visualization/plot_generator.py

# visualization/plot_generator.py
"""Main plotting functionality for similarity analysis"""

import pandas as pd
import plotly.graph_objects as go
from plotly.subplots import make_subplots
from typing import Tuple, Optional, Union
import sys
import os

# Add parent directory to path for imports
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))

class PlotGenerator:
    """Handles creation of plotly visualizations"""

    def __init__(self, data_loader):
        self.data_loader = data_loader

    def compute_category_correlation_method2(self, category_key: str, target_series: pd.Series) -> float:
        """

        Compute correlation using Method 2: Correlate each model, then average correlations.

        This matches the bar chart methodology.



        Args:

            category_key: Category name like 'vision', 'captions_neural', etc.

            target_series: The series to correlate with (e.g., brain measure or human judgement)



        Returns:

            Average correlation across all models in the category

        """
        import numpy as np

        # Get models in this category
        models = [model[0] for model in self.data_loader.model_categories[category_key]]

        if not models:
            return 0.0

        # Filter to available models
        data = self.data_loader.data
        available_models = [m for m in models if m in data.columns]

        if not available_models:
            return 0.0

        # Compute correlation for each model
        correlations = []
        for model in available_models:
            corr = data[model].corr(target_series)
            if not np.isnan(corr):
                correlations.append(corr)

        # Return average correlation
        if correlations:
            return np.mean(correlations)
        else:
            return 0.0

    @staticmethod
    def add_image_hover_to_html(html_str: str) -> str:
        """Add custom JavaScript to enable image preview on hover for image pairs"""
        custom_code = """

<style>

    #image-hover-tooltip {

        position: fixed;

        display: none;

        background: white;

        border: 2px solid #333;

        border-radius: 8px;

        padding: 10px;

        box-shadow: 0 4px 6px rgba(0,0,0,0.1);

        z-index: 10000;

        max-width: 500px;

        pointer-events: none;

    }

    #image-hover-tooltip .tooltip-header {

        font-weight: bold;

        margin-bottom: 8px;

        font-size: 13px;

        color: #333;

    }

    #image-hover-tooltip .image-container {

        display: flex;

        gap: 10px;

        align-items: flex-start;

    }

    #image-hover-tooltip .image-wrapper {

        flex: 1;

        text-align: center;

    }

    #image-hover-tooltip .image-label {

        font-size: 10px;

        color: #666;

        margin-bottom: 3px;

        font-weight: bold;

    }

    #image-hover-tooltip img {

        max-width: 230px;

        max-height: 230px;

        display: block;

        margin: 0 auto;

        border: 1px solid #ddd;

    }

</style>



<div id="image-hover-tooltip">

    <div class="tooltip-header"></div>

    <div class="image-container">

        <div class="image-wrapper">

            <div class="image-label">Image 1</div>

            <img id="tooltip-img1" src="" alt="Image 1">

        </div>

        <div class="image-wrapper">

            <div class="image-label">Image 2</div>

            <img id="tooltip-img2" src="" alt="Image 2">

        </div>

    </div>

</div>



<script>

document.addEventListener('DOMContentLoaded', function() {

    const tooltip = document.getElementById('image-hover-tooltip');

    const tooltipHeader = tooltip.querySelector('.tooltip-header');

    const tooltipImg1 = document.getElementById('tooltip-img1');

    const tooltipImg2 = document.getElementById('tooltip-img2');



    // Get all plotly divs

    const plotDivs = document.querySelectorAll('.plotly-graph-div');



    plotDivs.forEach(plotDiv => {

        plotDiv.on('plotly_hover', function(data) {

            if (data.points && data.points.length > 0) {

                const point = data.points[0];



                // Check if customdata exists and has image URLs

                // Format: [idx, image_1_name, image_2_name, stim_1_url, stim_2_url]

                if (point.customdata && point.customdata.length >= 5) {

                    const img1Name = point.customdata[1];

                    const img2Name = point.customdata[2];

                    const img1Url = point.customdata[3];

                    const img2Url = point.customdata[4];



                    if (img1Url && img2Url) {

                        tooltipHeader.textContent = `${img1Name} vs ${img2Name}`;

                        tooltipImg1.src = img1Url;

                        tooltipImg2.src = img2Url;

                        tooltip.style.display = 'block';

                    }

                }

                // Handle 4-element customdata (old format from 3D plot)

                else if (point.customdata && point.customdata.length === 4) {

                    const img1Name = point.customdata[0];

                    const img2Name = point.customdata[1];

                    const img1Url = point.customdata[2];

                    const img2Url = point.customdata[3];



                    if (img1Url && img2Url) {

                        tooltipHeader.textContent = `${img1Name} vs ${img2Name}`;

                        tooltipImg1.src = img1Url;

                        tooltipImg2.src = img2Url;

                        tooltip.style.display = 'block';

                    }

                }

            }

        });



        plotDiv.on('plotly_unhover', function(data) {

            tooltip.style.display = 'none';

        });



        // Update tooltip position on mouse move

        plotDiv.addEventListener('mousemove', function(e) {

            if (tooltip.style.display === 'block') {

                const x = e.clientX + 15;

                const y = e.clientY + 15;



                // Keep tooltip on screen

                const tooltipRect = tooltip.getBoundingClientRect();

                const maxX = window.innerWidth - tooltipRect.width - 10;

                const maxY = window.innerHeight - tooltipRect.height - 10;



                tooltip.style.left = Math.min(x, maxX) + 'px';

                tooltip.style.top = Math.min(y, maxY) + 'px';

            }

        });

    });

});

</script>

"""
        # Insert custom code before closing body tag
        return html_str.replace('</body>', custom_code + '</body>')
    
    def get_model_data(self, ml_model_selection: Union[str, int]) -> Tuple[pd.Series, str]:
        """Get model data - either individual model or category average"""
        data = self.data_loader.data

        # Define category display names
        category_labels = {
            'vision': 'Vision Models (Images) - Average',
            'captions_neural': 'Neural Language (Captions) - Average',
            'captions_statistical': 'Statistical Text (Captions) - Average',
            'tags_statistical': 'Statistical Text (Tags) - Average'
        }

        # Handle category averages - USE STORED COLUMNS IF THEY EXIST FOR CONSISTENCY
        if ml_model_selection == "avg_vision":
            # Check if pre-calculated column exists
            if 'avg_vision' in self.data_loader.data.columns:
                return self.data_loader.data['avg_vision'], category_labels['vision']
            # Otherwise calculate AND STORE for consistency
            models = [model[0] for model in self.data_loader.model_categories['vision']]
            if models:
                available_models = [m for m in models if m in data.columns]
                if not available_models:
                    raise ValueError("No vision models available in data")
                avg_data = data[available_models].mean(axis=1)
                self.data_loader.data['avg_vision'] = avg_data  # Store in original DataFrame
                return self.data_loader.data['avg_vision'], category_labels['vision']
            else:
                raise ValueError("No vision models available")

        elif ml_model_selection == "avg_captions_neural":
            # Check if pre-calculated column exists
            if 'avg_captions_neural' in self.data_loader.data.columns:
                return self.data_loader.data['avg_captions_neural'], category_labels['captions_neural']
            # Otherwise calculate AND STORE for consistency
            models = [model[0] for model in self.data_loader.model_categories['captions_neural']]
            if models:
                available_models = [m for m in models if m in data.columns]
                if not available_models:
                    raise ValueError("No neural language models available in data")
                avg_data = data[available_models].mean(axis=1)
                self.data_loader.data['avg_captions_neural'] = avg_data  # Store in original DataFrame
                return self.data_loader.data['avg_captions_neural'], category_labels['captions_neural']
            else:
                raise ValueError("No neural language models available")

        elif ml_model_selection == "avg_captions_statistical":
            # Check if pre-calculated column exists
            if 'avg_captions_statistical' in self.data_loader.data.columns:
                return self.data_loader.data['avg_captions_statistical'], category_labels['captions_statistical']
            # Otherwise calculate AND STORE for consistency
            models = [model[0] for model in self.data_loader.model_categories['captions_statistical']]
            if models:
                available_models = [m for m in models if m in data.columns]
                if not available_models:
                    raise ValueError("No statistical caption models available in data")
                avg_data = data[available_models].mean(axis=1)
                self.data_loader.data['avg_captions_statistical'] = avg_data  # Store in original DataFrame
                return self.data_loader.data['avg_captions_statistical'], category_labels['captions_statistical']
            else:
                raise ValueError("No statistical caption models available")

        elif ml_model_selection == "avg_tags_statistical":
            # Check if pre-calculated column exists
            if 'avg_tags_statistical' in self.data_loader.data.columns:
                return self.data_loader.data['avg_tags_statistical'], category_labels['tags_statistical']
            # Otherwise calculate AND STORE for consistency
            models = [model[0] for model in self.data_loader.model_categories['tags_statistical']]
            if models:
                available_models = [m for m in models if m in data.columns]
                if not available_models:
                    raise ValueError("No statistical tag models available in data")
                avg_data = data[available_models].mean(axis=1)
                self.data_loader.data['avg_tags_statistical'] = avg_data  # Store in original DataFrame
                return self.data_loader.data['avg_tags_statistical'], category_labels['tags_statistical']
            else:
                raise ValueError("No statistical tag models available")

        # Handle individual models
        elif isinstance(ml_model_selection, int):
            ml_column = self.data_loader.ml_models[ml_model_selection]
            return data[ml_column], ml_column

        else:
            raise ValueError(f"Invalid model selection: {ml_model_selection}")
    
    @staticmethod
    def normalize_series(series: pd.Series) -> pd.Series:
        """Normalize a pandas series to 0-1 range"""
        min_val = series.min()
        max_val = series.max()
        if max_val == min_val:
            return pd.Series([0.5] * len(series))
        return (series - min_val) / (max_val - min_val)

    def create_3d_plot(self, brain_measure: str, ml_model_selection: Union[str, int], normalize: bool = False) -> Optional[go.Figure]:
        """Create 3D scatter plot"""
        data = self.data_loader.data
        
        try:
            ml_data, ml_name = self.get_model_data(ml_model_selection)
        except ValueError as e:
            print(f"Error getting model data: {e}")
            return None
        
        # Get data (normalized or raw)
        if normalize:
            human_data = self.normalize_series(data['human_judgement'])
            brain_data = self.normalize_series(data[brain_measure])
            ml_plot_data = self.normalize_series(ml_data)
            value_suffix = " (normalized)"
        else:
            human_data = data['human_judgement']
            brain_data = data[brain_measure]
            ml_plot_data = ml_data
            value_suffix = ""
        
        # Create hover text
        hover_text = []
        for idx, row in data.iterrows():
            text = f"Pair #{idx}<br>"
            text += f"Images: {row['image_1']} vs {row['image_2']}<br>"
            text += f"Human: {human_data.iloc[idx]:.3f}<br>"
            text += f"Brain: {brain_data.iloc[idx]:.3f}<br>"
            text += f"ML: {ml_plot_data.iloc[idx]:.3f}"
            hover_text.append(text)
        
        fig = go.Figure(data=go.Scatter3d(
            x=human_data,
            y=brain_data,
            z=ml_plot_data,
            mode='markers',
            marker=dict(
                size=6,
                color=human_data,
                colorscale='Viridis',
                opacity=0.7,
                colorbar=dict(title="Human Rating" + value_suffix)
            ),
            text=hover_text,
            hovertemplate='%{text}<extra></extra>',
            customdata=data[['image_1', 'image_2', 'stim_1', 'stim_2']].values
        ))
        
        # Determine measure type and name from new column naming
        if brain_measure.startswith("roi_"):
            parts = brain_measure.split("_")
            measure_type = parts[1].title()  # cosine or pearson
            roi_type = parts[2].title()  # common, early, late
            if "avg_sim" in brain_measure:
                brain_name = f"{measure_type} {roi_type} (Similarity)"
            elif "avg_roi" in brain_measure:
                brain_name = f"{measure_type} {roi_type} (Pattern)"
            else:
                brain_name = brain_measure
        elif brain_measure.startswith("voxel_") and not brain_measure.startswith("voxel_to_roi_"):
            if "cosine" in brain_measure:
                measure_type = "Cosine"
            else:
                measure_type = "Pearson"
            if "all_avg" in brain_measure:
                brain_name = f"{measure_type} All Voxels (Avg)"
            elif "subj" in brain_measure:
                subj_num = brain_measure.split("subj")[1]
                brain_name = f"{measure_type} Subject {subj_num} Voxels"
            else:
                brain_name = brain_measure
        elif brain_measure.startswith("voxel_to_roi_"):
            parts = brain_measure.replace("voxel_to_roi_", "").split("_")
            measure_type = parts[0].title()
            roi_type = parts[1].title()
            if "avg_sim" in brain_measure:
                brain_name = f"{measure_type} {roi_type} (V→R Sim)"
            elif "avg_roi" in brain_measure:
                brain_name = f"{measure_type} {roi_type} (V→R Pattern)"
            else:
                brain_name = brain_measure
        else:
            brain_name = brain_measure
            measure_type = "Unknown"

        
        x_title = f'Human Rating{value_suffix}'
        y_title = f'Brain Similarity ({measure_type} {brain_name}){value_suffix}'
        z_title = f'ML Model: {ml_name}{value_suffix}'
        
        fig.update_layout(
            title=f'3D Analysis: Human vs {measure_type} {brain_name} Brain vs {ml_name}{"" if not normalize else " (Normalized)"}',
            scene=dict(
                xaxis_title=x_title,
                yaxis_title=y_title,
                zaxis_title=z_title,
                camera=dict(eye=dict(x=1.5, y=1.5, z=1.5))
            ),
            width=800,
            height=600
        )
        
        return fig

    def create_2d_plots(self, brain_measure: str, ml_model_selection: Union[str, int], normalize: bool = False) -> Optional[go.Figure]:
        """Create three 2D scatter plots"""
        data = self.data_loader.data
        
        try:
            ml_data, ml_name = self.get_model_data(ml_model_selection)
        except ValueError as e:
            print(f"Error getting model data: {e}")
            return None
        
        # Get data (normalized or raw)
        if normalize:
            human_data = self.normalize_series(data['human_judgement'])
            brain_data = self.normalize_series(data[brain_measure])
            ml_plot_data = self.normalize_series(ml_data)
            value_suffix = " (norm)"
        else:
            human_data = data['human_judgement']
            brain_data = data[brain_measure]
            ml_plot_data = ml_data
            value_suffix = ""
        
        # Calculate correlations (always on raw data)
        corr_hb = data['human_judgement'].corr(data[brain_measure])

        # Debug: Print what we're working with
        print(f"\n[DEBUG generate_scatter]")
        print(f"  ml_model_selection = {ml_model_selection}")
        print(f"  ml_model_selection type = {type(ml_model_selection)}")
        print(f"  Starts with 'avg_'? {str(ml_model_selection).startswith('avg_')}")

        # Check if this is a category average - use Method 2 (correlate then average)
        # This matches the bar chart methodology
        if str(ml_model_selection).startswith('avg_'):
            # Determine which category
            category_map = {
                'avg_vision': 'vision',
                'avg_captions_neural': 'captions_neural',
                'avg_captions_statistical': 'captions_statistical',
                'avg_tags_statistical': 'tags_statistical'
            }

            if ml_model_selection in category_map:
                category_key = category_map[ml_model_selection]

                # Method 2: Correlate each model individually, then average correlations
                corr_hm = self.compute_category_correlation_method2(category_key, data['human_judgement'])
                corr_bm = self.compute_category_correlation_method2(category_key, data[brain_measure])

                # Update model name to indicate Method 2 is used
                ml_name = ml_name + " (Method 2: Avg of Correlations)"

                print(f"[METHOD 2] Using correlate-then-average for {ml_model_selection}")
                print(f"  Human vs Category: r = {corr_hm:.4f}")
                print(f"  Brain vs Category: r = {corr_bm:.4f}")
            else:
                # Fallback to Method 1
                corr_hm = data['human_judgement'].corr(ml_data)
                corr_bm = data[brain_measure].corr(ml_data)
        else:
            # Individual model - use regular Method 1
            corr_hm = data['human_judgement'].corr(ml_data)
            corr_bm = data[brain_measure].corr(ml_data)
        
        # Determine measure type and name from new column naming
        if brain_measure.startswith("roi_"):
            parts = brain_measure.split("_")
            measure_type = parts[1].title()  # cosine or pearson
            roi_type = parts[2].title()  # common, early, late
            if "avg_sim" in brain_measure:
                brain_name = f"{measure_type} {roi_type} (Similarity)"
            elif "avg_roi" in brain_measure:
                brain_name = f"{measure_type} {roi_type} (Pattern)"
            else:
                brain_name = brain_measure
        elif brain_measure.startswith("voxel_") and not brain_measure.startswith("voxel_to_roi_"):
            if "cosine" in brain_measure:
                measure_type = "Cosine"
            else:
                measure_type = "Pearson"
            if "all_avg" in brain_measure:
                brain_name = f"{measure_type} All Voxels (Avg)"
            elif "subj" in brain_measure:
                subj_num = brain_measure.split("subj")[1]
                brain_name = f"{measure_type} Subject {subj_num} Voxels"
            else:
                brain_name = brain_measure
        elif brain_measure.startswith("voxel_to_roi_"):
            parts = brain_measure.replace("voxel_to_roi_", "").split("_")
            measure_type = parts[0].title()
            roi_type = parts[1].title()
            if "avg_sim" in brain_measure:
                brain_name = f"{measure_type} {roi_type} (V→R Sim)"
            elif "avg_roi" in brain_measure:
                brain_name = f"{measure_type} {roi_type} (V→R Pattern)"
            else:
                brain_name = brain_measure
        else:
            brain_name = brain_measure
            measure_type = "Unknown"

        
        # Create subplot
        fig = make_subplots(
            rows=1, cols=3,
            subplot_titles=[
                f'Human vs Brain (r={corr_hb:.3f})', 
                f'Human vs ML (r={corr_hm:.3f})', 
                f'Brain vs ML (r={corr_bm:.3f})'
            ],
            horizontal_spacing=0.1
        )
        
        # Custom data for hover
        customdata = [[idx, row['image_1'], row['image_2']] for idx, row in data.iterrows()]
        
        # Add scatter plots with proper labels
        plot_configs = [
            {
                'x': human_data, 
                'y': brain_data, 
                'color': 'blue',
                'x_label': f'Human{value_suffix}',
                'y_label': f'Brain ({measure_type}){value_suffix}'
            },
            {
                'x': human_data, 
                'y': ml_plot_data, 
                'color': 'red',
                'x_label': f'Human{value_suffix}',
                'y_label': f'ML Model{value_suffix}'
            },
            {
                'x': brain_data, 
                'y': ml_plot_data, 
                'color': 'green',
                'x_label': f'Brain ({measure_type}){value_suffix}',
                'y_label': f'ML Model{value_suffix}'
            }
        ]
        
        for i, config in enumerate(plot_configs):
            fig.add_trace(
                go.Scatter(
                    x=config['x'], 
                    y=config['y'],
                    mode='markers',
                    marker=dict(color=config['color'], opacity=0.6, size=3),
                    hovertemplate=f'Pair #%{{customdata[0]}}<br>{config["x_label"]}: %{{x:.3f}}<br>{config["y_label"]}: %{{y:.3f}}<br>%{{customdata[1]}} vs %{{customdata[2]}}<extra></extra>',
                    customdata=customdata,
                    showlegend=False
                ),
                row=1, col=i+1
            ) 

        fig.update_layout(
            title=f'2D Comparisons: {measure_type} {brain_name} Brain vs {ml_name}{"" if not normalize else " (Normalized)"}',
            width=1300,
            height=500,
            margin=dict(l=60, r=60, t=80, b=80)
        )
        
        # Add axis labels to each subplot
        fig.update_xaxes(title_text="Human Similarity", row=1, col=1)
        fig.update_yaxes(title_text=f"Brain Similarity ({measure_type})", row=1, col=1)
        
        fig.update_xaxes(title_text="Human Similarity", row=1, col=2)
        fig.update_yaxes(title_text="ML Model Similarity", row=1, col=2)
        
        fig.update_xaxes(title_text=f"Brain Similarity ({measure_type})", row=1, col=3)
        fig.update_yaxes(title_text="ML Model Similarity", row=1, col=3)
        
        return fig