TEMA — Triple Exponential Moving Average¶
Further reduces lag with three EMA layers: 3 * EMA - 3 * EMA(EMA) + EMA(EMA(EMA)). Reacts to price changes more quickly than DEMA at the cost of additional sensitivity to noise.
Inputs: [real] | Options: [period] | Outputs: [tema]
Basic¶
use tulip_rs::indicators::tema::{Tema, TIndicatorState, Indicator};
let close = vec![81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36_f64];
let (outputs, _state) = Tema::indicator(&[close.as_slice()], &[14.0], None).unwrap();
println!("TEMA(14): {:?}", outputs[0]);
// State continuation
let partial = close[..8].to_vec();
let (outputs2, mut state) = Tema::indicator(&[partial.as_slice()], &[14.0], None).unwrap();
println!("Partial TEMA: {:?}", outputs2[0]);
let new_close = close[8..].to_vec();
let continued = state.batch_indicator(&[new_close.as_slice()], None).unwrap();
println!("Continued TEMA: {:?}", continued[0]);
#include "tulip_rs_ffi.h"
double close[] = {81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36};
double options[TEMA_OPTIONS] = {14.0}; // period
const double *inputs[TEMA_INPUTS] = {close};
/* Full computation (check r.error == C_INDICATOR_ERROR_OK in real code) */
CIndicatorResult r = tema_indicator(inputs, 10, options, NULL, 0);
/* r.outputs[0] -> the TEMA(14) series, length r.output_lens[0] */
tulip_ffi_result_free(r);
tema_state_free(r.state);
/* Partial computation + state continuation */
CIndicatorResult p = tema_indicator(inputs, 8, options, NULL, 0);
double new_close[] = {84.55, 84.36};
const double *new_inputs[TEMA_INPUTS] = {new_close};
CBatchResult b = tema_batch(p.state, new_inputs, 2, NULL, 0);
/* b.outputs[0] -> TEMA values for just the two new bars */
tulip_ffi_batch_result_free(b);
tulip_ffi_result_free(p);
tema_state_free(p.state);
import "github.com/me60732/tulip_rs_go/indicators"
close := []float64{81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36}
options := []float64{14.0} // period
// Full computation — Rows are zero-copy views, valid until Close.
res, st, _ := indicators.Tema.Indicator(close, options, nil)
fmt.Println(res.Rows[0]) // TEMA(14) values
res.Close()
st.Close()
// Partial computation + state continuation.
res2, st2, _ := indicators.Tema.Indicator(close[:8], options, nil)
res2.Close() // outputs consumed or closed; state stays live
batch, _ := st2.Batch(close[8:], nil)
fmt.Println(batch.Rows[0]) // continued TEMA values
batch.Close()
st2.Close()
import org.tuliprs.*;
import org.tuliprs.indicators.Tema;
double[] close = {81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36};
double[] options = {14.0}; // period
// Full computation — output rows are zero-copy views, valid until close().
Outcome oc = Tema.indicator(new double[][] {close}, options);
try (Result res = oc.result(); State st = oc.state()) {
System.out.println(java.util.Arrays.toString(res.toDoubleArray(0))); // TEMA(14) values
}
// Partial computation + state continuation.
Outcome p = Tema.indicator(
new double[][] {java.util.Arrays.copyOfRange(close, 0, 8)}, options);
try (Result pr = p.result(); State st = p.state()) {
Result br = st.batch(new double[][] {java.util.Arrays.copyOfRange(close, 8, 10)});
try (br) {
System.out.println(java.util.Arrays.toString(br.toDoubleArray(0))); // continued TEMA values
}
}
import numpy as np
import tulip_rs
close = np.array([81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36], dtype=np.float64)
outputs, state = tulip_rs.indicators.tema.indicator([close], [14.0])
print("TEMA(14):", outputs[0])
# State continuation
partial = close[:8]
outputs2, state = tulip_rs.indicators.tema.indicator([partial], [14.0])
new_close = close[8:]
continued = state.batch_indicator([new_close])
print("Continued TEMA:", continued[0])
import * as ti from 'tulip-rs-node';
const close = Float64Array.from([81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36,
85.53, 86.54, 86.89, 87.77, 87.29]);
const [outputs, state] = ti.tema.indicator([close], [14]);
console.log('TEMA(14):', outputs[0]);
// State continuation
const [, state2] = ti.tema.indicator([close.slice(0, -5)], [14]);
const continued = state2.batchIndicator([close.slice(-5)]);
console.log('Continued TEMA:', continued[0]);
import { init } from 'tulip-rs-wasm';
import * as ti from 'tulip-rs-wasm';
await init(); // bundler resolves the WASM asset automatically
const close = [81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36,
85.53, 86.54, 86.89, 87.77, 87.29];
const [outputs, state] = ti.tema.indicator([close], [14]);
console.log('TEMA(14):', outputs[0]);
// State continuation
const [, state2] = ti.tema.indicator([close.slice(0, -5)], [14]);
const continued = state2.batchIndicator([close.slice(-5)]);
console.log('Continued TEMA:', continued[0]);
Optional Outputs¶
tema exposes 2 optional outputs: "dema", "ema". Pass a boolean mask as the third argument — one bool per optional output, in order.
use tulip_rs::indicators::tema::{Tema, TIndicatorState, Indicator};
let close = vec![81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36_f64];
// Request dema but not ema
let mask = [true, false]; // one per optional output
let (outputs, _state) = Tema::indicator(&[close.as_slice()], &[5.0], Some(&mask)).unwrap();
let tema = &outputs[0]; // tema (primary)
let dema = &outputs[1]; // "dema" (optional — requested)
// "ema" not requested
tema exposes 2 optional outputs: dema, ema. Pass a boolean mask — one bool per optional output.
#include "tulip_rs_ffi.h"
double close[] = {81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36};
double options[TEMA_OPTIONS] = {5.0}; // period
const double *inputs[TEMA_INPUTS] = {close};
bool optional_outputs[] = {true, false}; // dema requested, ema not requested
CIndicatorResult r = tema_indicator(inputs, 10, options, optional_outputs, 2);
/* r.outputs[0] -> tema (primary) */
/* r.outputs[1] -> dema (optional — requested) */
tulip_ffi_result_free(r);
tema_state_free(r.state);
tema exposes 2 optional outputs: dema, ema. Pass a boolean mask — one bool per optional output.
import "github.com/me60732/tulip_rs_go/indicators"
close := []float64{81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36}
options := []float64{5.0} // period
// Request dema but not ema
mask := []bool{true, false} // one per optional output
res, st, _ := indicators.Tema.Indicator(close, options, mask)
tema := res.Rows[0] // tema (primary)
dema := res.Rows[1] // "dema" (optional — requested)
// "ema" not requested
res.Close()
st.Close()
tema exposes 2 optional outputs: dema, ema. Pass a boolean mask — one boolean per optional output, in order.
import org.tuliprs.*;
import org.tuliprs.indicators.Tema;
// (close series as in the Basic tab)
double[] options = {5.0}; // period
boolean[] mask = {true, false}; // dema requested, ema not requested
Outcome oc = Tema.indicator(new double[][] {close}, options, mask);
try (Result res = oc.result()) {
// row 0 = tema (primary), row 1 = dema; ema not requested
System.out.println(java.util.Arrays.toString(res.toDoubleArray(1))); // dema
}
oc.state().close();
import numpy as np
import tulip_rs
close = np.array([81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36], dtype=np.float64)
# Request dema but not ema
outputs, state = tulip_rs.indicators.tema.indicator(
[close], [5.0],
optional_outputs=[True, False],
)
tema = outputs[0] # tema (primary)
dema = outputs[1] # "dema" (optional — requested)
# "ema" not requested
tema exposes 2 optional outputs: dema, ema. Pass a boolean mask — one bool per optional output.
// Request all optional outputs
const [allOut] = ti.tema.indicator([close], [5], [true, true]);
const tema = allOut[0]; // primary
const dema = allOut[1]; // optional 0: dema
const ema = allOut[2]; // optional 1: ema
// Request only dema
const [partial] = ti.tema.indicator([close], [5], [true, false]);
The WASM API is identical to Node.js — pass the boolean mask as the third argument.
// Request all optional outputs
const [allOut] = ti.tema.indicator([close], [5], [true, true]);
const tema = allOut[0]; // primary
const dema = allOut[1]; // optional 0: dema
const ema = allOut[2]; // optional 1: ema
// Request only dema
const [partial] = ti.tema.indicator([close], [5], [true, false]);
SIMD¶
By assets — same period applied to 4 assets in parallel:
use tulip_rs::indicators::tema::{Tema, Indicator};
let a1 = vec![81.59, 81.06, 82.87, 83.00, 83.61, 83.15, 82.84, 83.99, 84.55, 84.36_f64];
let a2 = vec![72.10, 72.85, 73.40, 73.00, 74.20, 74.85, 75.10, 75.60, 76.00, 76.50_f64];
let a3 = vec![55.30, 55.80, 56.10, 56.40, 56.90, 57.20, 57.50, 57.80, 58.10, 58.40_f64];
let a4 = vec![100.1, 100.5, 101.0, 101.3, 101.8, 102.0, 102.5, 103.0, 103.3, 103.8_f64];
let inputs: [&[&[f64]; 1]; 4] = [
&[a1.as_slice()],
&[a2.as_slice()],
&[a3.as_slice()],
&[a4.as_slice()],
];
let results = Tema::indicator_by_assets::<4>(&inputs, &[14.0], None).unwrap();
for (i, asset_outputs) in results.iter().enumerate() {
println!("Asset {}: {:?}", i + 1, asset_outputs[0]);
}
By options — same asset, 4 different periods in parallel:
use tulip_rs::indicators::tema::{Tema, IndicatorByOptions};
let close = vec![81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36_f64];
let opts: [&[f64; 1]; 4] = [&[5.0], &[10.0], &[14.0], &[20.0]];
let results = Tema::indicator_by_options::<4>(&[close.as_slice()], &opts, None).unwrap();
for (i, opt_outputs) in results.iter().enumerate() {
println!("Period set {}: {:?}", i + 1, opt_outputs[0]);
}
By assets — same period applied to 4 assets in one call (N must be 2/4/8/16):
double a1[] = {81.59, 81.06, 82.87, 83.00, 83.61, 83.15, 82.84, 83.99, 84.55, 84.36};
double a2[] = {72.10, 72.85, 73.40, 73.00, 74.20, 74.85, 75.10, 75.60, 76.00, 76.50};
double a3[] = {55.30, 55.80, 56.10, 56.40, 56.90, 57.20, 57.50, 57.80, 58.10, 58.40};
double a4[] = {100.1, 100.5, 101.0, 101.3, 101.8, 102.0, 102.5, 103.0, 103.3, 103.8};
/* one [INPUTS]-long pointer array per asset */
const double *asset1[TEMA_INPUTS] = {a1};
const double *asset2[TEMA_INPUTS] = {a2};
const double *asset3[TEMA_INPUTS] = {a3};
const double *asset4[TEMA_INPUTS] = {a4};
const double *const *const simd_inputs[4] = {asset1, asset2, asset3, asset4};
CSimdResult r = tema_simd_by_assets(simd_inputs, 4, 10, options, NULL, 0);
for (uintptr_t i = 0; i < r.num_results; i++) {
/* r.outputs[i][0] -> asset i's series, length r.output_lens[i][0] */
tema_state_free(r.states[i]);
}
tulip_ffi_simd_result_free(r);
By options — same asset, 4 different periods in one call:
double o5[] = {5.0}, o10[] = {10.0}, o14[] = {14.0}, o20[] = {20.0};
const double *const simd_opts[4] = {o5, o10, o14, o20};
CSimdResult r = tema_simd_by_options(inputs, 10, simd_opts, 4, NULL, 0);
/* r.outputs[i] -> results for option set i (periods 5/10/14/20) */
for (uintptr_t i = 0; i < r.num_results; i++) tema_state_free(r.states[i]);
tulip_ffi_simd_result_free(r);
By assets — same period applied to 4 assets in parallel (lane counts 2/4/8/16):
a1 := []float64{81.59, 81.06, 82.87, 83.00, 83.61, 83.15, 82.84, 83.99, 84.55, 84.36}
// Reuse the same data for assets 2–4 in this example
a2, a3, a4 := a1, a1, a1
assets := [][indicators.TemaInputs][]float64{{a1}, {a2}, {a3}, {a4}}
sim, _ := indicators.Tema.SimdByAssets(assets, []float64{14.0}, nil)
for i, lanes := range sim.Results {
fmt.Printf("Asset %d: %v\n", i+1, lanes[0])
}
sim.Close() // frees every lane state, then the SIMD buffers
By options — same asset, 4 different periods in parallel:
By assets — same period applied to 4 assets in parallel (N must be 2, 4, 8, or 16):
import org.tuliprs.*;
import org.tuliprs.indicators.Tema;
double[] a1 = {81.59, 81.06, 82.87, 83.00, 83.61, 83.15, 82.84, 83.99, 84.55, 84.36};
double[] a2 = {72.10, 72.85, 73.40, 73.00, 74.20, 74.85, 75.10, 75.60, 76.00, 76.50};
double[] a3 = {55.30, 55.80, 56.10, 56.40, 56.90, 57.20, 57.50, 57.80, 58.10, 58.40};
double[] a4 = {100.1, 100.5, 101.0, 101.3, 101.8, 102.0, 102.5, 103.0, 103.3, 103.8};
// One entry per asset; each asset lists its INPUTS series.
double[][][] assets = {{a1}, {a2}, {a3}, {a4}};
try (SimdResult sim = Tema.simdByAssets(assets, new double[] {14.0})) {
for (int i = 0; i < sim.numResults(); i++) {
System.out.printf("Asset %d: %s%n", i + 1,
java.util.Arrays.toString(sim.toDoubleArray(i, 0)));
}
} // frees every lane state, then the SIMD buffers (contractual order)
By options — same asset, 4 different periods in parallel:
import org.tuliprs.*;
import org.tuliprs.indicators.Tema;
double[] close = {81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36};
try (SimdResult sim = Tema.simdByOptions(new double[][] {close},
new double[][] {{5}, {10}, {14}, {20}}, null)) {
for (int i = 0; i < sim.numResults(); i++) {
System.out.printf("Period set %d: %s%n", i + 1,
java.util.Arrays.toString(sim.toDoubleArray(i, 0)));
}
}
By assets — same period applied to N assets in parallel (must be 2, 4, 8, or 16):
import numpy as np
import tulip_rs
close = np.array([81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36], dtype=np.float64)
simd_inputs = [[close], [close + 5.0], [close - 5.0], [close * 1.02]]
outputs_list, states = tulip_rs.indicators.tema.simd_by_assets(simd_inputs, [14.0])
for i, out in enumerate(outputs_list):
print(f"Asset {i + 1}: {out[0]}")
By options — same asset, N different periods in parallel:
import numpy as np
import tulip_rs
close = np.array([81.59, 81.06, 82.87, 83.00, 83.61,
83.15, 82.84, 83.99, 84.55, 84.36], dtype=np.float64)
simd_options = [[5.0], [10.0], [14.0], [20.0]]
outputs_list, states = tulip_rs.indicators.tema.simd_by_options([close], simd_options)
for i, out in enumerate(outputs_list):
print(f"Period set {i + 1}: {out[0]}")
By assets — same period applied to 4 assets in parallel:
const simdInputs = [
[close.slice()],
[close.map(v => v * 1.1)],
[close.map(v => v * 0.9)],
[close.map(v => v * 1.02)],
];
const [results] = ti.tema.simdByAssets(simdInputs, [14]);
results.forEach((out, i) => console.log(`Asset ${i + 1}:`, out[0]));
By options — same asset, 4 different periods in parallel: