Continued thread
Let's Make Forecasting Great Again.
Drop a comment or DM me if you're interested.
Let’s make forecasting history together!
Recent searches
Search options
#TimeSeries
9 posts6 participants1 post today
Continued thread
As they say it is not a drill. Enroll now if you want the first cohort with special perks.
https://maven.com/valeriy-manokhin/modern-forecasting-mastery
maven.comModern Forecasting Mastery by Valery Manokhin on MavenMaster forecasting with real-world, no-hype insights! Get hands-on with cutting-edge tools and learn what truly works. No fluff—just results
Continued thread
No more toy datasets, just real impact. Learn more: [Course Link](https://maven.com/valeriy-manokhin/modern-forecasting-master)
Continued thread
Check my repo “ Transformers_And_LLM_Are_What_You_Dont_Need”
Continued thread
Bottom line: R² can be informative in static regression, but in forecasting, it's a distraction at best — and deceptive at worst.
Forecasting is a craft. Planet 
time series is an unforgiving place.
#Day6 of #30DayChartChallenge, theme: #FlorenceNightingale
Estacionalidad promedio de compraventas de vivienda en España (2007-2024). Coxcomb plot muestra meses con ventas típicamente >100 (altas) o <100 (bajas) vs media anual.
Método: X-13ARIMA-SEATS con #rstats {seasonal}. Datos: INE.
Código: https://t.ly/lC34V
For anyone who thinks transformers and LLMs can manage gravity on planet time series, shows them this.
Everyone knows ARIMA as Box and Jenkins framework.
However the research on linear models for time series goes back almost 50 years before Box and Jenkins to the works of Kolmogorov (1941) and Wiener (1949).
Continued thread
I’ll be pairing it with a book by a former head of analytics at Zillow—let’s call it part of the “Dodo Book Collection”.
Stay tuned—I'll be sharing some forecasting insights soon, especially from the “simple methods often outperform” school of thought.
Continued thread
#Forecasting #DataScience #MachineLearning #TimeSeries
*[Engage below! What’s your go-to metric for forecasting?]*
Continued thread
No more toy datasets, just real impact. Learn more: [Course Link](https://maven.com/valeriy-manokhin/modern-forecasting-master)
If I ever decide to publish a book on time series, I can assure my reader it won’t be reviewed by a chap who wrote a book about …. Facebook prophet.
Continued thread
When they don’t forecast house prices for Zillow they moonlight writing medium articles how to predict stock prices and make you rich
Continued thread
- Learning about #timeseries and #forecasting is boring; just ask ChatGPT to produce your forecast
- Forecasting has no business value whatsoever
- The best way to land data science job is to spin LSTM on 30 data points
- Transformers and LLMs rule supreme.
Continued thread
The paper prove that even when data isn't exchangeable (like in time series), coverage loss can be bounded—so in many real-world cases, it doesn’t matter much!
These methods power tools like NeuralProphet and Nixtla, and now they’ve got solid math backing them up.