components/prediction_components.py

"""Components for AC prediction and visualization"""
import pandas as pd
import numpy as np
from datetime import datetime, timedelta
import plotly.graph_objects as go

try:
    from scipy.optimize import curve_fit
    from scipy import stats
    HAS_SCIPY = True
except ImportError:
    HAS_SCIPY = False
    # Fallback to numpy polynomial fitting
    def curve_fit(func, xdata, ydata, p0=None, maxfev=5000, bounds=None):
        # Simple fallback - just use polynomial fitting
        if func.__name__ == 'exponential_growth':
            # Linearize exponential: log(y) = log(a) + b*x
            log_y = np.log(ydata + 0.01)  # Add small constant to avoid log(0)
            coeffs = np.polyfit(xdata, log_y, 1)
            a = np.exp(coeffs[1])
            b = coeffs[0]
            c = 0.01
            return [a, b, c], None
        elif func.__name__ == 'logistic_growth':
            # Better fallback for logistic using data characteristics
            # Estimate L (max value) as slightly above current max
            L = min(1.0, max(ydata) * 1.2)  # Cap at 1.0

            # Estimate x0 (midpoint) - when growth would be fastest
            # For now, project forward from current trend
            if len(xdata) > 1:
                # Simple linear projection to estimate when we'd hit midpoint
                slope = (ydata[-1] - ydata[0]) / (xdata[-1] - xdata[0])
                if slope > 0:
                    # Estimate days to reach L/2
                    midpoint_value = L / 2
                    if ydata[-1] < midpoint_value:
                        days_to_midpoint = (midpoint_value - ydata[-1]) / slope
                        x0 = xdata[-1] + days_to_midpoint
                    else:
                        x0 = np.median(xdata)
                else:
                    x0 = np.median(xdata)
            else:
                x0 = np.median(xdata)

            # Estimate k (growth rate) based on current growth
            k = 0.003  # Conservative default

            return [L, k, x0], None
        elif func.__name__ == 'power_law':
            # Linearize power law: log(y) = log(a) + b*log(x)
            log_x = np.log(xdata + 1)
            log_y = np.log(ydata + 0.01)
            coeffs = np.polyfit(log_x, log_y, 1)
            return [np.exp(coeffs[1]), coeffs[0]], None
        return p0, None

def exponential_growth(x, a, b, c):
    """Exponential growth function: y = a * exp(b * x) + c"""
    return a * np.exp(b * x) + c

def logistic_growth(x, L, k, x0):
    """Logistic growth function: y = L / (1 + exp(-k*(x-x0)))"""
    return L / (1 + np.exp(-k * (x - x0)))

def power_law(x, a, b):
    """Power law function: y = a * x^b"""
    return a * np.power(x, b)

def create_ac_prediction_chart(df, domain_filter="All", model_type_filter="All"):
    """Create a prediction chart showing when AC will reach 99%

    Args:
        df: DataFrame with model data
        domain_filter: Domain to filter by (All, Banking, Healthcare, etc.)
        model_type_filter: Model type to filter by (All, Open Source, Proprietary)
    """

    # Clean up domain filter (remove emoji prefix if present)
    if domain_filter.startswith('🌐'):
        domain_filter = "All"
    elif domain_filter.startswith('🏦'):
        domain_filter = "Banking"
    elif domain_filter.startswith('🏥'):
        domain_filter = "Healthcare"
    elif domain_filter.startswith('🛡️'):
        domain_filter = "Insurance"
    elif domain_filter.startswith('💰'):
        domain_filter = "Investment"
    elif domain_filter.startswith('📱'):
        domain_filter = "Telecom"

    # Determine which AC column to use based on domain filter
    if domain_filter != "All":
        ac_column = f'{domain_filter} AC'
        # Check if domain-specific column exists
        if ac_column not in df.columns:
            ac_column = 'Avg AC'
    else:
        ac_column = 'Avg AC'

    # Filter data to only include models with valid release dates and AC scores
    df_clean = df.dropna(subset=['Release Date', ac_column])
    df_clean = df_clean[df_clean[ac_column] > 0]

    # Apply model type filter
    if model_type_filter == "Open Source":
        df_clean = df_clean[df_clean['Model Type'] == 'Open source']
    elif model_type_filter == "Proprietary":
        df_clean = df_clean[df_clean['Model Type'] == 'Proprietary']

    # Rename the AC column to 'Avg AC' for consistent processing (only if different)
    if ac_column != 'Avg AC':
        # Drop the original 'Avg AC' column if it exists to avoid duplicates
        if 'Avg AC' in df_clean.columns:
            df_clean = df_clean.drop(columns=['Avg AC'])
        df_clean = df_clean.rename(columns={ac_column: 'Avg AC'})

    # Make a copy to avoid any issues with the original data
    df_clean = df_clean.copy()

    # Handle both YYYY-MM and YYYY-MM-DD formats
    if df_clean['Release Date'].str.contains('-').all():
        # Check if it's YYYY-MM format (no day component)
        if df_clean['Release Date'].str.count('-').iloc[0] == 1:
            # Add '-01' to make it a valid date
            df_clean['Release Date'] = pd.to_datetime(df_clean['Release Date'] + '-01')
        else:
            df_clean['Release Date'] = pd.to_datetime(df_clean['Release Date'])
    else:
        df_clean['Release Date'] = pd.to_datetime(df_clean['Release Date'])

    # Sort by release date
    df_clean = df_clean.sort_values('Release Date')

    # Create a running maximum (best performance achieved up to each date)
    df_clean['Cumulative_Max_AC'] = df_clean['Avg AC'].expanding().max()

    # Group by date and take the cumulative maximum for each date
    df_best = df_clean.groupby('Release Date')['Cumulative_Max_AC'].max().reset_index()
    df_best.columns = ['Release Date', 'Avg AC']

    # Apply cumulative maximum again to ensure monotonic increase
    df_best['Avg AC'] = df_best['Avg AC'].cummax()

    # Convert dates to days since first release for curve fitting
    first_date = df_best['Release Date'].min()
    df_best['Days'] = (df_best['Release Date'] - first_date).dt.days

    # Prepare data for fitting
    x_data = df_best['Days'].values
    y_data = df_best['Avg AC'].values

    # With limited data (only 2 performance levels), use simple conservative linear projection
    # Don't try to fit complex curves that will overfit

    best_model = 'linear'

    # Calculate simple linear trend
    if len(x_data) > 1:
        # Basic linear regression
        z = np.polyfit(x_data, y_data, 1)
        slope = z[0]

        # Apply conservative adjustment (assume diminishing returns)
        conservative_slope = slope * 0.5  # Assume 50% slower future improvements

        # Create conservative linear projection
        best_fit = [conservative_slope, y_data[-1] - conservative_slope * x_data[-1]]

        # Calculate R² for the linear fit
        p = np.poly1d(z)
        y_pred = p(x_data)
        best_r2 = 1 - (np.sum((y_data - y_pred)**2) / np.sum((y_data - y_data.mean())**2))
    else:
        # Single data point - use minimal growth
        best_fit = [0.0001, y_data[0]]
        best_r2 = 0.0

    # Generate prediction timeline
    future_days = np.arange(0, 5475, 30)  # 15 years in 30-day intervals

    # Simple conservative linear projection
    p = np.poly1d(best_fit)
    future_ac = p(future_days)

    # Cap predictions at 1.0
    future_ac = np.minimum(future_ac, 1.0)

    # Find when we reach 99%
    target_ac = 0.99
    crossing_idx = np.where(future_ac >= target_ac)[0]

    if len(crossing_idx) > 0:
        days_to_99 = future_days[crossing_idx[0]]
        date_99 = first_date + timedelta(days=int(days_to_99))
        months_from_now = (date_99 - datetime.now()).days / 30.4
    else:
        date_99 = None
        months_from_now = None

    # Convert future days to dates
    future_dates = [first_date + timedelta(days=int(d)) for d in future_days]

    # Create the plot
    fig = go.Figure()

    # Add confidence bands FIRST (so they appear behind other traces)
    if best_model and best_fit is not None:
        # Generate smooth confidence bands
        future_std = 0.05  # Base uncertainty
        confidence_multiplier = np.linspace(1.0, 2.0, len(future_dates))

        upper_bound = np.minimum(future_ac + future_std * confidence_multiplier, 1.0)
        lower_bound = np.maximum(future_ac - future_std * confidence_multiplier, 0)

        # Add confidence band as filled area
        fig.add_trace(go.Scatter(
            x=future_dates + future_dates[::-1],
            y=list(upper_bound) + list(lower_bound[::-1]),
            fill='toself',
            fillcolor='rgba(16, 152, 247, 0.05)',
            line=dict(width=0),
            showlegend=False,
            hoverinfo='skip',
            name='Uncertainty'
        ))

    # Add vendor info and additional metrics
    df_with_vendor = df_clean.copy()
    if 'Vendor' in df.columns:
        vendor_map = df.set_index('Model')['Vendor'].to_dict()
        df_with_vendor['Vendor'] = df_with_vendor['Model'].map(vendor_map).fillna('Unknown')
    else:
        df_with_vendor['Vendor'] = 'Unknown'

    if 'Model Type' in df.columns:
        type_map = df.set_index('Model')['Model Type'].to_dict()
        df_with_vendor['Model Type'] = df_with_vendor['Model'].map(type_map).fillna('Unknown')
    else:
        df_with_vendor['Model Type'] = 'Unknown'

    # Calculate additional metrics for each model
    df_with_vendor['Gap_to_99'] = 0.99 - df_with_vendor['Avg AC']
    df_with_vendor['Gap_to_Best'] = df_with_vendor['Cumulative_Max_AC'] - df_with_vendor['Avg AC']

    # Get cost info if available
    if 'Avg Total Cost' in df.columns:
        cost_map = df.set_index('Model')['Avg Total Cost'].to_dict()
        df_with_vendor['Cost'] = df_with_vendor['Model'].map(cost_map).fillna(0)
    else:
        df_with_vendor['Cost'] = 0

    # Check if each model improved the frontier
    df_with_vendor['Is_Frontier'] = df_with_vendor['Avg AC'] >= df_with_vendor['Cumulative_Max_AC'] - 0.001  # Small tolerance for float comparison

    # Create frontier status text
    frontier_status = []
    for idx, row in df_with_vendor.iterrows():
        if row['Is_Frontier']:
            frontier_status.append('✅ Advanced SOTA')
        else:
            frontier_status.append('❌ Below existing best')

    vendor_info = df_with_vendor['Vendor'].values
    model_type = df_with_vendor['Model Type'].values
    gap_to_99 = df_with_vendor['Gap_to_99'].values
    gap_to_best = df_with_vendor['Gap_to_Best'].values
    cost_info = df_with_vendor['Cost'].values

    # Add historical data points with comprehensive hover
    fig.add_trace(go.Scatter(
        x=df_clean['Release Date'],
        y=df_clean['Avg AC'],
        mode='markers',
        name='Individual Models',
        marker=dict(
            size=14,
            color='rgba(227, 84, 84, 0.8)',
            line=dict(width=2, color='rgba(255, 255, 255, 0.6)'),
            symbol='circle'
        ),
        customdata=list(zip(vendor_info, model_type, gap_to_99, gap_to_best, cost_info, frontier_status)),
        hovertemplate=(
            '<b style="font-size: 18px; color: #E35454;">%{text}</b><br>'
            '<br>'
            '<b style="color: #1098F7;">Model Information:</b><br>'
            '• <b>Vendor:</b> %{customdata[0]}<br>'
            '• <b>Type:</b> %{customdata[1]}<br>'
            '• <b>Released:</b> %{x|%B %Y}<br>'
            '• <b>Frontier Status:</b> %{customdata[5]}<br>'
            '<br>'
            '<b style="color: #FFD700;">Performance Metrics:</b><br>'
            '• <b>Action Completion:</b> <span style="font-size: 20px; color: #FFD700;">%{y:.1%}</span><br>'
            '• <b>Gap to 99%:</b> <span style="color: #FF6B6B;">-%{customdata[2]:.1%}</span><br>'
            '• <b>Behind Best:</b> <span style="color: #FFA500;">-%{customdata[3]:.1%}</span><br>'
            '<br>'
            '<b style="color: #28a745;">Cost Efficiency:</b><br>'
            '• <b>Avg Session Cost:</b> $%{customdata[4]:.4f}<br>'
            '<br>'
            '<i style="color: #B1B5B9; font-size: 13px;">Performance at release time</i>'
            '<extra></extra>'
        ),
        text=df_clean['Model'].values,
        hoverlabel=dict(
            bgcolor='rgba(26, 26, 46, 0.95)',
            bordercolor='rgba(227, 84, 84, 0.5)',
            font=dict(size=14, color='#F5F6F7', family='Geist, sans-serif'),
            align='left',
            namelength=-1
        )
    ))

    # Calculate improvement metrics for hover
    df_best['Improvement'] = df_best['Avg AC'].diff().fillna(0)
    df_best['Improvement_Pct'] = (df_best['Avg AC'].pct_change() * 100).fillna(0)
    df_best['Gap_to_99'] = 0.99 - df_best['Avg AC']

    # Find which model is responsible for the best performance at each date
    best_model_at_date = []
    for date_val in df_best['Release Date']:
        # Find all models up to and including this date
        models_up_to_date = df_clean[df_clean['Release Date'] <= date_val]
        if not models_up_to_date.empty:
            # Find the model with the highest AC score up to this date
            best_idx = models_up_to_date['Avg AC'].idxmax()
            best_model_at_date.append(models_up_to_date.loc[best_idx, 'Model'])
        else:
            best_model_at_date.append('Unknown')

    # Add best performance line with enhanced metrics
    fig.add_trace(go.Scatter(
        x=df_best['Release Date'],
        y=df_best['Avg AC'],
        mode='lines+markers',
        name='Best Performance Trend',
        line=dict(color='#E35454', width=4, shape='linear'),
        marker=dict(
            size=16,
            color='#E35454',
            symbol='diamond',
            line=dict(width=2, color='white')
        ),
        customdata=list(zip(
            df_best['Improvement'].values,
            df_best['Improvement_Pct'].values,
            df_best['Gap_to_99'].values,
            best_model_at_date
        )),
        hovertemplate=(
            '<b style="font-size: 20px; color: #E35454;">📈 Best Performance Frontier</b><br>'
            '<br>'
            '<b>Date:</b> %{x|%B %Y}<br>'
            '<b>Leading Model:</b> %{customdata[3]}<br>'
            '<b>Cumulative Best AC:</b> <span style="font-size: 20px; color: #FFD700;">%{y:.1%}</span><br>'
            '<br>'
            '<b>Progress Metrics:</b><br>'
            '• Improvement: <span style="color: #28a745;">+%{customdata[0]:.1%}</span><br>'
            '• Growth Rate: <span style="color: #28a745;">+%{customdata[1]:.1f}%</span><br>'
            '• Gap to 99%: <span style="color: #1098F7;">%{customdata[2]:.1%}</span><br>'
            '<br>'
            '<i style="color: #B1B5B9; font-size: 13px;">This represents the best performance achieved by any model up to this date</i>'
            '<extra></extra>'
        ),
        hoverlabel=dict(
            bgcolor='rgba(26, 26, 46, 0.95)',
            bordercolor='rgba(227, 84, 84, 0.5)',
            font=dict(size=14, color='#F5F6F7', family='Geist, sans-serif'),
            align='left',
            namelength=-1
        )
    ))

    # Calculate months from now for each prediction point
    months_from_now_list = [(date - datetime.now()).days / 30.4 for date in future_dates]
    years_from_now_list = [m / 12 for m in months_from_now_list]

    # Add prediction line with comprehensive metrics
    fig.add_trace(go.Scatter(
        x=future_dates,
        y=future_ac,
        mode='lines',
        name=f'Prediction ({best_model.capitalize()})',
        line=dict(color='#1098F7', width=4, dash='dash'),
        opacity=0.8,
        customdata=list(zip(
            [max(0, 0.99 - y) for y in future_ac],
            months_from_now_list,
            years_from_now_list,
            [best_r2] * len(future_ac)
        )),
        hovertemplate=(
            '<b style="font-size: 18px; color: #1098F7;">🔮 AI Performance Prediction</b><br>'
            '<br>'
            '<b style="color: #FFD700;">Forecast Details:</b><br>'
            '• <b>Date:</b> %{x|%B %Y}<br>'
            '• <b>Predicted AC:</b> <span style="font-size: 20px; color: #FFD700;">%{y:.1%}</span><br>'
            '• <b>Gap to 99%:</b> <span style="color: #FF6B6B;">-%{customdata[0]:.1%}</span><br>'
            '<br>'
            '<b style="color: #28a745;">Timeline:</b><br>'
            '• <b>Months from now:</b> %{customdata[1]:.0f} months<br>'
            '• <b>Years from now:</b> %{customdata[2]:.1f} years<br>'
            '<br>'
            '<b style="color: #1098F7;">Model Confidence:</b><br>'
            f'• <b>Algorithm:</b> {best_model.capitalize()}<br>'
            '• <b>R² Score:</b> %{customdata[3]:.3f}<br>'
            '<br>'
            '<i style="color: #B1B5B9; font-size: 13px;">Based on historical performance trends</i>'
            '<extra></extra>'
        ),
        hoverlabel=dict(
            bgcolor='rgba(26, 26, 46, 0.95)',
            bordercolor='rgba(16, 152, 247, 0.5)',
            font=dict(size=14, color='#F5F6F7', family='Geist, sans-serif'),
            align='left',
            namelength=-1
        )
    ))

    # Add 99% threshold line with enhanced styling
    fig.add_hline(
        y=0.99,
        line_dash="dash",
        line_color="rgba(40, 167, 69, 0.4)",
        line_width=2,
        annotation=dict(
            text="<b>Enterprise-Grade Threshold (99%)</b>",
            font=dict(size=13, color='#28a745', family='Geist, sans-serif'),
            bgcolor='rgba(40, 167, 69, 0.15)',
            bordercolor='#28a745',
            borderwidth=1,
            borderpad=4
        ),
        annotation_position="right"
    )

    # Add marker for 99% crossing point with enhanced visibility
    if date_99:
        # Calculate days until achievement
        days_until = (date_99 - datetime.now()).days

        fig.add_trace(go.Scatter(
            x=[date_99],
            y=[0.99],
            mode='markers+text',
            name='🎯 99% Achievement',
            marker=dict(
                size=28,
                color='#28a745',
                symbol='star',
                line=dict(width=3, color='white')
            ),
            text=[f'<b>{date_99.strftime("%b %Y")}</b>'],
            textposition='top center',
            textfont=dict(size=16, color='#28a745', family='Geist, sans-serif'),
            hovertemplate=(
                '<b style="font-size: 18px; color: #28a745;">🎯 ENTERPRISE-READY MILESTONE</b><br>'
                '<br>'
                f'<b>Achievement Date:</b> <span style="font-size: 16px;">{date_99.strftime("%B %Y")}</span><br>'
                f'<b>Time from today:</b> <span style="font-size: 16px; color: #FFD700;">{months_from_now:.0f} months</span><br>'
                f'<b>Days remaining:</b> {days_until} days<br>'
                f'<b>Years:</b> {months_from_now/12:.1f} years<br>'
                '<br>'
                '<b style="color: #1098F7;">Strategic Implications:</b><br>'
                f'• Early adopters gain {months_from_now:.0f}-month advantage<br>'
                '• Infrastructure investment critical now<br>'
                '• 99% reliability enables production deployment<br>'
                '<extra></extra>'
            ),
            hoverlabel=dict(
                bgcolor='rgba(26, 26, 46, 0.95)',
                bordercolor='rgba(40, 167, 69, 0.5)',
                font=dict(size=14, color='#F5F6F7', family='Geist, sans-serif'),
                align='left',
                namelength=-1
            )
        ))

    # Update layout with improved title showing active filters
    filter_text = ""
    if domain_filter != "All":
        filter_text += f" - {domain_filter} Domain"
    if model_type_filter != "All":
        if filter_text:
            filter_text += f", {model_type_filter} Models"
        else:
            filter_text += f" - {model_type_filter} Models"

    title_text = f"<span style='font-size: 24px;'>🚀 When Will AI Agents Reach Enterprise-Grade Reliability?</span>"
    if filter_text:
        title_text += f"<br><span style='font-size: 14px; color: #1098F7;'>{filter_text}</span>"

    if date_99 and months_from_now:
        if months_from_now > 0:
            title_text += f"<br><span style='font-size: 16px; color: #B1B5B9;'>Prediction: <b style='color: #FFD700;'>{date_99.strftime('%B %Y')}</b> (~{months_from_now:.0f} months)</span>"
        else:
            title_text += f"<br><span style='font-size: 16px; color: #28a745;'>Already achieved!</span>"
    else:
        title_text += f"<br><span style='font-size: 16px; color: #B1B5B9;'>Tracking performance improvements...</span>"

    fig.update_layout(
        title=dict(
            text=title_text,
            font=dict(size=20, family="Geist, sans-serif", color="#F5F6F7"),
            x=0.5,
            xanchor='center'
        ),
        xaxis=dict(
            title=dict(
                text="<b>Release Date</b>",
                font=dict(size=16, family="Geist, sans-serif", color="#F5F6F7"),
                standoff=20
            ),
            tickfont=dict(size=12, family="Geist Mono, monospace", color="#B1B5B9"),
            gridcolor="rgba(245, 246, 247, 0.08)",
            zerolinecolor="rgba(245, 246, 247, 0.15)",
            showgrid=True,
            gridwidth=1,
            tickangle=0,
            tickformat='%b %Y',
            showspikes=True,
            spikecolor="rgba(245, 246, 247, 0.3)",
            spikethickness=1,
            spikemode='across',
            spikedash='dot',
            range=[df_clean['Release Date'].min() - timedelta(days=60),
                   min(datetime.now() + timedelta(days=800), future_dates[-1] if future_dates else datetime.now())]
        ),
        yaxis=dict(
            title=dict(
                text="<b>Action Completion (AC)</b>",
                font=dict(size=16, family="Geist, sans-serif", color="#F5F6F7"),
                standoff=20
            ),
            tickfont=dict(size=12, family="Geist Mono, monospace", color="#B1B5B9"),
            gridcolor="rgba(245, 246, 247, 0.08)",
            zerolinecolor="rgba(245, 246, 247, 0.15)",
            showgrid=True,
            gridwidth=1,
            tickformat='.0%',
            dtick=0.1,
            showspikes=True,
            spikecolor="rgba(245, 246, 247, 0.3)",
            spikethickness=1,
            spikemode='across',
            spikedash='dot',
            range=[-0.05, 1.08]
        ),
        plot_bgcolor="rgba(1, 9, 26, 0.98)",
        paper_bgcolor="rgba(1, 9, 26, 0.98)",
        height=650,
        margin=dict(l=90, r=100, t=120, b=90),
        hovermode='closest',
        hoverdistance=30,
        spikedistance=50,
        legend=dict(
            bgcolor="rgba(1, 9, 26, 0.9)",
            bordercolor="rgba(245, 246, 247, 0.3)",
            borderwidth=2,
            font=dict(size=12, family="Geist, sans-serif", color="#F5F6F7"),
            x=0.02,
            y=0.98,
            xanchor='left',
            yanchor='top',
            orientation='v',
            itemsizing='constant',
            itemwidth=40,
            tracegroupgap=5,
            title=dict(
                text='<b>Legend</b>',
                font=dict(size=13, color='#F5F6F7')
            )
        ),
        showlegend=True,
        annotations=[
            dict(
                text=f"<b>Model:</b> Conservative Linear | <b>Note:</b> Limited data - projection assumes diminishing returns",
                xref="paper", yref="paper",
                x=0.01, y=-0.12,
                showarrow=False,
                font=dict(size=11, color="#B1B5B9", family="Geist, sans-serif"),
                bgcolor="rgba(1, 9, 26, 0.9)",
                bordercolor="rgba(245, 246, 247, 0.3)",
                borderwidth=1,
                borderpad=4
            )
        ]
    )


    # Get the current best AC value (last value in y_data)
    current_best_ac = y_data[-1] if len(y_data) > 0 else None

    return fig, date_99, months_from_now, current_best_ac