"""EEGFaceSem - Minimal Usage Examples"""

import EEGFaceSem
import numpy as np
from sklearn.discriminant_analysis import LinearDiscriminantAnalysis
from sklearn.metrics import accuracy_score

# =============================================================================
# Example 1: Quick test with single subject (downloads ~300MB)
# =============================================================================

X, Y, ids = EEGFaceSem.load_data(task='female', subjects=[1])

X_flat = X.reshape(X.shape[0], -1)
(X_train, Y_train), (X_test, Y_test) = EEGFaceSem.split_random(X_flat, Y)

model = LinearDiscriminantAnalysis()
model.fit(X_train, Y_train)

acc = accuracy_score(Y_test, model.predict(X_test))
print(f"Accuracy: {acc:.2f}")

# =============================================================================
# Example 2: Full dataset with all subjects (downloads ~9GB)
# =============================================================================

# X, Y, ids = EEGFaceSem.load_data(task='female')  # All 30 subjects
# 
# # Leave-one-subject-out cross-validation
# for test_subj in range(30):
#     (X_train, Y_train), (X_test, Y_test) = EEGFaceSem.split_by_subject(X, Y, ids, test_subj)
#     # ... train and evaluate

# =============================================================================
# Example 3: Run full benchmark
# =============================================================================

# EEGFaceSem.benchmark(model='LDA', task_id=0)

# =============================================================================
# Example 4: Generate face images from latent vectors
# =============================================================================

# EEGFaceSem.download_models()  # Download PGAN model (~277MB)
# images = EEGFaceSem.generate(np.random.randn(1, 512))
# images[0].save("generated_face.png")