link = make_clickable_model(model_name, link)
        except: model_link = model_name
        
        # Rozetleri oluştur
        cv_wer = get_badge(row.get('CommonVoice_WER', '-'))
        cv_cer = row.get('CommonVoice_CER', '-')
        fl_wer = get_badge(row.get('FLEURS_WER', '-'))
        fl_cer = row.get('FLEURS_CER', '-')
        avg_wer = get_badge(row.get('Average_WER', '-'))
        rtf = get_badge(row.get('RTF', '-'), is_rtf=True)
        
        html += f"""<tr>
            <td style='text-align:center;'>{medal}</td><td>{model_link}</td>
            <td>{cv_wer}</td><td>{cv_cer}</td><td>{fl_wer}</td><td>{fl_cer}</td>
            <td>{avg_wer}</td><td>{row.get('Params','-')}</td><td>{rtf}</td>
            <td>{row.get('Model_Type','-')}</td><td>{row.get('License','-')}</td>
        </tr>"""
    html += "</tbody></table>"
    return html

# --- PLOTLY GRAFİKLERİ ---
def create_plots(df):
    if df.empty: return None, None

    # 1. Boyut vs. Başarım (Bubble Chart)
    fig_size = px.scatter(
        df, x="Params_Num", y="Average_WER",
        size="Params_Num", color="Model_Type",
        hover_name="Model", text="Model",
        labels={"Params_Num": "Parameters (Log Scale)", "Average_WER": "Average WER (%)", "Model_Type": "Type"},
        title="🤖 Model Size vs. Performance (Lower-Left is Better)",
        log_x=True, height=400,
        color_discrete_sequence=px.colors.qualitative.Prism # Mavi-Yeşil temaya uygun renkler
    )
    fig_size.update_traces(textposition='top center', marker=dict(opacity=0.7, line=dict(width=1, color='DarkSlateGrey')))
    fig_size.update_layout(
        plot_bgcolor='rgba(0,0,0,0)', paper_bgcolor='rgba(0,0,0,0)',
        font_color="var(--text-primary)",
        xaxis=dict(showgrid=True, gridcolor='var(--border-color)'),
        yaxis=dict(showgrid=True, gridcolor='var(--border-color)')
    )

    # 2. Hız vs. Başarım (Scatter Plot)
    fig_speed = px.scatter(
        df, x="RTF", y="Average_WER",
        color="Model_Type", hover_name="Model", text="Model",
        labels={"RTF": "Speed (RTF - Lower is Faster)", "Average_WER": "Average WER (%)"},
        title="⚡ Speed vs. Performance (Lower-Left is Better)",
        height=400, color_discrete_sequence=px.colors.qualitative.Prism
    )
    fig_speed.update_traces(textposition='top center', marker=dict(size=12, opacity=0.8))
    fig_speed.update_layout(
        plot_bgcolor='rgba(0,0,0,0)', paper_bgcolor='rgba(0,0,0,0)',
        font_color="var(--text-primary)",
        xaxis=dict(showgrid=True, gridcolor='var(--border-color)'),
        yaxis=dict(showgrid=True, gridcolor='var(--border-color)')
    )
    
    return fig_size, fig_speed

# --- METİN SANDBOX HESAPLAMA ---
def calculate_metrics_text(ref, hyp, norm=True):
    if not ref or not hyp: return "-","-","Please fill both areas."
    if norm:
        trans = jiwer.Compose([jiwer.ToLowerCase(), jiwer.RemovePunctuation(), jiwer.RemoveMultipleSpaces(), jiwer.Strip()])
        ref = trans(ref); hyp = trans(hyp)
    try:
        wer = jiwer.wer(ref, hyp)*100; cer = jiwer.cer(ref, hyp)*100
        return f"{wer:.2f}%", f"{cer:.2f}%", f"**Ref:** {ref}\n\n**Hyp:** {hyp}"
    except: return "Error", "Error", "Calculation failed."

# --- SESLİ SANDBOX (DEMO MODEL) ---
# Gerçek bir model yüklemek CPU'da yavaş olabilir, bu yüzden 'tiny' kullanıyoruz.
from transformers import pipeline

# Modeli global olarak bir kere yükle (Hafif model)
try:
    asr_pipeline = pipeline("automatic-speech-recognition", model="openai/whisper-tiny", device="cpu")
except:
    asr_pipeline = None
    print("Warning: Could not load Whisper-tiny for audio sandbox.")

def transcribe_audio(audio_path):
    if not asr_pipeline: return "ASR Model not loaded successfully."
    if audio_path is None: return "Please record or upload audio."
    
    try:
        # Gradio audio path'ini doğrudan pipeline'a ver
        result = asr_pipeline(audio_path)
        return result["text"]
    except Exception as e:
        return f"Transcription Error: {e}"