jszitas Goto Github PK

followers: 19.0 following: 47.0 repos: 50.0 gists: 3.0

Name: Juraj Szitás

Type: User

Company: Barclays

Bio: Credit Quant Developer Interested in Optimization |Time Series Forecasting | Machine Learning | Trees

Location: Prague

Hello! I’m Juraj Szitas, a somewhat mad Econometrician turned Data Scientist. I keep my open source stuff here - in the hopes that someone finds it useful. I work primarily on Time Series and Time Series ensembling.

Check out:

soothsayer if you like the fable framework and like the idea of meta learning for time series
blaze (WIP) A full fledged time series forecasting and analysis toolkit in modern C++.
- contains a new (S)ARIMA(X) implementation leveraging SIMD
- fully capable AR and AutoAR
- Benchmark methods (Integrated Noise)
- miscellaneous time series utility functions (seasonality identification, stationarity tests)
stack for some general notes on model stacking
gpvolatility for an implementation of a funky volatility model
treecoding for some interesting Trees and Forests (particularly an implementation of 'Autoencoder by Forest')
nlsolver Nonlinear optimizers as header-only, C++17 library.

I am currently building a lot of things in C++, for fun and profit :)

Juraj Szitás's Projects

autodep

autodep automates writing roxygen tags - and tracking dependencies.

autodepfailexample

Document unexpected behaviour when calling devtools::document()

autostsm

:exclamation: This is a read-only mirror of the CRAN R package repository. autostsm — Automatic Structural Time Series Models

blaze

A C++17 implementation of ARIMA following R

c-tokenizer

tokenize c/c++ source code

categoryencodings

Multiple methods to (quickly) encode factor variables, using data.table

circulant

A stupid simple circular data structure for recursive time series prediction

contexttree

Contextual Decision Tree in Python

crucible

Agent based modelling in R

devtools

Tools to make an R developer's life easier

es_rnn

The repository contains current, slightly updated, version of ES_RNN - a hybrid Exponential Smoothing/Recurrent NN method that won M4 Forecasting Competition