Modern ML for Trading
The full gradient-boosting-to-neural-operator toolkit - ensembles, AutoML, probabilistic forecasting, distillation and pruning for latency, plus physics-informed architectures like Neural ODEs and Fourier operators.
XGBoost for Return Direction: Class Imbalance and Decision Thresholds
Return-direction classifiers are imbalanced problems, and 0.5 is the wrong decision threshold. Comparing scale_pos_weight, focal loss, and precision-constrained threshold optimization on crypto data - plus the engineering differences between XGBoost, LightGBM, and CatBoost.
Ansambl usullari: mustahkam alfa uchun zaif o'quvchilarni birlashtirish
Savdo ansambllari haqida uchta narsa bu blogda yoritilmagan: nega modellar soni emas, kovariatsiya atamasi ansambl xatosida ustunlik qiladi, stacking oʻta meta-xususiyatlarga ega boʻlgan signal-kombinatsiya qatlami sifatida qanday ishlaydi va koʻplab signallar boʻylab aylanma tarmogʻi amalda bajarish narxini pasaytiradimi yoki yoʻqmi.
Tizimli savdo quvurlari uchun AutoML
Avtomatlashtirilgan funksiyalarni yaratish (tsfresh, Featuretools), byudjetdan xabardor modellarni qidirish (FLAMLning tejamkor optimizatori) va WorldQuant formulali alfa zavodi - tadqiqot yoʻnalishining qismlari ushbu blogning qidiruv va haddan tashqari moslama yoyi hech qachon qamrab olinmagan va arcning oʻz natijalari ular haqida nima deydi.
Bir vaqtning o'zida narx, hajm va o'zgaruvchanlikni bashorat qilish uchun ko'p vazifani o'rganish
Qaytish, hajm va o'zgaruvchanlikni birgalikda bashorat qilish haqiqatan ham yordam beradimi? Yo'qotishlarni muvozanatlash sxemalarini o'lchash va gradient kosinus o'xshashligi orqali salbiy uzatish diagnostikasi - klassik asosiy chiziq va tozalangan oldinga burmalar bilan.
Savdo botlarini himoya qilish uchun anomaliyalarni aniqlash: Z-Score'dan Transformer'gacha
Kripto algo-treydingda qaysi anomaliya aniqlash usullari haqiqatan ishlaydi, kaskadli himoya arxitekturasini qanday qurish kerak va nega bu asos bo'lmasa algo-treyding qimor o'yiniga aylanadi.
Scoring Probabilistic Forecasts: CRPS, PIT Calibration, and DeepAR
How to evaluate a predictive distribution honestly - CRPS as a proper scoring rule, the PIT histogram as a calibration diagnostic, and DeepAR sampling in GluonTS.
Irregular Time in Tick Models: Continuous-Time Encodings vs. Plain Positional Embeddings
Sequence models fed tick data still assume regular spacing. Three ways to tell a Transformer when a tick actually happened - learnable-timescale continuous encoding, ODE-RNN latent state, and delta_t as a plain feature - and the ablation that decides between them.
To'liq kunlik kontekst o'n daqiqani uradimi? Diqqat va ketma-ketlik bo'yicha savol
Flash Attention 23 400 bosqichli savdo kontekstini hisoblash uchun bepul qiladi - plitka qo‘yish, onlayn softmax va IO chegarasi N^2 d^2/M. Bu uzoqroq kontekst modelni yaxshiroq qiladimi yoki yo‘qmi, bu alohida savol va uni tasdiqlash emas, o‘lchash kerak.
Knowledge Distillation: Compressing Trading Models for Low-Latency Deployment
The blog's standing answer to the accuracy-vs-latency tension is a two-stage fast/slow split. Distillation is a different answer: train one small model to mimic the ensemble. The KD loss, temperature, born-again nets, early exits for a variable latency budget, and the distill-to-FPGA pipeline - plus the measurements that would decide whether it beats the two-stage split.
Model Pruning for Low-Latency Trading Inference
Magnitude and structured pruning, the Lottery Ticket Hypothesis, movement pruning, distillation and 2:4 sparsity - the methods behind shrinking a trading model, and what still has to be measured before any of it ships.
Parametrik bo'lmagan narxlarni modellashtirish uchun Gauss jarayonlari
Moliyaviy vaqt seriyalari uchun yadro dizayni - onaning pürüzlülüğü, mahalliy davriy tarkibi, spektral aralashmalar - va hech qanday tekshirish to'plamiga muhtoj bo'lmagan tartibga soluvchi sifatida chekli ehtimollik. Bundan tashqari, biron bir sotilishidan oldin o'lchanishi kerak bo'lgan narsalarning halol ro'yxati.
Epistemik va Aleatorik: Qaytarish Modelining Bilmedagini O'chash
Ushbu blogdagi har bir pozitsiya o'lcham qoida noaniqlikni bitta skalerga qisqartiradi. MC Dropout va chuqur ansambllar uni modelning bilmasligi va bozor shovqini sifatida ajratadi - va bu ikki turi turli pozitsiya o'lchamlarini nazarda tutadi.
Neyron ODElar: Uzluksiz vaqt Delta-Vaqt xususiyatidan ustun turadimi?
Neyron ODE'lar, Neyron SDE'lar va tartibsiz tanlangan bozor ma'lumotlari uchun doimiy normalizatsiya oqimlari - va uzluksiz dinamika ularning hal qiluvchi narxiga arziydimi yoki yo'qligini hal qiladigan bitta ablasyon.
Hamiltonian Neural Networks: Does a Financial System Conserve Anything?
Hamiltonian Neural Networks are provably stable - but stability is worthless if the conserved quantity does not exist. Learning a scalar H via autograd, symplectic integration, and the falsification test that decides whether a financial (q, p) pair is canonical at all.
PDE-ga asoslangan moliyaviy modellashtirish uchun Furye neyron operatori
Operatorni o'rganish nuqtalarni emas, balki butun funktsiya bo'shliqlarini xaritalaydi - Furye Neyron Operatori chastotalar bo'shlig'ida PDE yechim operatorini qanday parametrlashtirgani, opsion narxlari uchun nimani sotib olishi va uning da'volaridan qaysi biri hali ham haqiqiy uskunada o'lchashga muhtoj.
