Add Redis rolling stats for flow packets and z-score driven classifiers

This commit is contained in:
dirtydishes 2025-12-30 13:24:48 -05:00
parent fc7065792f
commit 163ab1039e
10 changed files with 389 additions and 32 deletions

View file

@ -10,8 +10,12 @@ type ParsedContract = {
export type ClassifierConfig = {
sweepMinPremium: number;
sweepMinCount: number;
sweepMinPremiumZ: number;
spikeMinPremium: number;
spikeMinSize: number;
spikeMinPremiumZ: number;
spikeMinSizeZ: number;
zMinSamples: number;
};
const clamp = (value: number, min = 0, max = 1): number => {
@ -120,8 +124,14 @@ const buildSweepHit = (
const firstPrice = getNumberFeature(packet, "first_price");
const lastPrice = getNumberFeature(packet, "last_price");
const windowMs = getNumberFeature(packet, "window_ms");
const premiumZ = getNumberFeature(packet, "total_premium_z");
const premiumBaseline = getNumberFeature(packet, "total_premium_baseline_n");
if (count < config.sweepMinCount || totalPremium < config.sweepMinPremium) {
const baselineReady = premiumBaseline >= config.zMinSamples;
const passesAbsolute = totalPremium >= config.sweepMinPremium;
const passesZ = baselineReady && premiumZ >= config.sweepMinPremiumZ;
if (count < config.sweepMinCount || (!passesAbsolute && !passesZ)) {
return null;
}
@ -138,9 +148,19 @@ const buildSweepHit = (
if (totalPremium >= config.sweepMinPremium * 2) {
confidence += 0.15;
}
if (passesZ) {
confidence += 0.1;
if (premiumZ >= config.sweepMinPremiumZ + 1) {
confidence += 0.05;
}
}
confidence = clamp(confidence, 0, 0.95);
const baselineNote = baselineReady
? `Baseline premium z-score ${premiumZ.toFixed(2)} over ${Math.round(premiumBaseline)} samples.`
: "Baseline premium z-score unavailable.";
return {
classifier_id: direction === "bullish" ? "large_bullish_call_sweep" : "large_bearish_put_sweep",
confidence,
@ -148,7 +168,8 @@ const buildSweepHit = (
explanations: [
`Likely ${direction === "bullish" ? "call" : "put"} sweep: ${count} prints in ${Math.round(windowMs)}ms for ${packet.features.option_contract_id ?? packet.id}.`,
`Premium ${formatUsd(totalPremium)} across ${Math.round(totalSize)} contracts; price ${priceTrend}.`,
`Thresholds: >=${config.sweepMinCount} prints and >=${formatUsd(config.sweepMinPremium)} premium.`
`Thresholds: >=${config.sweepMinCount} prints and >=${formatUsd(config.sweepMinPremium)} premium or z>=${config.sweepMinPremiumZ.toFixed(1)}.`,
baselineNote
]
};
};
@ -158,8 +179,19 @@ const buildSpikeHit = (packet: FlowPacket, config: ClassifierConfig): Classifier
const totalPremium = getNumberFeature(packet, "total_premium");
const totalSize = getNumberFeature(packet, "total_size");
const windowMs = getNumberFeature(packet, "window_ms");
const premiumZ = getNumberFeature(packet, "total_premium_z");
const sizeZ = getNumberFeature(packet, "total_size_z");
const premiumBaseline = getNumberFeature(packet, "total_premium_baseline_n");
const sizeBaseline = getNumberFeature(packet, "total_size_baseline_n");
if (totalSize < config.spikeMinSize || totalPremium < config.spikeMinPremium) {
const premiumBaselineReady = premiumBaseline >= config.zMinSamples;
const sizeBaselineReady = sizeBaseline >= config.zMinSamples;
const passesAbsolute = totalSize >= config.spikeMinSize && totalPremium >= config.spikeMinPremium;
const passesZ =
(premiumBaselineReady && premiumZ >= config.spikeMinPremiumZ) ||
(sizeBaselineReady && sizeZ >= config.spikeMinSizeZ);
if (!passesAbsolute && !passesZ) {
return null;
}
@ -173,9 +205,20 @@ const buildSpikeHit = (packet: FlowPacket, config: ClassifierConfig): Classifier
if (count >= 3) {
confidence += 0.1;
}
if (passesZ) {
confidence += 0.1;
if (premiumZ >= config.spikeMinPremiumZ + 1 || sizeZ >= config.spikeMinSizeZ + 1) {
confidence += 0.05;
}
}
confidence = clamp(confidence, 0, 0.9);
const baselineNote =
premiumBaselineReady || sizeBaselineReady
? `Baseline z-scores: premium ${premiumZ.toFixed(2)}, size ${sizeZ.toFixed(2)}.`
: "Baseline z-scores unavailable.";
return {
classifier_id: "unusual_contract_spike",
confidence,
@ -183,7 +226,8 @@ const buildSpikeHit = (packet: FlowPacket, config: ClassifierConfig): Classifier
explanations: [
`Unusual contract spike: ${count} prints in ${Math.round(windowMs)}ms for ${packet.features.option_contract_id ?? packet.id}.`,
`Premium ${formatUsd(totalPremium)} across ${Math.round(totalSize)} contracts.`,
`Thresholds: >=${config.spikeMinSize} contracts and >=${formatUsd(config.spikeMinPremium)} premium.`
`Thresholds: >=${config.spikeMinSize} contracts and >=${formatUsd(config.spikeMinPremium)} premium or z>=${config.spikeMinPremiumZ.toFixed(1)}.`,
baselineNote
]
};
};