foxhunt/ml/tests/real_data_helpers.rs

//! Real Market Data Helpers for ML Model Unit Tests
//!
//! Provides utilities to load real BTC/ETH Parquet data and convert it to formats
//! used by ML model tests (MAMBA-2, DQN, PPO, TFT).
//!
//! # Architecture
//!
//! - Loads real market data from test_data/real/parquet/*.parquet files
//! - Converts to model-specific formats (states, features, time series)
//! - Provides realistic data distributions for proper model testing
//!
//! # Usage
//!
//! ```ignore
//! use real_data_helpers::{RealDataLoader, load_dqn_states, load_tft_sequences};
//!
//! // DQN/PPO tests
//! let states = load_dqn_states(100).await?;
//!
//! // TFT/MAMBA-2 tests
//! let sequences = load_tft_sequences(50, 20).await?;
//! ```

use anyhow::{Context, Result};
use data::parquet_persistence::{MarketDataEvent, ParquetMarketDataReader};
use std::path::PathBuf;

/// Path to real test data directory (relative to workspace root)
const REAL_DATA_PATH: &str = "test_data/real/parquet";
const BTC_FILE: &str = "BTC-USD_30day_2024-09.parquet";
const ETH_FILE: &str = "ETH-USD_30day_2024-09.parquet";

/// Real market data loader for ML tests
pub(crate) struct RealDataLoader {
    base_path: String,
}

impl RealDataLoader {
    /// Create new loader with workspace-relative path
    pub(crate) fn new() -> Self {
        let manifest_dir = env!("CARGO_MANIFEST_DIR");
        let base_path = PathBuf::from(manifest_dir)
            .parent()
            .expect("Failed to get workspace root")
            .join(REAL_DATA_PATH);

        Self {
            base_path: base_path.to_string_lossy().to_string(),
        }
    }

    /// Check if real data files exist
    pub(crate) fn files_exist(&self) -> bool {
        let btc_path = PathBuf::from(&self.base_path).join(BTC_FILE);
        let eth_path = PathBuf::from(&self.base_path).join(ETH_FILE);
        btc_path.exists() && eth_path.exists()
    }

    /// Load raw market events from BTC data
    pub(crate) async fn load_btc_events(&self, count: usize) -> Result<Vec<MarketDataEvent>> {
        self.load_events(BTC_FILE, count).await
    }

    /// Load raw market events from ETH data
    pub(crate) async fn load_eth_events(&self, count: usize) -> Result<Vec<MarketDataEvent>> {
        self.load_events(ETH_FILE, count).await
    }

    /// Load events from a specific file
    async fn load_events(&self, filename: &str, count: usize) -> Result<Vec<MarketDataEvent>> {
        let reader = ParquetMarketDataReader::new(self.base_path.clone());
        let mut events = reader
            .read_file(filename)
            .await
            .with_context(|| format!("Failed to load {}", filename))?;

        // Take only the requested number of events
        events.truncate(count);
        Ok(events)
    }
}

impl Default for RealDataLoader {
    fn default() -> Self {
        Self::new()
    }
}

/// Convert market events to DQN/PPO state vectors
///
/// Each state vector contains [price, volume, bid_ask_spread, momentum, volatility]
pub(crate) fn events_to_dqn_states(events: &[MarketDataEvent], state_dim: usize) -> Vec<Vec<f32>> {
    if events.is_empty() {
        return vec![];
    }

    events
        .windows(5) // Need 5 events to calculate momentum/volatility
        .map(|window| {
            let current = &window[4];
            let prev = &window[0];

            // Extract OHLCV features
            let price = current.price.unwrap_or(0.0) as f32;
            let volume = current.quantity.unwrap_or(0.0) as f32;
            let high = current.high.unwrap_or(price as f64) as f32;
            let low = current.low.unwrap_or(price as f64) as f32;

            // Calculate technical indicators
            let bid_ask_spread = (high - low) / price.max(1e-6);
            let momentum =
                (price - prev.price.unwrap_or(price as f64) as f32) / price.max(1e-6);

            // Calculate volatility from window
            let prices: Vec<f32> = window.iter().map(|e| e.price.unwrap_or(0.0) as f32).collect();
            let mean_price = prices.iter().sum::<f32>() / prices.len() as f32;
            let variance = prices
                .iter()
                .map(|p| (p - mean_price).powi(2))
                .sum::<f32>()
                / prices.len() as f32;
            let volatility = variance.sqrt();

            // Build state vector (pad or truncate to state_dim)
            let mut state = vec![price, volume, bid_ask_spread, momentum, volatility];
            state.resize(state_dim, 0.0);
            state
        })
        .collect()
}

/// Convert market events to TFT/MAMBA-2 time series sequences
///
/// Returns (sequences, targets) where:
/// - sequences: [batch, seq_len, features]
/// - targets: [batch, prediction_horizon]
pub(crate) fn events_to_time_series(
    events: &[MarketDataEvent],
    seq_len: usize,
    features_per_event: usize,
) -> Vec<Vec<f32>> {
    if events.is_empty() || events.len() < seq_len {
        return vec![];
    }

    events
        .windows(seq_len)
        .map(|window| {
            window
                .iter()
                .flat_map(|event| {
                    let price = event.price.unwrap_or(0.0) as f32;
                    let volume = event.quantity.unwrap_or(0.0) as f32;
                    let high = event.high.unwrap_or(price as f64) as f32;
                    let low = event.low.unwrap_or(price as f64) as f32;
                    let open = event.open.unwrap_or(price as f64) as f32;

                    // Build feature vector [open, high, low, close, volume, ...]
                    let mut features = vec![open, high, low, price, volume];
                    features.resize(features_per_event, 0.0);
                    features
                })
                .collect()
        })
        .collect()
}

/// Load DQN states (convenience wrapper)
pub(crate) async fn load_dqn_states(count: usize, state_dim: usize) -> Result<Vec<Vec<f32>>> {
    let loader = RealDataLoader::new();

    // Check if files exist, otherwise return empty (tests will skip)
    if !loader.files_exist() {
        return Ok(vec![]);
    }

    let events = loader.load_btc_events(count + 5).await?; // +5 for window calculation
    Ok(events_to_dqn_states(&events, state_dim))
}

/// Load TFT/MAMBA-2 sequences (convenience wrapper)
pub(crate) async fn load_tft_sequences(
    count: usize,
    seq_len: usize,
    features_per_event: usize,
) -> Result<Vec<Vec<f32>>> {
    let loader = RealDataLoader::new();

    // Check if files exist, otherwise return empty (tests will skip)
    if !loader.files_exist() {
        return Ok(vec![]);
    }

    let events = loader.load_btc_events(count + seq_len).await?;
    Ok(events_to_time_series(&events, seq_len, features_per_event))
}

/// Check if real data is available for tests
pub(crate) async fn real_data_available() -> bool {
    let loader = RealDataLoader::new();
    loader.files_exist()
}

#[cfg(test)]
mod tests {
    use super::*;

    #[tokio::test]
    async fn test_real_data_loader_creation() {
        let loader = RealDataLoader::new();
        assert!(!loader.base_path.is_empty());
    }

    #[tokio::test]
    #[ignore = "requires real Parquet files"]
    async fn test_load_btc_events() {
        let loader = RealDataLoader::new();
        if !loader.files_exist() {
            eprintln!("Skipping test: real data files not found");
            return;
        }

        let events = loader.load_btc_events(100).await.unwrap();
        assert_eq!(events.len(), 100);
        assert!(events[0].price.is_some());
        assert!(events[0].quantity.is_some());
    }

    #[tokio::test]
    #[ignore = "requires real Parquet files"]
    async fn test_events_to_dqn_states() {
        let loader = RealDataLoader::new();
        if !loader.files_exist() {
            eprintln!("Skipping test: real data files not found");
            return;
        }

        let events = loader.load_btc_events(50).await.unwrap();
        let states = events_to_dqn_states(&events, 32);

        assert!(!states.is_empty());
        assert_eq!(states[0].len(), 32);

        // Verify state contains non-zero values
        assert!(states[0].iter().any(|&v| v != 0.0));
    }

    #[tokio::test]
    #[ignore = "requires real Parquet files"]
    async fn test_events_to_time_series() {
        let loader = RealDataLoader::new();
        if !loader.files_exist() {
            eprintln!("Skipping test: real data files not found");
            return;
        }

        let events = loader.load_btc_events(50).await.unwrap();
        let sequences = events_to_time_series(&events, 10, 5);

        assert!(!sequences.is_empty());
        assert_eq!(sequences[0].len(), 10 * 5); // seq_len * features_per_event

        // Verify sequences contain non-zero values
        assert!(sequences[0].iter().any(|&v| v != 0.0));
    }

    #[tokio::test]
    async fn test_load_dqn_states_wrapper() {
        let states = load_dqn_states(50, 32).await.unwrap();

        // If no data, should return empty (tests will skip)
        if states.is_empty() {
            eprintln!("No real data available, test will skip");
            return;
        }

        assert_eq!(states[0].len(), 32);
    }

    #[tokio::test]
    async fn test_load_tft_sequences_wrapper() {
        let sequences = load_tft_sequences(30, 10, 5).await.unwrap();

        // If no data, should return empty (tests will skip)
        if sequences.is_empty() {
            eprintln!("No real data available, test will skip");
            return;
        }

        assert_eq!(sequences[0].len(), 10 * 5);
    }

    #[tokio::test]
    async fn test_real_data_available() {
        let available = real_data_available().await;
        println!("Real data available: {}", available);
    }
}