foxhunt/tests/unit/ml/model_tests.rs

//! ML Model Integration and Accuracy Validation Tests
//! 
//! Comprehensive test suite for all ML models in the Foxhunt HFT system.
//! Tests model loading, inference accuracy, performance, and integration.

use std::time::{Duration, Instant};
use tokio::time::timeout;
// Note: ML models not yet available - commenting out for compilation
// use ml::{
//     MLModel, ModelRegistry, TLOBTransformer, MAMBAModel, 
//     LiquidNeuralNetwork, TemporalFusionTransformer, DQNAgent, PPOAgent
// };
// use risk::{RiskEngine, PositionTracker};

// Mock ML types for testing
pub struct MockMLModel;
pub struct MockModelRegistry;
pub struct MockTLOBTransformer;
pub struct MockMAMBAModel;
pub struct MockLiquidNeuralNetwork;
pub struct MockTemporalFusionTransformer;
pub struct MockDQNAgent;
pub struct MockPPOAgent;
pub struct MockRiskEngine;
pub struct MockPositionTracker;
// Simple test configuration for this file
#[derive(Debug, Clone)]
struct UnifiedTestConfig {
    initial_capital: common::prelude::Decimal,
    enable_logging: bool,
}

fn create_test_config() -> UnifiedTestConfig {
    UnifiedTestConfig {
        initial_capital: common::prelude::Decimal::from(100000),
        enable_logging: false,
    }
}

/// Configuration for ML model testing
#[derive(Debug, Clone)]
pub struct MLTestConfig {
    pub inference_timeout: Duration,
    pub accuracy_threshold: f64,
    pub max_inference_latency: Duration,
    pub model_warmup_iterations: usize,
    pub test_batch_size: usize,
    pub performance_iterations: usize,
}

impl Default for MLTestConfig {
    fn default() -> Self {
        Self {
            inference_timeout: Duration::from_millis(100),
            accuracy_threshold: 0.75, // 75% accuracy minimum
            max_inference_latency: Duration::from_micros(50),
            model_warmup_iterations: 10,
            test_batch_size: 100,
            performance_iterations: 1000,
        }
    }
}

/// ML model accuracy measurement
#[derive(Debug, Clone)]
pub struct AccuracyMeasurement {
    pub model_name: String,
    pub correct_predictions: usize,
    pub total_predictions: usize,
    pub accuracy_percentage: f64,
    pub precision: f64,
    pub recall: f64,
    pub f1_score: f64,
}

impl AccuracyMeasurement {
    pub fn new(model_name: String) -> Self {
        Self {
            model_name,
            correct_predictions: 0,
            total_predictions: 0,
            accuracy_percentage: 0.0,
            precision: 0.0,
            recall: 0.0,
            f1_score: 0.0,
        }
    }

    pub fn calculate_metrics(&mut self, true_positives: usize, false_positives: usize, false_negatives: usize) {
        self.accuracy_percentage = if self.total_predictions > 0 {
            (self.correct_predictions as f64 / self.total_predictions as f64) * 100.0
        } else {
            0.0
        };

        self.precision = if true_positives + false_positives > 0 {
            true_positives as f64 / (true_positives + false_positives) as f64
        } else {
            0.0
        };

        self.recall = if true_positives + false_negatives > 0 {
            true_positives as f64 / (true_positives + false_negatives) as f64
        } else {
            0.0
        };

        self.f1_score = if self.precision + self.recall > 0.0 {
            2.0 * (self.precision * self.recall) / (self.precision + self.recall)
        } else {
            0.0
        };
    }
}

/// ML inference performance measurement
#[derive(Debug, Clone)]
pub struct InferencePerformance {
    pub model_name: String,
    pub avg_inference_time: Duration,
    pub p50_latency: Duration,
    pub p95_latency: Duration,
    pub p99_latency: Duration,
    pub throughput_per_second: f64,
    pub memory_usage_mb: f64,
}

/// ML model test suite
pub struct MLModelTestSuite {
    config: MLTestConfig,
    model_registry: ModelRegistry,
    test_data: Vec<MarketSnapshot>,
    expected_signals: Vec<TradingSignal>,
}

impl MLModelTestSuite {
    pub async fn new(config: MLTestConfig) -> Result<Self, Box<dyn std::error::Error + Send + Sync>> {
        let model_registry = ModelRegistry::new().await?;
        let (test_data, expected_signals) = Self::generate_test_dataset(config.test_batch_size).await?;

        Ok(Self {
            config,
            model_registry,
            test_data,
            expected_signals,
        })
    }

    async fn generate_test_dataset(
        batch_size: usize
    ) -> Result<(Vec<MarketSnapshot>, Vec<TradingSignal>), Box<dyn std::error::Error + Send + Sync>> {
        let mut test_data = Vec::with_capacity(batch_size);
        let mut expected_signals = Vec::with_capacity(batch_size);

        for i in 0..batch_size {
            let timestamp = std::time::SystemTime::now();
            let price = Price::new(100 + (i as f64 * 0.01))?;
            let volume = Quantity::new(1000 + i as i64)?;

            let snapshot = MarketSnapshot {
                symbol: Symbol::new("EURUSD")?,
                timestamp,
                bid: price,
                ask: price + Price::new(0.0001)?,
                volume,
                spread: Price::new(0.0001)?,
            };

            // Generate expected signal based on simple pattern
            let signal_strength = if i % 4 == 0 { 0.8 } else { 0.3 };
            let direction = if i % 2 == 0 { 
                SignalDirection::Buy 
            } else { 
                SignalDirection::Sell 
            };

            let signal = TradingSignal {
                symbol: snapshot.symbol.clone(),
                direction,
                strength: signal_strength,
                timestamp,
                confidence: 0.75,
                source: "test_data".to_string(),
            };

            test_data.push(snapshot);
            expected_signals.push(signal);
        }

        Ok((test_data, expected_signals))
    }

    pub async fn test_model_accuracy<T: MLModel + Send + Sync>(
        &self,
        model: &T,
        model_name: &str
    ) -> Result<AccuracyMeasurement, Box<dyn std::error::Error + Send + Sync>> {
        let mut measurement = AccuracyMeasurement::new(model_name.to_string());
        let mut true_positives = 0;
        let mut false_positives = 0;
        let mut false_negatives = 0;

        for (i, (snapshot, expected)) in self.test_data.into_iter().zip(self.expected_signals.into_iter()).enumerate() {
            let inference_future = model.predict(snapshot);
            let result = timeout(self.config.inference_timeout, inference_future).await;

            match result {
                Ok(Ok(prediction)) => {
                    measurement.total_predictions += 1;

                    // Compare prediction with expected signal
                    let prediction_correct = self.evaluate_prediction(&prediction, expected);
                    
                    if prediction_correct {
                        measurement.correct_predictions += 1;
                        true_positives += 1;
                    } else {
                        if prediction.direction == SignalDirection::Buy || prediction.direction == SignalDirection::Sell {
                            false_positives += 1;
                        } else {
                            false_negatives += 1;
                        }
                    }
                }
                Ok(Err(e)) => {
                    eprintln!("Model prediction error for sample {}: {}", i, e);
                    false_negatives += 1;
                }
                Err(_) => {
                    eprintln!("Model prediction timeout for sample {}", i);
                    false_negatives += 1;
                }
            }
        }

        measurement.calculate_metrics(true_positives, false_positives, false_negatives);
        Ok(measurement)
    }

    fn evaluate_prediction(&self, prediction: &TradingSignal, expected: &TradingSignal) -> bool {
        // Check if direction matches and confidence is reasonable
        prediction.direction == expected.direction && 
        prediction.confidence >= 0.5 &&
        (prediction.strength - expected.strength).abs() <= 0.3
    }

    pub async fn test_inference_performance<T: MLModel + Send + Sync>(
        &self,
        model: &T,
        model_name: &str
    ) -> Result<InferencePerformance, Box<dyn std::error::Error + Send + Sync>> {
        // Warmup
        for _ in 0..self.config.model_warmup_iterations {
            if let Some(sample) = self.test_data.first() {
                let _ = model.predict(sample).await;
            }
        }

        let mut latencies = Vec::with_capacity(self.config.performance_iterations);
        let start_time = Instant::now();
        let mut successful_inferences = 0;

        for i in 0..self.config.performance_iterations {
            let sample_idx = i % self.test_data.len();
            let sample = &self.test_data[sample_idx];

            let inference_start = Instant::now();
            let result = model.predict(sample).await;
            let inference_duration = inference_start.elapsed();

            if result.is_ok() {
                latencies.push(inference_duration);
                successful_inferences += 1;
            }

            // Validate latency requirement
            assert!(
                inference_duration <= self.config.max_inference_latency,
                "Model {} inference latency {}μs exceeds maximum {}μs",
                model_name,
                inference_duration.as_micros(),
                self.config.max_inference_latency.as_micros()
            );
        }

        let total_duration = start_time.elapsed();
        latencies.sort();

        let avg_inference_time = if !latencies.is_empty() {
            latencies.iter().sum::<Duration>() / latencies.len() as u32
        } else {
            Duration::ZERO
        };

        let p50_latency = latencies.get(latencies.len() * 50 / 100).copied().unwrap_or(Duration::ZERO);
        let p95_latency = latencies.get(latencies.len() * 95 / 100).copied().unwrap_or(Duration::ZERO);
        let p99_latency = latencies.get(latencies.len() * 99 / 100).copied().unwrap_or(Duration::ZERO);

        let throughput_per_second = if total_duration.as_secs_f64() > 0.0 {
            successful_inferences as f64 / total_duration.as_secs_f64()
        } else {
            0.0
        };

        Ok(InferencePerformance {
            model_name: model_name.to_string(),
            avg_inference_time,
            p50_latency,
            p95_latency,
            p99_latency,
            throughput_per_second,
            memory_usage_mb: Self::estimate_memory_usage(),
        })
    }

    fn estimate_memory_usage() -> f64 {
        // Simplified memory usage estimation
        // In real implementation, this would use system monitoring
        64.0 // MB estimate
    }
}

#[tokio::test]
async fn test_tlob_transformer_integration() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    let config = MLTestConfig::default();
    let test_suite = MLModelTestSuite::new(config.clone()).await?;
    
    let tlob_model = TLOBTransformer::new().await?;
    
    // Test accuracy
    let accuracy = test_suite.test_model_accuracy(&tlob_model, "TLOB").await?;
    assert!(
        accuracy.accuracy_percentage >= config.accuracy_threshold * 100.0,
        "TLOB accuracy {}% below threshold {}%",
        accuracy.accuracy_percentage,
        config.accuracy_threshold * 100.0
    );

    // Test performance
    let performance = test_suite.test_inference_performance(&tlob_model, "TLOB").await?;
    assert!(
        performance.avg_inference_time <= config.max_inference_latency,
        "TLOB average inference time {}μs exceeds limit {}μs",
        performance.avg_inference_time.as_micros(),
        config.max_inference_latency.as_micros()
    );

    println!("✅ TLOB Transformer: {:.2}% accuracy, {:.2}μs avg latency", 
             accuracy.accuracy_percentage, performance.avg_inference_time.as_micros());
    
    Ok(())
}

#[tokio::test]
async fn test_mamba_model_integration() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    let config = MLTestConfig::default();
    let test_suite = MLModelTestSuite::new(config.clone()).await?;
    
    let mamba_model = MAMBAModel::new().await?;
    
    // Test accuracy
    let accuracy = test_suite.test_model_accuracy(&mamba_model, "MAMBA").await?;
    assert!(
        accuracy.accuracy_percentage >= config.accuracy_threshold * 100.0,
        "MAMBA accuracy {}% below threshold {}%",
        accuracy.accuracy_percentage,
        config.accuracy_threshold * 100.0
    );

    // Test performance
    let performance = test_suite.test_inference_performance(&mamba_model, "MAMBA").await?;
    assert!(
        performance.throughput_per_second >= 100.0,
        "MAMBA throughput {:.2} inferences/sec too low",
        performance.throughput_per_second
    );

    println!("✅ MAMBA Model: {:.2}% accuracy, {:.2} inferences/sec", 
             accuracy.accuracy_percentage, performance.throughput_per_second);
    
    Ok(())
}

#[tokio::test]
async fn test_liquid_neural_network_integration() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    let config = MLTestConfig::default();
    let test_suite = MLModelTestSuite::new(config.clone()).await?;
    
    let liquid_model = LiquidNeuralNetwork::new().await?;
    
    // Test accuracy and adaptability
    let accuracy = test_suite.test_model_accuracy(&liquid_model, "Liquid").await?;
    assert!(
        accuracy.f1_score >= 0.7,
        "Liquid Neural Network F1 score {:.3} below threshold 0.7",
        accuracy.f1_score
    );

    // Test performance
    let performance = test_suite.test_inference_performance(&liquid_model, "Liquid").await?;
    assert!(
        performance.p99_latency <= Duration::from_micros(100),
        "Liquid NN P99 latency {}μs exceeds 100μs",
        performance.p99_latency.as_micros()
    );

    println!("✅ Liquid Neural Network: F1={:.3}, P99={}μs", 
             accuracy.f1_score, performance.p99_latency.as_micros());
    
    Ok(())
}

#[tokio::test]
async fn test_temporal_fusion_transformer_integration() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    let config = MLTestConfig::default();
    let test_suite = MLModelTestSuite::new(config.clone()).await?;
    
    let tft_model = TemporalFusionTransformer::new().await?;
    
    // Test accuracy
    let accuracy = test_suite.test_model_accuracy(&tft_model, "TFT").await?;
    assert!(
        accuracy.precision >= 0.75 && accuracy.recall >= 0.75,
        "TFT precision={:.3} or recall={:.3} below 0.75",
        accuracy.precision, accuracy.recall
    );

    // Test performance
    let performance = test_suite.test_inference_performance(&tft_model, "TFT").await?;
    assert!(
        performance.memory_usage_mb <= 128.0,
        "TFT memory usage {:.1}MB exceeds 128MB limit",
        performance.memory_usage_mb
    );

    println!("✅ Temporal Fusion Transformer: P={:.3}, R={:.3}, Mem={:.1}MB", 
             accuracy.precision, accuracy.recall, performance.memory_usage_mb);
    
    Ok(())
}

#[tokio::test]
async fn test_dqn_agent_integration() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    let config = MLTestConfig::default();
    let test_suite = MLModelTestSuite::new(config.clone()).await?;
    
    let dqn_agent = DQNAgent::new().await?;
    
    // Test decision-making accuracy
    let accuracy = test_suite.test_model_accuracy(&dqn_agent, "DQN").await?;
    assert!(
        accuracy.accuracy_percentage >= 70.0,
        "DQN agent accuracy {}% below 70%",
        accuracy.accuracy_percentage
    );

    // Test performance and action selection speed
    let performance = test_suite.test_inference_performance(&dqn_agent, "DQN").await?;
    assert!(
        performance.avg_inference_time <= Duration::from_micros(25),
        "DQN action selection {}μs exceeds 25μs limit",
        performance.avg_inference_time.as_micros()
    );

    println!("✅ DQN Agent: {:.2}% accuracy, {}μs action selection", 
             accuracy.accuracy_percentage, performance.avg_inference_time.as_micros());
    
    Ok(())
}

#[tokio::test]
async fn test_ppo_agent_integration() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    let config = MLTestConfig::default();
    let test_suite = MLModelTestSuite::new(config.clone()).await?;
    
    let ppo_agent = PPOAgent::new().await?;
    
    // Test policy optimization
    let accuracy = test_suite.test_model_accuracy(&ppo_agent, "PPO").await?;
    assert!(
        accuracy.accuracy_percentage >= 75.0,
        "PPO agent accuracy {}% below 75%",
        accuracy.accuracy_percentage
    );

    // Test performance
    let performance = test_suite.test_inference_performance(&ppo_agent, "PPO").await?;
    assert!(
        performance.throughput_per_second >= 50.0,
        "PPO throughput {:.2} decisions/sec too low",
        performance.throughput_per_second
    );

    println!("✅ PPO Agent: {:.2}% accuracy, {:.2} decisions/sec", 
             accuracy.accuracy_percentage, performance.throughput_per_second);
    
    Ok(())
}

#[tokio::test]
async fn test_model_registry_integration() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    let config = MLTestConfig::default();
    let mut registry = ModelRegistry::new().await?;

    // Test model registration and loading
    let model_names = vec!["TLOB", "MAMBA", "Liquid", "TFT", "DQN", "PPO"];
    
    for model_name in &model_names {
        let model_exists = registry.has_model(model_name).await?;
        assert!(model_exists, "Model {} not found in registry", model_name);

        let model = registry.get_model(model_name).await?;
        assert!(model.is_some(), "Failed to load model {}", model_name);
    }

    // Test model ensemble prediction
    let test_suite = MLModelTestSuite::new(config.clone()).await?;
    if let Some(sample) = test_suite.test_data.first() {
        let ensemble_prediction = registry.predict_ensemble(sample).await?;
        assert!(
            ensemble_prediction.confidence >= 0.5,
            "Ensemble prediction confidence {:.3} too low",
            ensemble_prediction.confidence
        );
    }

    println!("✅ Model Registry: {} models loaded, ensemble prediction working", model_names.len());
    
    Ok(())
}

#[tokio::test]
async fn test_ml_risk_integration() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    let config = create_test_config();
    let ml_config = MLTestConfig::default();
    let test_suite = MLModelTestSuite::new(ml_config.clone()).await?;
    
    // Create integrated ML-Risk system
    let mut risk_engine = RiskEngine::new(config.risk.clone()).await?;
    let tlob_model = TLOBTransformer::new().await?;

    // Test ML signal integration with risk management
    if let Some(sample) = test_suite.test_data.first() {
        let ml_signal = tlob_model.predict(sample).await?;
        
        // Validate ML signal passes risk checks
        let risk_result = risk_engine.validate_signal(&ml_signal).await?;
        assert!(risk_result.is_valid, "ML signal failed risk validation: {:?}", risk_result.reason);

        // Test signal strength adjustment based on risk
        let adjusted_signal = risk_engine.adjust_signal_strength(&ml_signal).await?;
        assert!(
            adjusted_signal.strength <= ml_signal.strength,
            "Risk adjustment should not increase signal strength"
        );
    }

    println!("✅ ML-Risk Integration: Signal validation and adjustment working");
    
    Ok(())
}

#[tokio::test]
async fn test_concurrent_ml_inference() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    let config = MLTestConfig::default();
    let test_suite = MLModelTestSuite::new(config.clone()).await?;
    
    let tlob_model = std::sync::Arc::new(TLOBTransformer::new().await?);
    let mamba_model = std::sync::Arc::new(MAMBAModel::new().await?);

    // Test concurrent inference from multiple models
    let mut tasks = vec![];
    let num_concurrent = 10;

    for i in 0..num_concurrent {
        let tlob = tlob_model.clone();
        let mamba = mamba_model.clone();
        let sample = test_suite.test_data[i % test_suite.test_data.len()].clone();

        tasks.push(tokio::spawn(async move {
            let start = Instant::now();
            
            let (tlob_result, mamba_result) = tokio::join!(
                tlob.predict(&sample),
                mamba.predict(&sample)
            );
            
            let duration = start.elapsed();
            (tlob_result, mamba_result, duration)
        }));
    }

    // Wait for all concurrent inferences
    let results = futures::future::join_all(tasks).await;
    
    for (i, result) in results.into_iter().enumerate() {
        let (tlob_result, mamba_result, duration) = result?;
        
        assert!(tlob_result.is_ok(), "TLOB concurrent inference {} failed", i);
        assert!(mamba_result.is_ok(), "MAMBA concurrent inference {} failed", i);
        assert!(
            duration <= Duration::from_millis(100),
            "Concurrent inference {} took {}ms",
            i, duration.as_millis()
        );
    }

    println!("✅ Concurrent ML Inference: {} parallel inferences completed", num_concurrent);
    
    Ok(())
}

#[tokio::test]
async fn test_ml_model_memory_safety() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    let config = MLTestConfig::default();
    let test_suite = MLModelTestSuite::new(config.clone()).await?;
    
    // Test memory allocation patterns
    let tlob_model = TLOBTransformer::new().await?;
    
    // Run many inferences to test memory stability
    let iterations = 1000;
    let initial_memory = Self::estimate_memory_usage();
    
    for i in 0..iterations {
        let sample_idx = i % test_suite.test_data.len();
        let sample = &test_suite.test_data[sample_idx];
        
        let result = tlob_model.predict(sample).await;
        assert!(result.is_ok(), "Memory safety test failed at iteration {}", i);
        
        // Check for memory leaks every 100 iterations
        if i % 100 == 0 {
            let current_memory = Self::estimate_memory_usage();
            assert!(
                current_memory <= initial_memory * 1.5,
                "Potential memory leak detected: {}MB -> {}MB at iteration {}",
                initial_memory, current_memory, i
            );
        }
    }

    println!("✅ ML Memory Safety: {} inferences completed without memory issues", iterations);
    
    Ok(())
}

#[tokio::test]
async fn test_comprehensive_ml_validation() -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
    let config = MLTestConfig::default();
    let test_suite = MLModelTestSuite::new(config.clone()).await?;
    
    let models = vec![
        ("TLOB", Box::new(TLOBTransformer::new().await?) as Box<dyn MLModel + Send + Sync>),
        ("MAMBA", Box::new(MAMBAModel::new().await?) as Box<dyn MLModel + Send + Sync>),
        ("Liquid", Box::new(LiquidNeuralNetwork::new().await?) as Box<dyn MLModel + Send + Sync>),
    ];

    let mut total_accuracy = 0.0;
    let mut total_throughput = 0.0;

    for (name, model) in &models {
        // Test accuracy
        let accuracy = test_suite.test_model_accuracy(model.as_ref(), name).await?;
        assert!(
            accuracy.accuracy_percentage >= config.accuracy_threshold * 100.0,
            "{} accuracy {}% below threshold",
            name, accuracy.accuracy_percentage
        );

        // Test performance
        let performance = test_suite.test_inference_performance(model.as_ref(), name).await?;
        assert!(
            performance.avg_inference_time <= config.max_inference_latency,
            "{} latency {}μs exceeds limit",
            name, performance.avg_inference_time.as_micros()
        );

        total_accuracy += accuracy.accuracy_percentage;
        total_throughput += performance.throughput_per_second;

        println!("✅ {}: {:.2}% accuracy, {:.2} inf/sec, {}μs avg", 
                 name, accuracy.accuracy_percentage, 
                 performance.throughput_per_second,
                 performance.avg_inference_time.as_micros());
    }

    let avg_accuracy = total_accuracy / models.len() as f64;
    let total_system_throughput = total_throughput;

    assert!(
        avg_accuracy >= 75.0,
        "Overall ML system accuracy {:.2}% below 75%",
        avg_accuracy
    );

    assert!(
        total_system_throughput >= 200.0,
        "Total ML system throughput {:.2} inf/sec below 200",
        total_system_throughput
    );

    println!("🎯 COMPREHENSIVE ML VALIDATION PASSED");
    println!("   Average Accuracy: {:.2}%", avg_accuracy);
    println!("   Total Throughput: {:.2} inferences/sec", total_system_throughput);
    println!("   All {} models meet production requirements", models.len());
    
    Ok(())
}