dexbot 1.6.9 → 1.6.11
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +34 -0
- package/README.md +1 -0
- package/analysis/README.md +50 -12
- package/analysis/bot_fitting/README.md +12 -0
- package/analysis/tradingview/README.md +15 -10
- package/analysis/trend_detection/DYNAMIC_WEIGHT_RESEARCH.md +10 -36
- package/analysis/trend_detection/README.md +31 -0
- package/claw/modules/chain_queries.ts +9 -4
- package/claw/modules/claw_bridge.ts +9 -9
- package/claw/package.json +1 -1
- package/claw/runtimes/openclaw-plugin/openclaw.plugin.json +1 -1
- package/claw/runtimes/openclaw-plugin/package.json +1 -1
- package/dist/analysis/account_resolver.js +13 -1
- package/dist/analysis/ama_fitting/optimizer_high_resolution.js +2 -1
- package/dist/analysis/analyze_dynamic_weight.js +0 -3
- package/dist/analysis/bot_fitting/backtest_ama_sweep.js +52 -8
- package/dist/analysis/bot_fitting/backtest_bot_fitting.js +68 -10
- package/dist/analysis/bot_fitting/backtest_lookback_drawdown.d.ts +21 -0
- package/dist/analysis/bot_fitting/backtest_lookback_drawdown.js +315 -0
- package/dist/analysis/bot_key_utils.js +17 -17
- package/dist/analysis/fills_cache.d.ts +64 -0
- package/dist/analysis/fills_cache.js +254 -0
- package/dist/analysis/fills_source.js +5 -2
- package/dist/analysis/pnl_report.d.ts +39 -0
- package/dist/analysis/pnl_report.js +325 -0
- package/dist/analysis/resolve_source.d.ts +0 -1
- package/dist/analysis/trade_profitability.d.ts +11 -2
- package/dist/analysis/trade_profitability.js +187 -61
- package/dist/analysis/tradingview/analyze_tradingview.d.ts +0 -3
- package/dist/analysis/tradingview/analyze_tradingview.js +6 -19
- package/dist/analysis/tradingview/grid_reset_config.d.ts +2 -0
- package/dist/analysis/tradingview/grid_reset_config.js +18 -1
- package/dist/analysis/tradingview/grid_reset_sim.d.ts +12 -1
- package/dist/analysis/tradingview/grid_reset_sim.js +60 -11
- package/dist/analysis/tradingview/tradingview_uplot_chart_generator.js +180 -33
- package/dist/analysis/trend_detection/backtest_ama_slope_huber.d.ts +102 -0
- package/dist/analysis/trend_detection/backtest_ama_slope_huber.js +725 -0
- package/dist/analysis/trend_detection/dynamic_weight_chart_generator.js +29 -110
- package/dist/analysis/trend_detection/huber_scale_variants.d.ts +34 -0
- package/dist/analysis/trend_detection/huber_scale_variants.js +176 -0
- package/dist/dexbot.js +19 -11
- package/dist/market_adapter/core/market_adapter_service.d.ts +15 -0
- package/dist/market_adapter/core/market_adapter_service.js +64 -5
- package/dist/market_adapter/core/strategies/ama.d.ts +1 -1
- package/dist/market_adapter/core/strategies/ama.js +5 -28
- package/dist/market_adapter/core/strategies/ama_slope_model.d.ts +2 -2
- package/dist/market_adapter/core/strategies/ama_slope_model.js +12 -10
- package/dist/market_adapter/core/strategies/dynamic_weight_series.d.ts +41 -14
- package/dist/market_adapter/core/strategies/dynamic_weight_series.js +138 -30
- package/dist/market_adapter/inputs/fetch_cex_synthetic_data.js +1 -2
- package/dist/market_adapter/log_format.js +1 -2
- package/dist/market_adapter/lp_chart_strategy_loader.js +2 -1
- package/dist/market_adapter/market_adapter.d.ts +0 -1
- package/dist/market_adapter/market_adapter.js +2 -27
- package/dist/market_adapter/utils/chain.d.ts +1 -1
- package/dist/market_adapter/utils/chain.js +8 -6
- package/dist/modules/account_bots.js +5 -4
- package/dist/modules/bitshares-native/resolvers.js +5 -0
- package/dist/modules/bot_settings.js +13 -1
- package/dist/modules/chain_orders.js +2 -1
- package/dist/modules/constants.d.ts +9 -1
- package/dist/modules/constants.js +55 -9
- package/dist/modules/credential_policy.js +4 -1
- package/dist/modules/credit_runtime.js +5 -1
- package/dist/modules/dexbot_maintenance_runtime.js +12 -0
- package/dist/modules/order/accounting.js +25 -0
- package/dist/modules/order/format.d.ts +35 -15
- package/dist/modules/order/format.js +62 -16
- package/dist/modules/order/manager.d.ts +1 -0
- package/dist/modules/order/manager.js +22 -1
- package/dist/modules/order/sync_engine.js +15 -1
- package/dist/modules/order/utils/order.d.ts +21 -1
- package/dist/modules/order/utils/order.js +80 -2
- package/dist/modules/order/utils/system.js +11 -1
- package/dist/modules/paths.d.ts +2 -0
- package/dist/modules/paths.js +2 -0
- package/dist/modules/utils/asset_symbols.d.ts +59 -0
- package/dist/modules/utils/asset_symbols.js +128 -0
- package/dist/modules/utils/time_range.d.ts +15 -0
- package/dist/modules/utils/time_range.js +18 -0
- package/dist/scripts/analyze-credit.js +10 -7
- package/dist/scripts/analyze-orders.js +1 -44
- package/dist/scripts/chart_command.js +15 -8
- package/dist/scripts/pnl.d.ts +3 -0
- package/dist/scripts/pnl.js +40 -0
- package/dist/scripts/verify-browser-bundle.js +8 -0
- package/dist/unlock.js +34 -0
- package/docs/DEXBOT_COMPARISON.md +3 -3
- package/docs/EVOLUTION.md +12 -10
- package/docs/FUND_MOVEMENT_AND_ACCOUNTING.md +1 -1
- package/docs/GRID_RECALCULATION.md +33 -17
- package/docs/README.md +2 -2
- package/docs/WORKFLOW.md +1 -0
- package/market_adapter/README.md +6 -28
- package/package.json +1 -1
- package/scripts/README.md +14 -0
|
@@ -2,7 +2,7 @@
|
|
|
2
2
|
import { DEFAULT_CONFIG, MARKET_ADAPTER } from '../../modules/constants.js';
|
|
3
3
|
import { getAmaWarmupBars } from '../../market_adapter/core/strategies/ama.js';
|
|
4
4
|
import { bilinearInterpolate } from '../../market_adapter/core/strategies/regime_interp.js';
|
|
5
|
-
import { computeDynamicWeightSeries,
|
|
5
|
+
import { computeDynamicWeightSeries, computeHuberWindowSlopePct, computeAmaSlopeClipThreshold, echoLatchSeries, roundToN } from '../../market_adapter/core/strategies/dynamic_weight_series.js';
|
|
6
6
|
import { buildKalmanVelocitySeries, computeAbsolutePercentileThreshold, resolveKalmanVelocitySmoothingConfig, smoothKalmanVelocityPoint, } from '../../market_adapter/core/signals/kalman_velocity_smoothing.js';
|
|
7
7
|
import { embedFunctionSources, escapeHtml, serializeJsonForScript, toEpochSeconds, UPLOT_SHARED_SCRIPT, uplotInlineTags } from '../chart_utils.js';
|
|
8
8
|
import { sharedChartCSS } from '../chart_css.js';
|
|
@@ -10,7 +10,7 @@ import { Y_AXIS_SIZE, makeCursorConfig, bindHoverStateFn, wireChartEvents, zoomR
|
|
|
10
10
|
// Browser-embedded shared functions — the interactive chart runs the exact same
|
|
11
11
|
// pure logic as the live market adapter service instead of a hand-copied copy.
|
|
12
12
|
const EMBEDDED_SHARED_FUNCS = embedFunctionSources([
|
|
13
|
-
|
|
13
|
+
computeHuberWindowSlopePct,
|
|
14
14
|
computeAmaSlopeClipThreshold,
|
|
15
15
|
bilinearInterpolate,
|
|
16
16
|
echoLatchSeries,
|
|
@@ -32,7 +32,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
32
32
|
const AMA_MS_MIN = 0.06, AMA_MS_MAX = 0.12;
|
|
33
33
|
const KAL_MS_MIN = 0.5, KAL_MS_MAX = 1.5;
|
|
34
34
|
const CLIP_PCT_MAX = 20;
|
|
35
|
-
const EMA_MIN = 0, EMA_MAX = 32;
|
|
36
35
|
const TH_MIN = 0, TH_MAX = 0.5;
|
|
37
36
|
const KF_MIN = 0, KF_MAX = 200;
|
|
38
37
|
const KFD_MIN = 1.0, KFD_MAX = 3.0;
|
|
@@ -46,7 +45,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
46
45
|
dispWeight: [0, 1],
|
|
47
46
|
neutralZonePct: [0, 1],
|
|
48
47
|
lookbackBars: [LB_MIN, LB_MAX],
|
|
49
|
-
amaEmaSpan: [EMA_MIN, EMA_MAX],
|
|
50
48
|
amaMaxSlopePct: [AMA_MS_MIN, AMA_MS_MAX],
|
|
51
49
|
kalmanMaxSlopePct: [KAL_MS_MIN, KAL_MS_MAX],
|
|
52
50
|
clipPct: [0, CLIP_PCT_MAX],
|
|
@@ -106,10 +104,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
106
104
|
const defaultKalmanDispThresholdMult = data.kalmanDispThresholdMult ?? MARKET_ADAPTER.DYNAMIC_WEIGHT_KALMAN_DISP_THRESHOLD_MULT_DEFAULT;
|
|
107
105
|
const defaultKalmanSmoothSpanPct = data.kalmanSmoothSpanPct ?? MARKET_ADAPTER.DYNAMIC_WEIGHT_KALMAN_SMOOTH_SPAN_PCT_DEFAULT;
|
|
108
106
|
const defaultSignalConfirmBars = data.signalConfirmBars ?? MARKET_ADAPTER.DYNAMIC_WEIGHT_SIGNAL_CONFIRM_BARS_DEFAULT;
|
|
109
|
-
// Research-only `ema` knob: EMA span in bars over the AMA input before the
|
|
110
|
-
// slope is taken. 0 (default) = off, keeps the raw AMA input.
|
|
111
|
-
const defaultAmaEmaSpanRaw = data.amaEmaSpan ?? ma.amaEmaSpan ?? 0;
|
|
112
|
-
const defaultAmaEmaSpan = Math.round(Math.min(EMA_MAX, Math.max(EMA_MIN, Number.isFinite(defaultAmaEmaSpanRaw) ? defaultAmaEmaSpanRaw : EMA_MIN)));
|
|
113
107
|
const regimeInitSlider = Math.round(defaultRegimeSensitivity * 100);
|
|
114
108
|
const interval = results.length > 1 ?
|
|
115
109
|
(new Date(results[1].timestamp).getTime() - new Date(results[0].timestamp).getTime()) / 1000 : 3600;
|
|
@@ -136,8 +130,7 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
136
130
|
const amaErPeriod = data.amaConfig?.erPeriod ?? MARKET_ADAPTER.AMAS[defaultAmaKey].erPeriod;
|
|
137
131
|
const amaSlowPeriod = data.amaConfig?.slowPeriod ?? MARKET_ADAPTER.AMAS[defaultAmaKey].slowPeriod;
|
|
138
132
|
const amaFastPeriod = data.amaConfig?.fastPeriod ?? MARKET_ADAPTER.AMAS[defaultAmaKey].fastPeriod;
|
|
139
|
-
const
|
|
140
|
-
const amaWarmupBars = getAmaWarmupBars(amaErPeriod, amaSlowPeriod, lookbackBars, amaFastPeriod, amaErSmoothPeriod);
|
|
133
|
+
const amaWarmupBars = getAmaWarmupBars(amaErPeriod, amaSlowPeriod, lookbackBars, amaFastPeriod);
|
|
141
134
|
const amaSlopeReadyBars = Math.ceil(amaErPeriod) + lookbackBars;
|
|
142
135
|
const amaSeedSmaLine = new Array(results.length).fill(null);
|
|
143
136
|
const amaSeedWindowEnd = Math.min(results.length - 1, Math.ceil(amaErPeriod));
|
|
@@ -175,16 +168,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
175
168
|
amaSeedSmaLine.push(null);
|
|
176
169
|
}
|
|
177
170
|
const realBarCount = results.length;
|
|
178
|
-
function maxAbsPct(arr) {
|
|
179
|
-
let m = 0;
|
|
180
|
-
for (let i = 0; i < realBarCount; i++) {
|
|
181
|
-
if (arr[i] != null && Math.abs(arr[i]) > m)
|
|
182
|
-
m = Math.abs(arr[i]);
|
|
183
|
-
}
|
|
184
|
-
return m || 4;
|
|
185
|
-
}
|
|
186
|
-
const amaPctMax = Math.ceil(maxAbsPct(amaSlopePct) * 1.15) || 5;
|
|
187
|
-
const kalPctMax = Math.ceil(Math.max(maxAbsPct(kalmanVelocityPct), maxAbsPct(kalmanDisplacementPct)) * 1.15) || 5;
|
|
188
171
|
function buildPercentiles(arr, startIndex = 0) {
|
|
189
172
|
const safeStartIndex = Math.max(0, Math.min(realBarCount, Math.ceil(startIndex)));
|
|
190
173
|
const sorted = [];
|
|
@@ -207,7 +190,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
207
190
|
}
|
|
208
191
|
const amaPercentiles = buildPercentiles(amaSlopePct, amaSlopeReadyBars);
|
|
209
192
|
const kalPercentiles = buildPercentiles(kalmanVelocityPct);
|
|
210
|
-
const maxDispPct = Math.ceil(maxAbsPct(kalmanDisplacementPct) * 1.15) || 5;
|
|
211
193
|
return `<!DOCTYPE html>
|
|
212
194
|
<html lang="en">
|
|
213
195
|
<head>
|
|
@@ -224,7 +206,7 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
224
206
|
.legend-item { gap: 3px; }
|
|
225
207
|
.legend-val { font-family: monospace; display: inline-block; text-align: right; min-width: 45px; }
|
|
226
208
|
#l-price, #l-ama3 { min-width: 62px; }
|
|
227
|
-
#l-ama-slope, #l-
|
|
209
|
+
#l-ama-slope, #l-kal-vel, #l-kal-disp { min-width: 54px; }
|
|
228
210
|
#l-signal, #l-signal-latched { min-width: 70px; text-align: left; }
|
|
229
211
|
#l-combined-raw, #l-combined-echo { min-width: 42px; }
|
|
230
212
|
#l-mult { min-width: 38px; }
|
|
@@ -269,9 +251,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
269
251
|
.ctrl.lb label { min-width: 20px; }
|
|
270
252
|
.ctrl.lb input[type="range"] { accent-color: #39d0d8; width: 80px; }
|
|
271
253
|
.ctrl.lb .val { color: #39d0d8; }
|
|
272
|
-
.ctrl.ema label { min-width: 20px; }
|
|
273
|
-
.ctrl.ema input[type="range"] { accent-color: #f0a000; width: 80px; }
|
|
274
|
-
.ctrl.ema .val { color: #f0a000; }
|
|
275
254
|
.ctrl.off input[type="range"] { accent-color: #3fb950; }
|
|
276
255
|
.ctrl.off .val { color: #3fb950; }
|
|
277
256
|
.ctrl.nz input[type="range"] { accent-color: #8b949e; }
|
|
@@ -317,7 +296,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
317
296
|
<div class="section-label">AMA SLOPE INPUT</div>
|
|
318
297
|
<div class="legend">
|
|
319
298
|
<div class="legend-item"><div class="dot" style="background:#f0a000;"></div>Slope‰: <span id="l-ama-slope" class="legend-val" style="font-weight:bold;">-</span></div>
|
|
320
|
-
<div class="legend-item"><div class="dot" style="background:#8b949e;"></div>Raw‰: <span id="l-ama-slope-raw" class="legend-val">-</span></div>
|
|
321
299
|
</div>
|
|
322
300
|
<div id="ama-chart"></div>
|
|
323
301
|
</div>
|
|
@@ -348,7 +326,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
348
326
|
<div class="group-sep"></div>
|
|
349
327
|
<div class="ctrl nz"><label for="nz-slider">nz%</label><input type="range" id="nz-slider" min="0" max="100" value="${Math.round(defaultNeutralZone * 100)}" title="Neutral Zone %"><span class="val" id="nz-value">${defaultNeutralZone.toFixed(2)}</span></div>
|
|
350
328
|
<div class="ctrl lb"><label for="lb-slider">lb</label><input type="range" id="lb-slider" min="0" max="1000" value="${lbInitSlider}" title="Lookback Bars (${LB_MIN}-${LB_MAX})"><span class="val" id="lb-value">${lookbackBars}</span></div>
|
|
351
|
-
<div class="ctrl ema"><label for="ema-slider">ema</label><input type="range" id="ema-slider" min="${EMA_MIN}" max="${EMA_MAX}" value="${defaultAmaEmaSpan}" title="AMA input EMA span in bars (${EMA_MIN} = off; noise std ~ 1/sqrt(span), lag ~ (span-1)/2 bars; span ~ lb is the balance point)"><span class="val" id="ema-value">${defaultAmaEmaSpan}</span></div>
|
|
352
329
|
<div class="ctrl ms-ama"><label for="ama-ms-slider">amaS%</label><input type="range" id="ama-ms-slider" min="0" max="1000" value="${amaMsInitSlider}" title="AMA Max Slope % per bar (${AMA_MS_MIN}-${AMA_MS_MAX})"><span class="val" id="ama-ms-value">${defaultAmaMaxSlopePct.toFixed(4)}</span></div>
|
|
353
330
|
<div class="ctrl clip"><label for="clip-slider">clip%</label><input type="range" id="clip-slider" min="0" max="${CLIP_PCT_MAX}" value="${Math.min(defaultClipPct, CLIP_PCT_MAX)}" title="Outlier Clip %"><span class="val" id="clip-value">${Math.min(defaultClipPct, CLIP_PCT_MAX)}%</span></div>
|
|
354
331
|
|
|
@@ -382,11 +359,15 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
382
359
|
</div>
|
|
383
360
|
</div>
|
|
384
361
|
|
|
385
|
-
<script id="payload" type="application/json">${serializeJsonForScript({ dates, prices, hurstArr, peArr, hurstSegments, peSegments, ama3Prices, amaSeedSmaLine,
|
|
362
|
+
<script id="payload" type="application/json">${serializeJsonForScript({ dates, prices, hurstArr, peArr, hurstSegments, peSegments, ama3Prices, amaSeedSmaLine, kalmanVelocityPctRaw, kalmanVelocityPct, kalmanDisplacementPct, kalmanIsReady, signals, alpha: defaultAlpha, gain: defaultGain, kalmanSmoothPct: defaultKalmanSmoothPct, kalmanDispScaleMult: defaultKalmanDispScaleMult, kalmanDispThresholdMult: defaultKalmanDispThresholdMult, kalmanSmoothSpanPct: defaultKalmanSmoothSpanPct, signalConfirmBars: defaultSignalConfirmBars, neutralZonePct: defaultNeutralZone, dispWeight: defaultDispWeight, amaMaxSlopePct: defaultAmaMaxSlopePct, kalmanMaxSlopePct: defaultKalmanMaxSlopePct, clipPct: defaultClipPct, minOutputThreshold: defaultMinOutputThreshold, outputClamp: defaultOutputClamp, regimeSensitivity: defaultRegimeSensitivity, absoluteThreshold: defaultAbsoluteThreshold, lookbackBars, amaErPeriod, amaSlowPeriod, amaWarmupBars, amaSlopeReadyBars, realBarCount, amaPercentiles, kalPercentiles, amaSlopeLogMin: AMA_MS_LOG_MIN_N, amaSlopeLogMax: AMA_MS_LOG_MAX_N, kalSlopeLogMin: KAL_MS_LOG_MIN_N, kalSlopeLogMax: KAL_MS_LOG_MAX_N, lbLogMin: LB_LOG_MIN_N, lbLogMax: LB_LOG_MAX_N, gainLogMin: GAIN_LOG_MIN_N, gainLogMax: GAIN_LOG_MAX_N, sliderRanges: SLIDER_RANGES, dispScaleMinPct: defaultDispScaleMinPct, weightMin: MARKET_ADAPTER.DYNAMIC_WEIGHT_MIN_WEIGHT, weightMax: MARKET_ADAPTER.DYNAMIC_WEIGHT_MAX_WEIGHT, marketAdapter: ma, amaWeightConfig, hNodes: [0.5 + MARKET_ADAPTER.HURST_ZONE_BAND, 0.5, 0.5 - MARKET_ADAPTER.HURST_ZONE_BAND], pNodes: MARKET_ADAPTER.PE_NODES, regimeTable: MARKET_ADAPTER.REGIME_TABLE })}</script>
|
|
386
363
|
|
|
387
364
|
<script>
|
|
388
365
|
const data = JSON.parse(document.getElementById('payload').textContent);
|
|
389
366
|
|
|
367
|
+
// Canonical Huber-slope parameters, injected from MARKET_ADAPTER so the
|
|
368
|
+
// embedded estimator runs the same values as the live adapter.
|
|
369
|
+
const AMA_SLOPE_HUBER = ${serializeJsonForScript(MARKET_ADAPTER.DYNAMIC_WEIGHT_AMA_HUBER)};
|
|
370
|
+
|
|
390
371
|
${EMBEDDED_SHARED_FUNCS}
|
|
391
372
|
|
|
392
373
|
const SYNC_KEY = "dyn-wt-res-v4";
|
|
@@ -424,8 +405,7 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
424
405
|
const kalMsSliderToVal = (pos) => Math.exp(KAL_MS_LOG_MIN + (pos / 1000) * (KAL_MS_LOG_MAX - KAL_MS_LOG_MIN));
|
|
425
406
|
|
|
426
407
|
let currentLookbackBars = data.lookbackBars ?? ${JSON.stringify(lookbackBars)};
|
|
427
|
-
|
|
428
|
-
let currentAmaSource = data.ama3Prices;
|
|
408
|
+
const currentAmaSource = data.ama3Prices;
|
|
429
409
|
const LB_LOG_MIN = data.lbLogMin;
|
|
430
410
|
const LB_LOG_MAX = data.lbLogMax;
|
|
431
411
|
const lbSliderToVal = (pos) => Math.round(Math.exp(LB_LOG_MIN + (pos / 1000) * (LB_LOG_MAX - LB_LOG_MIN)));
|
|
@@ -437,11 +417,10 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
437
417
|
const gainValToSlider = (val) => Math.round((Math.log(Math.max(Math.exp(GAIN_LOG_MIN), val)) - GAIN_LOG_MIN) / (GAIN_LOG_MAX - GAIN_LOG_MIN) * 1000);
|
|
438
418
|
|
|
439
419
|
let currentClipPct = data.clipPct ?? ${JSON.stringify(defaultClipPct)};
|
|
440
|
-
//
|
|
441
|
-
//
|
|
442
|
-
//
|
|
443
|
-
|
|
444
|
-
let currentAmaClipThreshold = currentClipPct === 0 ? maxAmaSlope : Infinity;
|
|
420
|
+
// The AMA clip threshold always comes from dynamicClipThreshold, which
|
|
421
|
+
// recalcInputs() recomputes via the canonical
|
|
422
|
+
// computeAmaSlopeClipThreshold (Infinity when clipping is disabled,
|
|
423
|
+
// matching the live adapter). No separate clipPct==0 branch is needed.
|
|
445
424
|
let currentKalClipThreshold = currentClipPct === 0 ? Infinity : data.kalPercentiles[100 - currentClipPct];
|
|
446
425
|
|
|
447
426
|
// Regime table and axis nodes from payload — sourced from MARKET_ADAPTER in constants.ts
|
|
@@ -455,7 +434,10 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
455
434
|
*/
|
|
456
435
|
function getRegimeMultiplier(H, PE) {
|
|
457
436
|
if (H == null || PE == null) return 1.0;
|
|
458
|
-
|
|
437
|
+
// NOTE: peNodes is the canonical option name read by
|
|
438
|
+
// bilinearInterpolate; a pNodes key silently falls back to the
|
|
439
|
+
// inline defaults.
|
|
440
|
+
return bilinearInterpolate(H, PE, REGIME_TABLE, { hNodes: H_NODES, peNodes: P_NODES });
|
|
459
441
|
}
|
|
460
442
|
|
|
461
443
|
let currentRegimeSensitivity = data.regimeSensitivity ?? ma.regimeSensitivity;
|
|
@@ -464,7 +446,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
464
446
|
const currentLatchedSignals = new Array(data.dates.length).fill(null);
|
|
465
447
|
const dynamicAmaOff = new Array(data.dates.length).fill(null);
|
|
466
448
|
const dynamicAmaSlopePct = new Array(data.dates.length).fill(null);
|
|
467
|
-
const dynamicAmaSlopePctRaw = new Array(data.dates.length).fill(null);
|
|
468
449
|
const dynamicKalOff = new Array(data.dates.length).fill(null);
|
|
469
450
|
const combinedOff = new Array(data.dates.length).fill(null);
|
|
470
451
|
const combinedSell = new Array(data.dates.length).fill(null);
|
|
@@ -579,40 +560,9 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
579
560
|
}
|
|
580
561
|
}
|
|
581
562
|
|
|
582
|
-
/**
|
|
583
|
-
* Research-only ema knob — EMA low-pass over the AMA input before the
|
|
584
|
-
* slope is taken (EMA-then-difference == difference-then-EMA, both are
|
|
585
|
-
* linear filters). span bars, alpha = 2/(span+1): noise std shrinks ~
|
|
586
|
-
* 1/sqrt(span), lag grows ~ (span-1)/2 bars. Seeded with an SMA over the
|
|
587
|
-
* first 'span' finite values (mirrors the AMA seed SMA) so the transient
|
|
588
|
-
* is gone before any slope bar can go ready. Nulls (warmup / trailing
|
|
589
|
-
* padding) stay null so no fabricated values reach the clip-percentile
|
|
590
|
-
* pool or the offset pipeline. span <= 0 returns the input untouched.
|
|
591
|
-
*/
|
|
592
|
-
function buildEmaSeries(values, span) {
|
|
593
|
-
if (!(span > 0) || !Array.isArray(values)) return values;
|
|
594
|
-
const n = values.length;
|
|
595
|
-
const out = new Array(n);
|
|
596
|
-
const alpha = 2 / (span + 1);
|
|
597
|
-
let seedSum = 0, seedCount = 0;
|
|
598
|
-
for (let i = 0; i < n && seedCount < span; i++) {
|
|
599
|
-
const v = values[i];
|
|
600
|
-
if (v == null || !Number.isFinite(v)) continue;
|
|
601
|
-
seedSum += v; seedCount++;
|
|
602
|
-
}
|
|
603
|
-
let prev = seedCount > 0 ? seedSum / seedCount : null;
|
|
604
|
-
for (let i = 0; i < n; i++) {
|
|
605
|
-
const v = values[i];
|
|
606
|
-
if (v == null || !Number.isFinite(v)) { out[i] = null; continue; }
|
|
607
|
-
prev = prev == null ? v : alpha * v + (1 - alpha) * prev;
|
|
608
|
-
out[i] = prev;
|
|
609
|
-
}
|
|
610
|
-
return out;
|
|
611
|
-
}
|
|
612
|
-
|
|
613
563
|
function computeSlopeAtIndex(idx, lb, values) {
|
|
614
|
-
// Canonical AMA slope % (
|
|
615
|
-
const sp =
|
|
564
|
+
// Canonical AMA slope % (computeHuberWindowSlopePct injected above).
|
|
565
|
+
const sp = computeHuberWindowSlopePct(values, idx, lb);
|
|
616
566
|
return sp == null ? 0 : sp;
|
|
617
567
|
}
|
|
618
568
|
|
|
@@ -620,7 +570,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
620
570
|
recalcKalmanVelocity();
|
|
621
571
|
recalcKalmanClipThreshold();
|
|
622
572
|
recalcLatchedSignals();
|
|
623
|
-
currentAmaSource = buildEmaSeries(data.ama3Prices, currentAmaEmaSpan);
|
|
624
573
|
const lb = currentLookbackBars;
|
|
625
574
|
const amaErWarmup = Math.max(0, Number.isFinite(data.amaErPeriod) ? Math.ceil(data.amaErPeriod) : ${JSON.stringify(MARKET_ADAPTER.AMAS[MARKET_ADAPTER.DEFAULT_AMA_KEY].erPeriod)});
|
|
626
575
|
const amaReadyBar = Math.max(lb, amaErWarmup + lb);
|
|
@@ -629,20 +578,13 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
629
578
|
// implementation (computeAmaSlopeClipThreshold injected above),
|
|
630
579
|
// shared with the live market adapter service. Trailing padded
|
|
631
580
|
// nulls beyond realBarCount are skipped by its finiteness guards.
|
|
632
|
-
// Clip pool follows the (possibly EMA-filtered) AMA input so the
|
|
633
|
-
// percentile threshold matches the series the offsets derive from.
|
|
634
581
|
dynamicClipThreshold = computeAmaSlopeClipThreshold(currentAmaSource, amaErWarmup, lb, currentClipPct);
|
|
635
582
|
|
|
636
583
|
for (let i = 0; i < data.realBarCount; i++) {
|
|
637
|
-
|
|
638
|
-
// recalcWeights): the raw reference line from the unfiltered AMA
|
|
639
|
-
// input, the plotted series from the EMA-filtered input.
|
|
640
|
-
dynamicAmaSlopePctRaw[i] = i < amaReadyBar ? null : computeSlopeAtIndex(i, lb, data.ama3Prices);
|
|
641
|
-
dynamicAmaSlopePct[i] = i < amaReadyBar ? null : computeSlopeAtIndex(i, lb, currentAmaSource);
|
|
584
|
+
dynamicAmaSlopePct[i] = i < amaReadyBar ? null : computeSlopeAtIndex(i, lb, currentAmaSource);
|
|
642
585
|
}
|
|
643
586
|
for (let i = data.realBarCount; i < data.dates.length; i++) {
|
|
644
587
|
dynamicAmaSlopePct[i] = null;
|
|
645
|
-
dynamicAmaSlopePctRaw[i] = null;
|
|
646
588
|
}
|
|
647
589
|
}
|
|
648
590
|
|
|
@@ -650,7 +592,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
650
592
|
// Normalize each channel by the configured clamp so alpha remains a pure ratio knob.
|
|
651
593
|
// Gain then scales the blended output linearly after the dead-band decision.
|
|
652
594
|
currentOutputAxisMax = Math.max(0.5, OUTPUT_CLAMP);
|
|
653
|
-
const channelNorm = Math.max(Math.abs(OUTPUT_CLAMP), 1e-9);
|
|
654
595
|
const outputThreshold = currentMinOutputThreshold;
|
|
655
596
|
const n = data.realBarCount;
|
|
656
597
|
|
|
@@ -658,11 +599,11 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
658
599
|
for (let i = 0; i < data.dates.length; i++) {
|
|
659
600
|
if (i >= n) { currentMults[i] = null; continue; }
|
|
660
601
|
const baseMult = getRegimeMultiplier(data.hurstArr[i], data.peArr[i]);
|
|
661
|
-
//
|
|
662
|
-
|
|
663
|
-
//
|
|
664
|
-
|
|
665
|
-
currentMults[i] = Math.abs(rawMult - 1.0) >= ABSOLUTE_THRESHOLD ?
|
|
602
|
+
// Match the live regime gate exactly: apply the sensitivity
|
|
603
|
+
// exponent, clamp to 1.0 (dampen-only), round the per-bar series
|
|
604
|
+
// to 3 decimals, then apply the absolute-threshold dead-band.
|
|
605
|
+
const rawMult = roundToN(Math.min(Math.pow(baseMult, currentRegimeSensitivity), 1.0), 1000);
|
|
606
|
+
currentMults[i] = Math.abs(rawMult - 1.0) >= ABSOLUTE_THRESHOLD ? rawMult : 1.0;
|
|
666
607
|
}
|
|
667
608
|
|
|
668
609
|
// Per-bar offset pipeline — canonical implementation (computeDynamicWeightSeries
|
|
@@ -675,7 +616,7 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
675
616
|
regimeMultipliers: currentMults,
|
|
676
617
|
lookbackBars: currentLookbackBars,
|
|
677
618
|
amaErPeriod: data.amaErPeriod,
|
|
678
|
-
amaClipThreshold:
|
|
619
|
+
amaClipThreshold: dynamicClipThreshold,
|
|
679
620
|
kalClipThreshold: currentKalClipThreshold,
|
|
680
621
|
neutralZonePct: currentNz,
|
|
681
622
|
amaMaxSlopePct: currentAmaMaxSlopePct,
|
|
@@ -752,7 +693,7 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
752
693
|
max: Number.isFinite(xs.max) ? xs.max : xMax,
|
|
753
694
|
} : null;
|
|
754
695
|
|
|
755
|
-
amaChart.setData([data.dates,
|
|
696
|
+
amaChart.setData([data.dates, dynamicAmaSlopePct], false);
|
|
756
697
|
kalmanChart.setData([data.dates, currentKalmanVelocityPct, data.kalmanDisplacementPct], false);
|
|
757
698
|
outputChart.setData([data.dates, combinedOff, echoCombinedOff], false);
|
|
758
699
|
|
|
@@ -787,11 +728,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
787
728
|
if (sp == null) { spEl.textContent = '-'; spEl.style.color = '#8b949e'; }
|
|
788
729
|
else { spEl.textContent = (sp >= 0 ? '+' : '') + (sp * 10).toFixed(3) + '‰'; spEl.style.color = sp > 0.01 ? '#2ea043' : sp < -0.01 ? '#f85149' : '#8b949e'; }
|
|
789
730
|
|
|
790
|
-
const spRaw = dynamicAmaSlopePctRaw[idx];
|
|
791
|
-
const spRawEl = document.getElementById('l-ama-slope-raw');
|
|
792
|
-
if (spRaw == null) { spRawEl.textContent = '-'; spRawEl.style.color = '#8b949e'; }
|
|
793
|
-
else { spRawEl.textContent = (spRaw >= 0 ? '+' : '') + (spRaw * 10).toFixed(3) + '‰'; spRawEl.style.color = '#8b949e'; }
|
|
794
|
-
|
|
795
731
|
const vp = currentKalmanVelocityPct[idx];
|
|
796
732
|
const vpEl = document.getElementById('l-kal-vel');
|
|
797
733
|
if (vp == null) { vpEl.textContent = '-'; vpEl.style.color = '#8b949e'; }
|
|
@@ -1015,7 +951,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
1015
951
|
document.getElementById('clip-slider').value = currentClipPct;
|
|
1016
952
|
document.getElementById('nz-slider').value = Math.round(currentNz * 100);
|
|
1017
953
|
document.getElementById('lb-slider').value = lbValToSlider(currentLookbackBars);
|
|
1018
|
-
document.getElementById('ema-slider').value = currentAmaEmaSpan;
|
|
1019
954
|
|
|
1020
955
|
const priceEl = document.getElementById('price-panel');
|
|
1021
956
|
const amaEl = document.getElementById('ama-panel');
|
|
@@ -1051,7 +986,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
1051
986
|
scales: { x: { time: true }, p: { auto: true } },
|
|
1052
987
|
series: [
|
|
1053
988
|
{ label: 'Time' },
|
|
1054
|
-
{ label: 'Raw‰', stroke: '#8b949e', width: 1, dash: [3, 3], scale: 'p', points: { show: false } },
|
|
1055
989
|
{ label: 'Slope‰', stroke: '#f0a000', width: 2, scale: 'p', points: { show: false } },
|
|
1056
990
|
],
|
|
1057
991
|
axes: [
|
|
@@ -1061,7 +995,7 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
1061
995
|
],
|
|
1062
996
|
cursor: cursorCfg,
|
|
1063
997
|
hooks: { draw: [makePctFillHook(dynamicAmaSlopePct, 'p', 'rgba(46,160,67,0.20)', 'rgba(248,81,73,0.20)'), makeSignalBgHook('p')] }
|
|
1064
|
-
}, [data.dates,
|
|
998
|
+
}, [data.dates, dynamicAmaSlopePct], document.getElementById('ama-chart'));
|
|
1065
999
|
|
|
1066
1000
|
kalmanChart = new uPlot({
|
|
1067
1001
|
width: kalmanEl.offsetWidth, height: kalmanEl.offsetHeight,
|
|
@@ -1198,7 +1132,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
1198
1132
|
document.getElementById('clip-slider').addEventListener('input', (e) => {
|
|
1199
1133
|
currentClipPct = parseInt(e.target.value, 10);
|
|
1200
1134
|
if (currentClipPct === 0) {
|
|
1201
|
-
currentAmaClipThreshold = data.amaPercentiles[data.amaPercentiles.length - 1];
|
|
1202
1135
|
currentKalClipThreshold = Infinity;
|
|
1203
1136
|
} else {
|
|
1204
1137
|
// AMA clip threshold for clipPct>0 comes from the canonical
|
|
@@ -1223,12 +1156,6 @@ function generateHTML(data, title = 'Dynamic Weight Research') {
|
|
|
1223
1156
|
onSliderChange();
|
|
1224
1157
|
});
|
|
1225
1158
|
|
|
1226
|
-
document.getElementById('ema-slider').addEventListener('input', (e) => {
|
|
1227
|
-
currentAmaEmaSpan = parseInt(e.target.value, 10);
|
|
1228
|
-
document.getElementById('ema-value').textContent = currentAmaEmaSpan;
|
|
1229
|
-
onSliderChange();
|
|
1230
|
-
});
|
|
1231
|
-
|
|
1232
1159
|
document.getElementById('regime-slider').addEventListener('input', (e) => {
|
|
1233
1160
|
currentRegimeSensitivity = parseInt(e.target.value, 10) / 100;
|
|
1234
1161
|
document.getElementById('regime-value').textContent = currentRegimeSensitivity.toFixed(2);
|
|
@@ -1303,7 +1230,6 @@ function applyParams(p, btn) {
|
|
|
1303
1230
|
document.getElementById('clip-slider').value = currentClipPct;
|
|
1304
1231
|
document.getElementById('clip-value').textContent = currentClipPct + '%';
|
|
1305
1232
|
if (currentClipPct === 0) {
|
|
1306
|
-
currentAmaClipThreshold = data.amaPercentiles[data.amaPercentiles.length - 1];
|
|
1307
1233
|
currentKalClipThreshold = Infinity;
|
|
1308
1234
|
} else {
|
|
1309
1235
|
recalcKalmanClipThreshold();
|
|
@@ -1324,11 +1250,6 @@ function applyParams(p, btn) {
|
|
|
1324
1250
|
document.getElementById('lb-slider').value = lbValToSlider(currentLookbackBars);
|
|
1325
1251
|
document.getElementById('lb-value').textContent = currentLookbackBars;
|
|
1326
1252
|
}
|
|
1327
|
-
if (p.amaEmaSpan != null) {
|
|
1328
|
-
currentAmaEmaSpan = Math.max(SR.amaEmaSpan[0], Math.min(SR.amaEmaSpan[1], Math.round(p.amaEmaSpan)));
|
|
1329
|
-
document.getElementById('ema-slider').value = currentAmaEmaSpan;
|
|
1330
|
-
document.getElementById('ema-value').textContent = currentAmaEmaSpan;
|
|
1331
|
-
}
|
|
1332
1253
|
recalcInputs();
|
|
1333
1254
|
recalcWeights();
|
|
1334
1255
|
refreshChartsPreservingZoom();
|
|
@@ -1355,7 +1276,6 @@ function applyParams(p, btn) {
|
|
|
1355
1276
|
regimeSensitivity: 'regime',
|
|
1356
1277
|
dispWeight: 'dw',
|
|
1357
1278
|
lookbackBars: 'lb',
|
|
1358
|
-
amaEmaSpan: 'ema',
|
|
1359
1279
|
};
|
|
1360
1280
|
document.getElementById('paste-confirm-vals').innerHTML = Object.entries(labels)
|
|
1361
1281
|
.filter(([k]) => p[k] != null)
|
|
@@ -1409,7 +1329,6 @@ function applyParams(p, btn) {
|
|
|
1409
1329
|
regimeSensitivity: +currentRegimeSensitivity.toFixed(2),
|
|
1410
1330
|
dispWeight: +currentDw.toFixed(2),
|
|
1411
1331
|
lookbackBars: currentLookbackBars,
|
|
1412
|
-
amaEmaSpan: currentAmaEmaSpan,
|
|
1413
1332
|
};
|
|
1414
1333
|
const json = JSON.stringify(params, null, 2);
|
|
1415
1334
|
localStorage.setItem(LS_KEY, json);
|
|
@@ -0,0 +1,34 @@
|
|
|
1
|
+
export type HuberScaleMode = 'none' | 'df' | 'mscale';
|
|
2
|
+
/** One-bar diagnostic snapshot returned alongside the slope. */
|
|
3
|
+
export interface HuberDiagnostics {
|
|
4
|
+
slopePct: number;
|
|
5
|
+
scale: number;
|
|
6
|
+
outlierFraction: number;
|
|
7
|
+
}
|
|
8
|
+
/** Accumulated stats attached to an estimator built by {@link createHuberEstimator}. */
|
|
9
|
+
export interface HuberEstimatorStats {
|
|
10
|
+
calls: number;
|
|
11
|
+
outlierFractionSum: number;
|
|
12
|
+
scaleSum: number;
|
|
13
|
+
}
|
|
14
|
+
/** Estimator function shape accepted by simulateGridResetSeries (cfg.slopeEstimator). */
|
|
15
|
+
export type HuberEstimator = ((amaValues: any, index: number, lookbackBars: any) => number | null) & {
|
|
16
|
+
stats: HuberEstimatorStats;
|
|
17
|
+
};
|
|
18
|
+
/**
|
|
19
|
+
* delta_C = E[rho_C(Z)] for Z ~ N(0,1), the proposal-2 consistency constant:
|
|
20
|
+
* delta_C = 2*Phi(c) - 1 + 2*c*phi(c) - 2*c^2*(1 - Phi(c)).
|
|
21
|
+
*/
|
|
22
|
+
export declare function huberConsistencyConstant(c: number): number;
|
|
23
|
+
/**
|
|
24
|
+
* Core fit. Reproduces `computeHuberWindowSlopePct` byte-for-byte for
|
|
25
|
+
* `mode='none'`; the other modes change only the scale estimate fed to the
|
|
26
|
+
* Huber weight `min(1, C*s/|r|)`.
|
|
27
|
+
*/
|
|
28
|
+
export declare function huberSlopeCore(amaValues: any, index: number, lookbackBars: any, mode?: HuberScaleMode, hub?: any): HuberDiagnostics | null;
|
|
29
|
+
/**
|
|
30
|
+
* Build the `(series, index, lookbackBars)` estimator the reset replay expects,
|
|
31
|
+
* accumulating the mean outlier fraction and scale across calls in `.stats`.
|
|
32
|
+
*/
|
|
33
|
+
export declare function createHuberEstimator(mode?: HuberScaleMode, hub?: any): HuberEstimator;
|
|
34
|
+
//# sourceMappingURL=huber_scale_variants.d.ts.map
|
|
@@ -0,0 +1,176 @@
|
|
|
1
|
+
'use strict';
|
|
2
|
+
/**
|
|
3
|
+
* RESEARCH-ONLY scale variants for the canonical Huber AMA-slope estimator.
|
|
4
|
+
*
|
|
5
|
+
* `computeHuberWindowSlopePct` (market_adapter/core/strategies/dynamic_weight_series.ts)
|
|
6
|
+
* estimates the robust scale as a PLUG-IN `1.4826 * MAD` of the residuals of the
|
|
7
|
+
* current fit, recomputed each IRLS pass. Those residuals are shrunk relative to
|
|
8
|
+
* the true errors (n - 2 degrees of freedom), so the plug-in scale is biased low
|
|
9
|
+
* — measured ~0.93x sigma at a 16-bar window, ~0.86x at 8 bars — which lowers the
|
|
10
|
+
* effective Huber constant below the nominal 1.345 and makes short windows more
|
|
11
|
+
* conservative than intended.
|
|
12
|
+
*
|
|
13
|
+
* scikit-learn's `HuberRegressor` avoids this by optimizing sigma JOINTLY with the
|
|
14
|
+
* coefficients (sklearn/linear_model/_huber.py); its reference is Huber &
|
|
15
|
+
* Ronchetti, "Concomitant scale estimates", p. 172. This module reproduces the
|
|
16
|
+
* baseline exactly and offers two corrected scale estimates so the effect can be
|
|
17
|
+
* measured without touching production:
|
|
18
|
+
*
|
|
19
|
+
* 'none' — exact production behaviour (plug-in 1.4826 * MAD).
|
|
20
|
+
* 'df' — plug-in MAD inflated by the finite-sample factor sqrt(n / (n - 2)).
|
|
21
|
+
* 'mscale' — Huber proposal-2 M-scale: solve mean(rho_C(r_i / s)) = delta_C,
|
|
22
|
+
* where delta_C = E[rho_C(Z)] under N(0, 1), iterated with the fit.
|
|
23
|
+
*
|
|
24
|
+
* Also returns an outlier fraction (share of |residual| > C * s), the sklearn
|
|
25
|
+
* `outliers_` analogue, for observability.
|
|
26
|
+
*
|
|
27
|
+
* Not imported by production code and not embedded in any chart — a research
|
|
28
|
+
* harness for the scale-bias hypothesis only.
|
|
29
|
+
*/
|
|
30
|
+
import { MARKET_ADAPTER } from '../../modules/constants.js';
|
|
31
|
+
// Abramowitz & Stegun 7.1.26 approximation — max abs error ~1.5e-7, ample for a
|
|
32
|
+
// consistency constant that only needs ~1e-6 relative accuracy.
|
|
33
|
+
function erf(x) {
|
|
34
|
+
const sign = x < 0 ? -1 : 1;
|
|
35
|
+
const ax = Math.abs(x);
|
|
36
|
+
const t = 1 / (1 + 0.3275911 * ax);
|
|
37
|
+
const poly = ((((1.061405429 * t - 1.453152027) * t + 1.421413741) * t - 0.284496736) * t + 0.254829592) * t;
|
|
38
|
+
return sign * (1 - poly * Math.exp(-ax * ax));
|
|
39
|
+
}
|
|
40
|
+
function standardNormalCdf(x) {
|
|
41
|
+
return 0.5 * (1 + erf(x / Math.SQRT2));
|
|
42
|
+
}
|
|
43
|
+
function standardNormalPdf(x) {
|
|
44
|
+
return Math.exp(-0.5 * x * x) / Math.sqrt(2 * Math.PI);
|
|
45
|
+
}
|
|
46
|
+
/** Huber rho: z^2 inside the band, linear beyond. */
|
|
47
|
+
function huberRho(z, c) {
|
|
48
|
+
const a = Math.abs(z);
|
|
49
|
+
return a <= c ? z * z : 2 * c * a - c * c;
|
|
50
|
+
}
|
|
51
|
+
/**
|
|
52
|
+
* delta_C = E[rho_C(Z)] for Z ~ N(0,1), the proposal-2 consistency constant:
|
|
53
|
+
* delta_C = 2*Phi(c) - 1 + 2*c*phi(c) - 2*c^2*(1 - Phi(c)).
|
|
54
|
+
*/
|
|
55
|
+
export function huberConsistencyConstant(c) {
|
|
56
|
+
const Phi = standardNormalCdf(c);
|
|
57
|
+
const phi = standardNormalPdf(c);
|
|
58
|
+
return 2 * Phi - 1 + 2 * c * phi - 2 * c * c * (1 - Phi);
|
|
59
|
+
}
|
|
60
|
+
/**
|
|
61
|
+
* Proposal-2 M-scale: the s satisfying mean(rho_C(r_i / s)) = delta_C.
|
|
62
|
+
* Fixed-point iteration s <- s * sqrt(mean(rho_C(r / s)) / delta_C), seeded from
|
|
63
|
+
* the plug-in MAD so it starts near the answer.
|
|
64
|
+
*/
|
|
65
|
+
function solveMScale(resid, c, seed, floor) {
|
|
66
|
+
const delta = huberConsistencyConstant(c);
|
|
67
|
+
let s = Math.max(seed, floor);
|
|
68
|
+
for (let it = 0; it < 60; it++) {
|
|
69
|
+
let acc = 0;
|
|
70
|
+
for (let i = 0; i < resid.length; i++)
|
|
71
|
+
acc += huberRho(resid[i] / s, c);
|
|
72
|
+
const ratio = acc / (resid.length * delta);
|
|
73
|
+
const next = s * Math.sqrt(Math.max(ratio, 1e-12));
|
|
74
|
+
if (!Number.isFinite(next))
|
|
75
|
+
return s;
|
|
76
|
+
if (Math.abs(next - s) <= 1e-13 * s)
|
|
77
|
+
return next;
|
|
78
|
+
s = next;
|
|
79
|
+
}
|
|
80
|
+
return s;
|
|
81
|
+
}
|
|
82
|
+
/**
|
|
83
|
+
* Core fit. Reproduces `computeHuberWindowSlopePct` byte-for-byte for
|
|
84
|
+
* `mode='none'`; the other modes change only the scale estimate fed to the
|
|
85
|
+
* Huber weight `min(1, C*s/|r|)`.
|
|
86
|
+
*/
|
|
87
|
+
export function huberSlopeCore(amaValues, index, lookbackBars, mode = 'none', hub = MARKET_ADAPTER.DYNAMIC_WEIGHT_AMA_HUBER) {
|
|
88
|
+
const bars = Number.isFinite(lookbackBars) && Number(lookbackBars) > 0
|
|
89
|
+
? Math.ceil(Number(lookbackBars))
|
|
90
|
+
: 0;
|
|
91
|
+
if (!Array.isArray(amaValues) || bars < 1)
|
|
92
|
+
return null;
|
|
93
|
+
if (!Number.isFinite(index) || index < bars || index >= amaValues.length)
|
|
94
|
+
return null;
|
|
95
|
+
const y = [];
|
|
96
|
+
for (let k = index - bars; k <= index; k++) {
|
|
97
|
+
const v = Number(amaValues[k]);
|
|
98
|
+
if (!Number.isFinite(v) || v <= 0)
|
|
99
|
+
return null;
|
|
100
|
+
y.push(Math.log(v));
|
|
101
|
+
}
|
|
102
|
+
const n = y.length;
|
|
103
|
+
const c = Number.isFinite(hub?.C) ? hub.C : 1.345;
|
|
104
|
+
const iterations = Number.isFinite(hub?.ITERATIONS) ? hub.ITERATIONS : 5;
|
|
105
|
+
const scaleFloor = Number.isFinite(hub?.SCALE_FLOOR) ? hub.SCALE_FLOOR : 1e-6;
|
|
106
|
+
const zeroEpsilon = Number.isFinite(hub?.ZERO_EPSILON) ? hub.ZERO_EPSILON : 1e-9;
|
|
107
|
+
const xMean = (n - 1) / 2;
|
|
108
|
+
const wls = (w) => {
|
|
109
|
+
let sw = 0, swx = 0, swy = 0, swxx = 0, swxy = 0;
|
|
110
|
+
for (let i = 0; i < n; i++) {
|
|
111
|
+
const xi = i - xMean;
|
|
112
|
+
sw += w[i];
|
|
113
|
+
swx += w[i] * xi;
|
|
114
|
+
swy += w[i] * y[i];
|
|
115
|
+
swxx += w[i] * xi * xi;
|
|
116
|
+
swxy += w[i] * xi * y[i];
|
|
117
|
+
}
|
|
118
|
+
const den = sw * swxx - swx * swx;
|
|
119
|
+
const b = den === 0 ? 0 : (sw * swxy - swx * swy) / den;
|
|
120
|
+
const a = sw === 0 ? 0 : (swy - b * swx) / sw;
|
|
121
|
+
return { a, b };
|
|
122
|
+
};
|
|
123
|
+
let fit = wls(new Array(n).fill(1));
|
|
124
|
+
let scale = scaleFloor;
|
|
125
|
+
for (let iter = 0; iter < iterations; iter++) {
|
|
126
|
+
const resid = y.map((v, i) => v - (fit.a + fit.b * (i - xMean)));
|
|
127
|
+
const abs = resid.map(Math.abs).sort((p, q) => p - q);
|
|
128
|
+
const mid = abs.length >> 1;
|
|
129
|
+
const mad = abs.length % 2 === 1 ? abs[mid] : (abs[mid - 1] + abs[mid]) / 2;
|
|
130
|
+
let s = 1.4826 * mad;
|
|
131
|
+
if (mode === 'df')
|
|
132
|
+
s = s * Math.sqrt(n / Math.max(1, n - 2));
|
|
133
|
+
else if (mode === 'mscale')
|
|
134
|
+
s = solveMScale(resid, c, s, scaleFloor);
|
|
135
|
+
s = Math.max(s, scaleFloor);
|
|
136
|
+
scale = s;
|
|
137
|
+
const w = resid.map((r) => Math.min(1, (c * s) / Math.max(Math.abs(r), 1e-12)));
|
|
138
|
+
const next = wls(w);
|
|
139
|
+
const moved = Math.abs(next.b - fit.b);
|
|
140
|
+
fit = next;
|
|
141
|
+
if (moved < 1e-12)
|
|
142
|
+
break;
|
|
143
|
+
}
|
|
144
|
+
// Outlier fraction at the final fit (sklearn outliers_ analogue).
|
|
145
|
+
let outliers = 0;
|
|
146
|
+
for (let i = 0; i < n; i++) {
|
|
147
|
+
const r = y[i] - (fit.a + fit.b * (i - xMean));
|
|
148
|
+
if (Math.abs(r) > c * scale)
|
|
149
|
+
outliers++;
|
|
150
|
+
}
|
|
151
|
+
let slope = fit.b * 100;
|
|
152
|
+
if (Number.isFinite(slope) && Math.abs(slope) < zeroEpsilon)
|
|
153
|
+
slope = 0;
|
|
154
|
+
if (!Number.isFinite(slope))
|
|
155
|
+
return null;
|
|
156
|
+
return { slopePct: slope, scale, outlierFraction: outliers / n };
|
|
157
|
+
}
|
|
158
|
+
/**
|
|
159
|
+
* Build the `(series, index, lookbackBars)` estimator the reset replay expects,
|
|
160
|
+
* accumulating the mean outlier fraction and scale across calls in `.stats`.
|
|
161
|
+
*/
|
|
162
|
+
export function createHuberEstimator(mode = 'none', hub = MARKET_ADAPTER.DYNAMIC_WEIGHT_AMA_HUBER) {
|
|
163
|
+
const stats = { calls: 0, outlierFractionSum: 0, scaleSum: 0 };
|
|
164
|
+
const fn = ((amaValues, index, lookbackBars) => {
|
|
165
|
+
const d = huberSlopeCore(amaValues, index, lookbackBars, mode, hub);
|
|
166
|
+
if (d == null)
|
|
167
|
+
return null;
|
|
168
|
+
stats.calls++;
|
|
169
|
+
stats.outlierFractionSum += d.outlierFraction;
|
|
170
|
+
stats.scaleSum += d.scale;
|
|
171
|
+
return d.slopePct;
|
|
172
|
+
});
|
|
173
|
+
fn.stats = stats;
|
|
174
|
+
return fn;
|
|
175
|
+
}
|
|
176
|
+
//# sourceMappingURL=huber_scale_variants.js.map
|