CoordinateNormalizer — rolling z-score normalizer for SpacetimeEvent. More...

#include "srfm/manifold.hpp"
#include <cassert>
#include <cstddef>
#include <deque>
#include <optional>

Classes
class	srfm::CoordinateNormalizer

Namespaces
namespace	srfm

Detailed Description

CoordinateNormalizer — rolling z-score normalizer for SpacetimeEvent.

Module: Coordinate Normalizer

Responsibility

Normalize the spatial coordinates of SpacetimeEvent instances using a rolling window of observed values before they are passed to SpacetimeInterval::compute.

Why This Matters

Raw market coordinates differ by many orders of magnitude:

price ~100
volume ~1e6
momentum ~0.01

Without normalization, volume dominates ds² by ~10 orders of magnitude, making interval classification (timelike/spacelike) meaningless. After z-score normalization each coordinate contributes equally.

Formula

For each spatial coordinate x, rolling window of n observations: x_norm = (x - rolling_mean(x)) / rolling_stddev(x)

Time coordinate passes through unchanged.

Warm-up Guard

Before at least min_samples data points have been accumulated (default = window_size, minimum 20), the z-scores are potentially unreliable due to insufficient statistical history. warmed_up() returns false during this period, and normalize() returns a zero-valued event (all spatial coordinates set to 0.0 with time unchanged) instead of a potentially wild z-score. Callers should check warmed_up() before using the output of normalize() for trading decisions.

Edge Cases

stddev < 1e-9 (flat / constant series): returns 0.0 (no variance to normalize)
Window not yet full OR min_samples not yet reached: returns 0.0 for spatial coords
Single-sample window: mean = x, stddev = 0 → returns 0.0

Guarantees

Stateful: call in time order for correct rolling statistics
Never divides by zero
Time coordinate always unchanged
No dynamic allocation on hot path (deque amortized)

Definition in file normalizer.hpp.

Classes

Namespaces