Trait DualModeFeature

Source

pub trait DualModeFeature<const N: usize = 32, const T: usize = 16>: Send + Sync {
    // Required methods
    fn calculate_incremental(
        &mut self,
        update: &MarketUpdate<N, T>,
    ) -> FeatureValue;
    fn calculate_batch(&self, data: &[MarketData]) -> Vec<FeatureValue>;
    fn name(&self) -> &'static str;
    fn reset(&mut self);
}

Expand description

Trait for features that support both real-time and batch calculation

This trait enables feature calculations to work in both real-time streaming mode and batch processing mode, ensuring identical results between modes.

§Type Parameters

N - The capacity for order book levels (default: 32)
- Controls the maximum number of bid/ask price and quantity pairs that can be stored inline
- Uses SmallVec to avoid heap allocation for typical order book depths
- Default value: 32 levels, suitable for most market making strategies
- Common values:
  - 5-10: Top-of-book strategies (L1 data only)
  - 16-32: Standard market making and arbitrage strategies
  - 64-128: Deep market making and full depth analysis
- Performance implications: Values up to N are stack-allocated; exceeding this causes heap allocation
T - The capacity for trades buffer (default: 16)
- Controls the maximum number of trade updates that can be stored inline per market update
- Uses SmallVec to avoid heap allocation for typical trade bursts
- Default value: 16 trades, suitable for most market conditions
- Common values:
  - 8-16: Normal market conditions and low-frequency strategies
  - 32-64: High-volume periods and burst trade handling
  - 128+: Extremely liquid assets or high-frequency aggregation
- Performance implications: Values up to T are stack-allocated; exceeding this causes heap allocation

§Buffer Sizing Guidelines

Choose buffer sizes based on your specific use case and market characteristics:

§Order Book Levels (N)

Level 1 only: Use N = 5 for top-of-book strategies
Standard depth: Use N = 16-32 for typical market making
Deep analysis: Use N = 64-128 for full order book strategies
Monitor usage: Track actual order book depths to optimize N

§Trade Buffer (T)

Low frequency: Use T = 8-16 for slower strategies
Standard flow: Use T = 16-32 for normal market conditions
High frequency: Use T = 64-128 for burst-heavy markets
Monitor usage: Track trade volumes per update to optimize T

§Performance Considerations

Stack allocation: Values within capacity are stored inline (fastest)
Heap allocation: Exceeding capacity triggers heap allocation (slower)
Memory locality: Smaller buffers improve cache performance
Compilation: Different const values create distinct types at compile time
SIMD optimization: Powers-of-2 sizes may enable better vectorization

§Usage Examples

// Default configuration (32 levels, 16 trades) - suitable for most strategies
impl DualModeFeature for MyFeature {
    fn calculate_incremental(&mut self, update: &MarketUpdate) -> FeatureValue {
        // Implementation using default 32 levels, 16 trades
    }
}

// Top-of-book strategy - minimal memory footprint
impl DualModeFeature<5, 8> for TopOfBookFeature {
    fn calculate_incremental(&mut self, update: &MarketUpdate<5, 8>) -> FeatureValue {
        // Implementation optimized for L1 data only
    }
}

// Deep market making - full order book analysis
impl DualModeFeature<64, 32> for DeepMMFeature {
    fn calculate_incremental(&mut self, update: &MarketUpdate<64, 32>) -> FeatureValue {
        // Implementation for deep order book strategies
    }
}

// High-frequency burst handling
impl DualModeFeature<32, 128> for BurstFeature {
    fn calculate_incremental(&mut self, update: &MarketUpdate<32, 128>) -> FeatureValue {
        // Implementation for handling trade bursts
    }
}