feat(aura-std): EMA node (SMA-seeded, O(1) per cycle)

Add Ema, the std-lib's first recursive/stateful producer: exponential moving average with alpha = 2/(length+1). The building block MACD (and DEMA, MACD-histogram, etc.) needs, and a worked example of a node whose param sizes its math, not its input window.

Seeding is ratified as SMA-seeded (ta-lib convention): the EMA stays silent until it has seen length samples, seeds with their SMA, then runs the recurrence. Chosen over a first-value seed for three substantive reasons -- warm-up consistency with Sma (both emit None until length samples), parity with the EMA traders expect (ta-lib MACD seeds this way), and statistical honesty (no average claimed from a single observation). A first-value seed was the proof-of-concept convenience; rejected.

Performance: recursive state means lookback = 1 (length sizes alpha, not the window); eval is O(1) time and state via ema += alpha*(x - ema), with no hot-path allocation (the output buffer is reused).

Tested: warm-up-then-smooth (a step input separates it from a plain SMA), length-1 identity, label, and factory/schema agreement.
This commit is contained in:
2026-06-07 23:20:25 +02:00
parent a4cfe5c53f
commit ae1d4564f5
2 changed files with 170 additions and 0 deletions
+168
View File
@@ -0,0 +1,168 @@
//! `Ema` — exponential moving average of one f64 input, smoothing constant
//! `alpha = 2 / (length + 1)` (the conventional length↔alpha identity, so an
//! `Ema` and an `Sma` of the same `length` are roughly comparable).
//!
//! Two design points worth stating:
//!
//! - **Seeding (ta-lib convention).** Like `Sma`, the EMA stays silent (`None`)
//! until it has seen `length` samples, then seeds itself with the *SMA* of those
//! first `length` values and runs the recurrence from there. This keeps warm-up
//! semantics uniform across the moving-average nodes, matches the EMA traders
//! know (ta-lib's MACD seeds its EMAs this way), and avoids claiming an average
//! from a single observation. (A first-value seed — emit from sample 1 — is the
//! other common convention; rejected here for the consistency/honesty reasons
//! above.)
//! - **Recursive, so `lookback = 1`.** The running average lives in internal
//! state, so the node reads only the newest sample each cycle — `length` sizes
//! `alpha`, not the input window. `eval` is O(1) time, O(1) state, and allocates
//! nothing on the hot path (the output buffer is reused).
use aura_core::{
Ctx, FieldSpec, Firing, InputSpec, LeafFactory, Node, NodeSchema, ParamSpec, Scalar, ScalarKind,
};
/// Exponential moving average of one f64 input, smoothing `alpha = 2/(length+1)`,
/// seeded with the SMA of the first `length` samples.
pub struct Ema {
alpha: f64,
length: usize,
// Warm-up accumulator: until `length` samples have arrived the node is silent
// and sums them; on the `length`-th sample it locks in the SMA seed (`seeded`)
// and `ema` carries the running value from then on.
seeded: bool,
warmup_sum: f64,
count: usize,
ema: f64,
out: [Scalar; 1],
}
impl Ema {
/// Build an EMA of window `length` (must be >= 1); `alpha = 2/(length+1)`
/// (`length == 1` gives `alpha == 1`, the identity).
pub fn new(length: usize) -> Self {
assert!(length >= 1, "EMA length must be >= 1");
Self {
alpha: 2.0 / (length as f64 + 1.0),
length,
seeded: false,
warmup_sum: 0.0,
count: 0,
ema: 0.0,
out: [Scalar::F64(0.0)],
}
}
/// The param-generic recipe for a blueprint leaf: declares `length` and builds
/// through `Ema::new` (the single sizing/validation gate; the slice is
/// kind-checked before `build` runs, so the typed read is total).
pub fn factory() -> LeafFactory {
LeafFactory::new(
"EMA",
vec![ParamSpec { name: "length".into(), kind: ScalarKind::I64 }],
|p| Box::new(Ema::new(p[0].as_i64().expect("length slot is I64") as usize)),
)
}
}
impl Node for Ema {
fn schema(&self) -> NodeSchema {
NodeSchema {
inputs: vec![InputSpec {
kind: ScalarKind::F64,
// recursive: the running average lives in internal state, so only
// the newest sample is read — `length` sizes alpha, not the window.
lookback: 1,
firing: Firing::Any,
}],
output: vec![FieldSpec { name: "value", kind: ScalarKind::F64 }],
params: vec![ParamSpec { name: "length".into(), kind: ScalarKind::I64 }],
}
}
fn eval(&mut self, ctx: Ctx<'_>) -> Option<&[Scalar]> {
let w = ctx.f64_in(0);
if w.is_empty() {
return None; // no sample yet
}
let x = w[0]; // index 0 = newest (financial indexing)
if !self.seeded {
// warm up over `length` samples, then seed with their SMA (ta-lib
// convention) — silent until ready, exactly like `Sma`.
self.warmup_sum += x;
self.count += 1;
if self.count < self.length {
return None;
}
self.ema = self.warmup_sum / self.length as f64;
self.seeded = true;
} else {
// O(1) recurrence: one sub, one mul, one add.
self.ema += self.alpha * (x - self.ema);
}
self.out[0] = Scalar::F64(self.ema);
Some(&self.out)
}
fn label(&self) -> String {
format!("EMA({})", self.length)
}
}
#[cfg(test)]
mod tests {
use super::*;
use aura_core::{AnyColumn, Timestamp};
#[test]
fn ema_warms_up_like_sma_then_smooths() {
// length 3 -> alpha = 0.5: silent for the first 2 samples, seeds with the
// SMA of the first 3, then runs the recurrence. The jump to 10 separates the
// EMA from a plain SMA (EMA = 7.0 here, SMA(4,6,10) would be 6.667); all
// magnitudes are exact in f64.
let mut ema = Ema::new(3);
let mut inputs = vec![AnyColumn::with_capacity(ScalarKind::F64, 1)];
let feed = [2.0_f64, 4.0, 6.0, 10.0];
// None, None, SMA(2,4,6) = 4.0, then 4.0 + 0.5*(10 - 4.0) = 7.0
let expect = [None, None, Some(4.0), Some(7.0)];
for (v, want) in feed.iter().zip(expect) {
inputs[0].push(Scalar::F64(*v)).unwrap();
let got = ema.eval(Ctx::new(&inputs, Timestamp(0)));
match want {
None => assert_eq!(got, None),
Some(m) => assert_eq!(got, Some([Scalar::F64(m)].as_slice())),
}
}
}
#[test]
fn ema_length_one_is_identity() {
// alpha = 2/2 = 1.0: the seed locks in on the first sample (length 1) and
// the recurrence reduces to ema = x.
let mut ema = Ema::new(1);
let mut inputs = vec![AnyColumn::with_capacity(ScalarKind::F64, 1)];
inputs[0].push(Scalar::F64(7.0)).unwrap();
assert_eq!(ema.eval(Ctx::new(&inputs, Timestamp(0))), Some([Scalar::F64(7.0)].as_slice()));
inputs[0].push(Scalar::F64(9.0)).unwrap();
assert_eq!(ema.eval(Ctx::new(&inputs, Timestamp(0))), Some([Scalar::F64(9.0)].as_slice()));
}
#[test]
fn label_carries_the_window() {
assert_eq!(Ema::new(12).label(), "EMA(12)");
assert_eq!(Ema::new(26).label(), "EMA(26)");
}
#[test]
fn factory_params_match_built_node_schema() {
let f = Ema::factory();
let built = f.build(&[Scalar::I64(9)]);
assert_eq!(f.params(), built.schema().params.as_slice());
// the built node carries the declared I64 `length` knob
assert_eq!(built.schema().params[0].name, "length");
assert_eq!(built.schema().params[0].kind, ScalarKind::I64);
}
}
+2
View File
@@ -16,6 +16,7 @@
//! average — a worked producer node proving the `aura-core` `Node` contract.
mod add;
mod ema;
mod exposure;
mod lincomb;
mod recorder;
@@ -23,6 +24,7 @@ mod sim_broker;
mod sma;
mod sub;
pub use add::Add;
pub use ema::Ema;
pub use exposure::Exposure;
pub use lincomb::LinComb;
pub use recorder::Recorder;