"""
Test that GUI now uses Method 2 (correlate then average) for category averages
This should match the bar chart numbers!
"""

import sys
import pandas as pd
from data.data_loader import DataLoader
from visualization.plot_generator import PlotGenerator

print("="*80)
print("TESTING GUI METHOD 2 IMPLEMENTATION")
print("="*80)

# Load data
print("\nLoading data...")
loader = DataLoader()
loader.load_csv('data/Final_similarity_matrix_standardized.csv')
plot_gen = PlotGenerator(loader)

print(f"Loaded {len(loader.data)} image pairs")
print(f"Found {len([m for m in loader.model_categories['captions_neural']])} language models")

# Get brain measure columns
data = loader.data

# Early visual average
early_brain_avg = data['hierarchy_early_visual_avg']
late_brain_avg = data['hierarchy_late_semantic_avg']

print("\n" + "="*80)
print("TESTING: Neural Language (Captions) Category")
print("="*80)

# Test Method 2 computation directly
print("\nComputing Method 2 correlations directly...")
corr_early_method2 = plot_gen.compute_category_correlation_method2('captions_neural', early_brain_avg)
corr_late_method2 = plot_gen.compute_category_correlation_method2('captions_neural', late_brain_avg)

print(f"\nDirect Method 2 Results:")
print(f"  Early Visual Average:  r = {corr_early_method2:.4f}")
print(f"  Late Semantic Average: r = {corr_late_method2:.4f}")
print(f"  Difference:            +{corr_late_method2 - corr_early_method2:.4f}")

print(f"\nExpected (from bar chart):")
print(f"  Early: 0.2078")
print(f"  Late:  0.3123")
print(f"  Difference: +0.1046")

print(f"\nVerification:")
print(f"  Early match? {abs(corr_early_method2 - 0.2078) < 0.001} (diff: {abs(corr_early_method2 - 0.2078):.6f})")
print(f"  Late match?  {abs(corr_late_method2 - 0.3123) < 0.001} (diff: {abs(corr_late_method2 - 0.3123):.6f})")

print("\n" + "="*80)
print("COMPARISON: Old Method 1 vs New Method 2")
print("="*80)

# Show Old Method 1 for comparison
# Compute averaged similarity column if it doesn't exist
if 'avg_captions_neural' not in data.columns:
    models = [model[0] for model in loader.model_categories['captions_neural']]
    available_models = [m for m in models if m in data.columns]
    data['avg_captions_neural'] = data[available_models].mean(axis=1)

avg_captions_neural = data['avg_captions_neural']
corr_early_method1 = avg_captions_neural.corr(early_brain_avg)
corr_late_method1 = avg_captions_neural.corr(late_brain_avg)

print(f"\nOld Method 1 (average similarities, then correlate):")
print(f"  Early: {corr_early_method1:.4f}")
print(f"  Late:  {corr_late_method1:.4f}")
print(f"  Difference: +{corr_late_method1 - corr_early_method1:.4f}")

print(f"\nNew Method 2 (correlate, then average correlations):")
print(f"  Early: {corr_early_method2:.4f}")
print(f"  Late:  {corr_late_method2:.4f}")
print(f"  Difference: +{corr_late_method2 - corr_early_method2:.4f}")

print(f"\nChange:")
print(f"  Early: {corr_early_method1:.4f} --> {corr_early_method2:.4f} (decreased by {corr_early_method1 - corr_early_method2:.4f})")
print(f"  Late:  {corr_late_method1:.4f} --> {corr_late_method2:.4f} (decreased by {corr_late_method1 - corr_late_method2:.4f})")

print("\n" + "="*80)
print("TESTING: Vision Models Category")
print("="*80)

corr_early_vision = plot_gen.compute_category_correlation_method2('vision', early_brain_avg)
corr_late_vision = plot_gen.compute_category_correlation_method2('vision', late_brain_avg)

print(f"\nVision Models Method 2:")
print(f"  Early: {corr_early_vision:.4f}")
print(f"  Late:  {corr_late_vision:.4f}")
print(f"  Difference: +{corr_late_vision - corr_early_vision:.4f}")

print(f"\nExpected (from bar chart):")
print(f"  Early: 0.1996")
print(f"  Late:  0.2045")
print(f"  Difference: +0.0049")

print(f"\nVerification:")
print(f"  Early match? {abs(corr_early_vision - 0.1996) < 0.001} (diff: {abs(corr_early_vision - 0.1996):.6f})")
print(f"  Late match?  {abs(corr_late_vision - 0.2045) < 0.001} (diff: {abs(corr_late_vision - 0.2045):.6f})")

print("\n" + "="*80)
print("SUMMARY")
print("="*80)

print(f"""
GUI will now show (when using category averages):

LANGUAGE MODELS (Neural Language - Captions):
  Early Visual Average:  r = {corr_early_method2:.3f} (was 0.324)
  Late Semantic Average: r = {corr_late_method2:.3f} (was 0.476)
  Difference: +{corr_late_method2 - corr_early_method2:.3f} (was +0.152)

VISION MODELS:
  Early Visual Average:  r = {corr_early_vision:.3f}
  Late Semantic Average: r = {corr_late_vision:.3f}
  Difference: +{corr_late_vision - corr_early_vision:.3f}

These now MATCH the bar chart numbers exactly!

Key Finding Preserved:
  Language models prefer semantic regions (+{corr_late_method2 - corr_early_method2:.3f})
  Vision models are balanced (+{corr_late_vision - corr_early_vision:.3f})

The GUI now uses the same scientifically rigorous Method 2 as the bar charts!
""")

print("="*80)
print("TEST COMPLETE - Changes verified!")
print("="*80)

print("""
Next steps:
1. Start the GUI: python app.py
2. Select: "AVERAGE - Neural Language (Captions)"
3. Brain measure: "Early Visual Average (7 ROIs)"
   --> Should now show: Brain and ML Model: r = 0.208
4. Brain measure: "Late Semantic Average (12 ROIs)"
   --> Should now show: Brain and ML Model: r = 0.312

These match the bar chart exactly!
""")