foxhunt/ml/tests/dqn_hyperopt_multiobjective_test.rs

//! DQN Hyperopt Multi-Objective Function Tests
//!
//! This module tests the multi-objective optimization function that combines:
//! - Reward component (80% weight): Primary metric - episode reward
//! - Diversity penalty (10% weight): Action distribution entropy
//! - Stability penalty (5% weight): Q-value health and gradient norms
//! - Completion penalty (5% weight): Epochs completed vs target
//!
//! ## Test Philosophy
//!
//! These tests validate the objective function BEFORE implementation (TDD).
//! They will fail initially and pass once the implementation is complete.
//!
//! ## Expected Behavior
//!
//! 1. **Hard Constraints** (trial rejection → objective = +1e6):
//!    - Q-value floor: > -10.0
//!    - Training loss ceiling: < 1000.0
//!    - Minimum epochs: ≥ 20
//!
//! 2. **Soft Penalties** (weighted composite):
//!    - Reward: -avg_episode_reward (negated for minimization)
//!    - Diversity: Entropy-based action distribution penalty
//!    - Stability: Q-value range + gradient norm penalties
//!    - Completion: Penalty for early stopping
//!
//! ## Note on DQNMetrics Structure
//!
//! This test file uses the PROPOSED DQNMetrics structure with added fields:
//! - `action_distribution: [f64; 3]` - [BUY%, SELL%, HOLD%]
//! - `avg_gradient_norm: f64` - Average gradient norm during training
//!
//! These fields need to be added to `ml/src/hyperopt/adapters/dqn.rs` before
//! the tests can compile. See `DQN_HYPEROPT_OBJECTIVE_ANALYSIS.md` Section 6.

// NOTE: This will not compile until DQNMetrics is extended with:
// - action_distribution: [f64; 3]
// - avg_gradient_norm: f64
//
// Temporary struct definition for test-driven development
#[derive(Debug, Clone)]
struct DQNMetrics {
    pub train_loss: f64,
    pub val_loss: f64,
    pub avg_q_value: f64,
    pub final_epsilon: f64,
    pub epochs_completed: usize,
    pub avg_episode_reward: f64,
    pub action_distribution: [f64; 3],  // ✅ NEW: [BUY%, SELL%, HOLD%]
    pub avg_gradient_norm: f64,         // ✅ NEW: Average gradient norm
}

// ============================================================================
// Test Helpers
// ============================================================================

/// Create test DQNMetrics with default "healthy" values
fn create_healthy_metrics() -> DQNMetrics {
    DQNMetrics {
        train_loss: 0.1,
        val_loss: 0.15,
        avg_q_value: 5.0,           // Healthy Q-value (0.5 to 10.0)
        final_epsilon: 0.01,
        epochs_completed: 50,        // Full training
        avg_episode_reward: 0.001,   // Positive reward
        action_distribution: [0.33, 0.33, 0.34],  // Balanced actions
        avg_gradient_norm: 2.0,      // Healthy gradient norm (< 10.0)
    }
}

/// Create test DQNMetrics with Q-value collapse
fn create_q_collapse_metrics() -> DQNMetrics {
    DQNMetrics {
        train_loss: 0.1,
        val_loss: 0.15,
        avg_q_value: -681.92,        // CATASTROPHIC collapse
        final_epsilon: 0.01,
        epochs_completed: 10,         // Early stop
        avg_episode_reward: 0.001,
        action_distribution: [0.994, 0.003, 0.003],  // Degenerate (99.4% BUY)
        avg_gradient_norm: 0.01,
    }
}

/// Create test DQNMetrics with high loss (numerical explosion)
fn create_high_loss_metrics() -> DQNMetrics {
    DQNMetrics {
        train_loss: 5000.0,          // EXPLOSION
        val_loss: 5500.0,
        avg_q_value: 5.0,
        final_epsilon: 0.01,
        epochs_completed: 5,          // Early stop due to explosion
        avg_episode_reward: 0.0001,
        action_distribution: [0.50, 0.30, 0.20],
        avg_gradient_norm: 150.0,     // Gradient explosion
    }
}

/// Create test DQNMetrics with early stopping
fn create_early_stop_metrics() -> DQNMetrics {
    DQNMetrics {
        train_loss: 0.1,
        val_loss: 0.15,
        avg_q_value: 5.0,
        final_epsilon: 0.50,          // High epsilon (not enough decay)
        epochs_completed: 15,         // Stopped at 15/50 epochs
        avg_episode_reward: 0.0005,
        action_distribution: [0.40, 0.35, 0.25],
        avg_gradient_norm: 3.0,
    }
}

/// Create test DQNMetrics with extreme HOLD bias
fn create_hold_bias_metrics() -> DQNMetrics {
    DQNMetrics {
        train_loss: 0.05,
        val_loss: 0.06,
        avg_q_value: 3.0,
        final_epsilon: 0.01,
        epochs_completed: 50,
        avg_episode_reward: 0.0003,   // Low reward (agent not acting)
        action_distribution: [0.05, 0.05, 0.90],  // 90% HOLD
        avg_gradient_norm: 1.5,
    }
}

// ============================================================================
// Component Tests (6 tests)
// ============================================================================

#[test]
fn test_reward_component_normalization() {
    // Test that reward component is correctly negated and scaled
    let metrics = create_healthy_metrics();

    // Reward component should be -avg_episode_reward
    // For avg_episode_reward = 0.001, reward_component = -0.001
    let reward_component = calculate_reward_component(&metrics);

    assert_eq!(
        reward_component,
        -0.001,
        "Reward component should be negated avg_episode_reward"
    );

    // Test with higher reward
    let high_reward_metrics = DQNMetrics {
        avg_episode_reward: 0.005,
        ..create_healthy_metrics()
    };
    let high_reward_component = calculate_reward_component(&high_reward_metrics);

    assert_eq!(
        high_reward_component,
        -0.005,
        "Higher reward should result in more negative component (better for minimization)"
    );
}

#[test]
fn test_diversity_penalty_balanced_actions() {
    // Test that balanced action distribution (33/33/33) results in zero/low penalty
    let metrics = create_healthy_metrics();

    let penalty = calculate_diversity_penalty(&metrics);

    // Balanced distribution should have entropy close to max_entropy = ln(3) ≈ 1.099
    // Diversity score = entropy / max_entropy ≈ 1.0
    // Penalty = 1.0 - diversity_score ≈ 0.0
    assert!(
        penalty < 0.05,
        "Balanced action distribution should have low penalty: {}",
        penalty
    );
}

#[test]
fn test_diversity_penalty_extreme_hold() {
    // Test that extreme HOLD bias (5/5/90) results in high penalty
    let metrics = create_hold_bias_metrics();

    let penalty = calculate_diversity_penalty(&metrics);

    // Degenerate distribution (90% HOLD) should have low entropy
    // Diversity score should be low
    // Penalty should be high (close to 1.0)
    assert!(
        penalty > 0.5,
        "Extreme HOLD bias should have high penalty: {}",
        penalty
    );
}

#[test]
fn test_stability_penalty_good_qvalues() {
    // Test that Q-values in healthy range (0.5 to 10.0) result in zero penalty
    let metrics = create_healthy_metrics();

    let penalty = calculate_stability_penalty(&metrics);

    // Q-value = 5.0 is in healthy range [0.5, 10.0]
    // Gradient norm = 2.0 is below threshold (10.0)
    // Penalty should be zero or very low
    assert!(
        penalty < 0.1,
        "Healthy Q-values and gradients should have low penalty: {}",
        penalty
    );
}

#[test]
fn test_stability_penalty_gradient_explosion() {
    // Test that gradient explosion (grad_norm > 10) results in high penalty
    let metrics = create_high_loss_metrics();

    let penalty = calculate_stability_penalty(&metrics);

    // Gradient norm = 150.0 >> 10.0 threshold
    // Should result in significant penalty
    assert!(
        penalty > 5.0,
        "Gradient explosion should have high penalty: {}",
        penalty
    );
}

#[test]
fn test_completion_penalty_early_stop() {
    // Test that early stopping (30/100 epochs) results in penalty
    let metrics = create_early_stop_metrics();

    let penalty = calculate_completion_penalty(&metrics, 50);

    // Completed 15/50 epochs = 30% completion
    // Should have penalty proportional to missing epochs
    assert!(
        penalty > 0.5,
        "Early stopping should have significant penalty: {}",
        penalty
    );
}

// ============================================================================
// Integration Tests (3 tests)
// ============================================================================

#[test]
fn test_multiobjective_optimal_trial() {
    // Test that a "perfect" trial gets a good (low) objective score
    let metrics = create_healthy_metrics();

    let objective = calculate_multiobjective(&metrics, 50);

    // Expected calculation:
    // reward_component = -0.001
    // diversity_penalty = ~0.0 (balanced actions)
    // stability_penalty = ~0.0 (healthy Q-values)
    // completion_penalty = ~0.0 (50/50 epochs)
    //
    // objective = 0.8 * (-0.001) + 0.1 * 0.0 + 0.05 * 0.0 + 0.05 * 0.0
    //           = -0.0008

    assert!(
        objective < -0.0005,
        "Optimal trial should have low (negative) objective: {}",
        objective
    );

    // Verify all constraints pass
    assert!(
        !violates_hard_constraints(&metrics),
        "Optimal trial should not violate hard constraints"
    );
}

#[test]
fn test_multiobjective_broken_trial() {
    // Test that a broken trial (99% HOLD) gets a bad (high) objective score
    let metrics = create_hold_bias_metrics();

    let objective = calculate_multiobjective(&metrics, 50);

    // Expected calculation:
    // reward_component = -0.0003 (low reward due to inaction)
    // diversity_penalty = ~0.6 (90% HOLD bias)
    // stability_penalty = ~0.0 (Q-values OK)
    // completion_penalty = ~0.0 (50/50 epochs)
    //
    // objective = 0.8 * (-0.0003) + 0.1 * 0.6 + 0.05 * 0.0 + 0.05 * 0.0
    //           = -0.00024 + 0.06
    //           = +0.05976

    assert!(
        objective > 0.03,
        "Broken trial (HOLD bias) should have high (positive) objective: {}",
        objective
    );
}

#[test]
fn test_weight_sensitivity() {
    // Test that weight changes affect trial ranking correctly

    let good_metrics = create_healthy_metrics();
    let hold_bias_metrics = create_hold_bias_metrics();

    // Calculate objectives with standard weights (reward=80%, diversity=10%, stability=5%, completion=5%)
    let good_objective = calculate_multiobjective(&good_metrics, 50);
    let hold_objective = calculate_multiobjective(&hold_bias_metrics, 50);

    // Good trial should have LOWER objective (minimization)
    assert!(
        good_objective < hold_objective,
        "Good trial (balanced actions) should rank better than HOLD-biased trial: {} vs {}",
        good_objective,
        hold_objective
    );

    // Verify ranking is correct
    let ranking_correct = good_objective < hold_objective;
    assert!(
        ranking_correct,
        "Weight sensitivity failed: good={}, hold={}",
        good_objective,
        hold_objective
    );
}

// ============================================================================
// Hard Constraint Tests (3 tests)
// ============================================================================

#[test]
fn test_hard_constraint_q_value_floor() {
    // Test that Q-value < -10.0 triggers hard constraint violation
    let metrics = create_q_collapse_metrics();

    let objective = calculate_multiobjective(&metrics, 50);

    // Hard constraint violation should return penalty = +1e6
    assert_eq!(
        objective,
        1e6,
        "Q-value floor violation should return penalty objective"
    );
}

#[test]
fn test_hard_constraint_training_loss_ceiling() {
    // Test that training loss > 1000.0 triggers hard constraint violation
    let metrics = create_high_loss_metrics();

    let objective = calculate_multiobjective(&metrics, 50);

    // Hard constraint violation should return penalty = +1e6
    assert_eq!(
        objective,
        1e6,
        "Training loss ceiling violation should return penalty objective"
    );
}

#[test]
fn test_hard_constraint_minimum_epochs() {
    // Test that epochs_completed < 20 triggers hard constraint violation
    let metrics = DQNMetrics {
        epochs_completed: 15,  // Below minimum of 20
        ..create_healthy_metrics()
    };

    let objective = calculate_multiobjective(&metrics, 50);

    // Hard constraint violation should return penalty = +1e6
    assert_eq!(
        objective,
        1e6,
        "Minimum epochs violation should return penalty objective"
    );
}

// ============================================================================
// Edge Case Tests (3 additional tests)
// ============================================================================

#[test]
fn test_edge_case_zero_entropy() {
    // Test that 100% single action (zero entropy) results in maximum penalty
    let metrics = DQNMetrics {
        train_loss: 0.1,
        val_loss: 0.15,
        avg_q_value: 5.0,
        final_epsilon: 0.01,
        epochs_completed: 50,
        avg_episode_reward: 0.001,
        action_distribution: [1.0, 0.0, 0.0],  // 100% BUY (degenerate)
        avg_gradient_norm: 2.0,
    };

    let penalty = calculate_diversity_penalty(&metrics);

    // Zero entropy → diversity_score = 0 → penalty = 1.0 (maximum)
    assert!(
        penalty > 0.95,
        "Zero entropy should have maximum penalty: {}",
        penalty
    );
}

#[test]
fn test_edge_case_q_value_boundaries() {
    // Test Q-value penalty at exact boundaries (0.5 and 10.0)

    // Test 1: Q-value = 0.5 (lower boundary of healthy range)
    let metrics_lower = DQNMetrics {
        avg_q_value: 0.5,
        avg_gradient_norm: 2.0,
        ..create_healthy_metrics()
    };
    let penalty_lower = calculate_stability_penalty(&metrics_lower);
    assert!(
        penalty_lower < 0.01,
        "Q-value at lower boundary (0.5) should have minimal penalty: {}",
        penalty_lower
    );

    // Test 2: Q-value = 10.0 (upper boundary of healthy range)
    let metrics_upper = DQNMetrics {
        avg_q_value: 10.0,
        avg_gradient_norm: 2.0,
        ..create_healthy_metrics()
    };
    let penalty_upper = calculate_stability_penalty(&metrics_upper);
    assert!(
        penalty_upper < 0.01,
        "Q-value at upper boundary (10.0) should have minimal penalty: {}",
        penalty_upper
    );

    // Test 3: Q-value = 0.49 (just below lower boundary)
    let metrics_below = DQNMetrics {
        avg_q_value: 0.49,
        avg_gradient_norm: 2.0,
        ..create_healthy_metrics()
    };
    let penalty_below = calculate_stability_penalty(&metrics_below);
    assert!(
        penalty_below > 0.0,
        "Q-value below lower boundary (0.49) should have penalty: {}",
        penalty_below
    );

    // Test 4: Q-value = 10.01 (just above upper boundary)
    let metrics_above = DQNMetrics {
        avg_q_value: 10.01,
        avg_gradient_norm: 2.0,
        ..create_healthy_metrics()
    };
    let penalty_above = calculate_stability_penalty(&metrics_above);
    assert!(
        penalty_above > 0.0,
        "Q-value above upper boundary (10.01) should have penalty: {}",
        penalty_above
    );
}

#[test]
fn test_edge_case_gradient_norm_threshold() {
    // Test gradient norm penalty at exact threshold (10.0)

    // Test 1: grad_norm = 10.0 (at threshold)
    let metrics_at = DQNMetrics {
        avg_q_value: 5.0,
        avg_gradient_norm: 10.0,
        ..create_healthy_metrics()
    };
    let penalty_at = calculate_stability_penalty(&metrics_at);
    assert!(
        penalty_at < 0.1,
        "Gradient norm at threshold (10.0) should have minimal penalty: {}",
        penalty_at
    );

    // Test 2: grad_norm = 9.99 (just below threshold)
    let metrics_below = DQNMetrics {
        avg_q_value: 5.0,
        avg_gradient_norm: 9.99,
        ..create_healthy_metrics()
    };
    let penalty_below = calculate_stability_penalty(&metrics_below);
    assert!(
        penalty_below < 0.1,
        "Gradient norm below threshold (9.99) should have minimal penalty: {}",
        penalty_below
    );

    // Test 3: grad_norm = 10.01 (just above threshold)
    let metrics_above = DQNMetrics {
        avg_q_value: 5.0,
        avg_gradient_norm: 10.01,
        ..create_healthy_metrics()
    };
    let penalty_above = calculate_stability_penalty(&metrics_above);
    assert!(
        penalty_above > 0.0,
        "Gradient norm above threshold (10.01) should have penalty: {}",
        penalty_above
    );
}

// ============================================================================
// Component Helper Functions (NOT IMPLEMENTED YET)
// ============================================================================

/// Calculate reward component (negated for minimization)
///
/// **NOT IMPLEMENTED**: This function is a placeholder for the actual implementation.
/// It will be implemented in `ml/src/hyperopt/adapters/dqn.rs` as part of the multi-objective function.
fn calculate_reward_component(_metrics: &DQNMetrics) -> f64 {
    panic!("calculate_reward_component() not implemented yet - implement in ml/src/hyperopt/adapters/dqn.rs");
}

/// Calculate diversity penalty using entropy
///
/// Entropy = -Σ(p_i * ln(p_i)) for i in [BUY, SELL, HOLD]
/// Max entropy = ln(3) ≈ 1.099 (uniform distribution)
/// Diversity score = entropy / max_entropy
/// Penalty = 1.0 - diversity_score
///
/// **NOT IMPLEMENTED**: This function is a placeholder for the actual implementation.
fn calculate_diversity_penalty(_metrics: &DQNMetrics) -> f64 {
    panic!("calculate_diversity_penalty() not implemented yet - implement in ml/src/hyperopt/adapters/dqn.rs");
}

/// Calculate stability penalty from Q-value health and gradient norms
///
/// Q-value penalty:
/// - If Q < 0.5: penalty = |0.5 - Q|
/// - If Q > 10.0: penalty = (Q - 10.0) * 0.1
/// - Else: penalty = 0.0
///
/// Gradient penalty:
/// - If grad_norm > 10.0: penalty = (grad_norm - 10.0) * 0.5
/// - Else: penalty = 0.0
///
/// **NOT IMPLEMENTED**: This function is a placeholder for the actual implementation.
fn calculate_stability_penalty(_metrics: &DQNMetrics) -> f64 {
    panic!("calculate_stability_penalty() not implemented yet - implement in ml/src/hyperopt/adapters/dqn.rs");
}

/// Calculate completion penalty from epochs completed vs target
///
/// Penalty = (target_epochs - completed_epochs) / target_epochs
///
/// **NOT IMPLEMENTED**: This function is a placeholder for the actual implementation.
fn calculate_completion_penalty(_metrics: &DQNMetrics, _target_epochs: usize) -> f64 {
    panic!("calculate_completion_penalty() not implemented yet - implement in ml/src/hyperopt/adapters/dqn.rs");
}

/// Check if metrics violate any hard constraints
///
/// Hard constraints:
/// 1. Q-value floor: avg_q_value > -10.0
/// 2. Training loss ceiling: train_loss < 1000.0
/// 3. Minimum epochs: epochs_completed >= 20
///
/// **NOT IMPLEMENTED**: This function is a placeholder for the actual implementation.
fn violates_hard_constraints(_metrics: &DQNMetrics) -> bool {
    panic!("violates_hard_constraints() not implemented yet - implement in ml/src/hyperopt/adapters/dqn.rs");
}

/// Calculate multi-objective function
///
/// If hard constraints violated: return +1e6
/// Else: return weighted composite of components
///
/// **NOT IMPLEMENTED**: This function is a placeholder for the actual implementation.
fn calculate_multiobjective(_metrics: &DQNMetrics, _target_epochs: usize) -> f64 {
    panic!("calculate_multiobjective() not implemented yet - implement in ml/src/hyperopt/adapters/dqn.rs");
}