Time Series Bootcamp, Packt (1/4)

Time Series Bootcamp, Packt (1/4) | Dr Krzysztof Ozimek

Packt banner for the Time Series Forecasting 2-Day Hands-On Bootcamp on November 28 to 29, 2026, listing instructors Graeme Davidson, Lei Ma and Jeffrey Tackes, with the text I'll be there from Dr Krzysztof Ozimek, PRM

Slide from the Packt Time Series Forecasting 2-Day Hands-On Bootcamp, 28 to 29 November 2026: the attendee Dr Krzysztof Ozimek expects to see how these model families can support trading signals and backtesting in the trading domain

I'll be attending Time Series Bootcamp: Forecasting with ML, Deep Learning, Foundation Models, a live, hands-on online bootcamp by Packt Publishing, on November 28–29, 2026.

10:00 AM – 2:00 PM EST / 4:00 PM – 8:00 PM CET on both days (four hours each).

Which forecasting model should you choose: statistical, machine learning, deep learning, or a foundation model? The bootcamp compares all four on a shared retail dataset, with a common evaluation framework and a shared leaderboard covering accuracy, bias, stability, uncertainty, computational cost, and business impact.

Instructors: Graeme Davidson and Lei Ma, co-authors of "Time Series with PyTorch" (Davidson & Ma, 2026), and Jeffrey Tackes of Forecast Academy, co-author of "Modern Time Series Forecasting with Python", Second Edition (Joseph & Tackes, 2024). Both books are published by Packt.

Day 1: baselines, evaluation, machine learning

Baselines, evaluation metrics, initial model selection
Global ML models with MLForecast; exogenous drivers and feature engineering with Nixtla
Direct vs recursive multi-horizon forecasting
Objective functions, model optimisation, model selection
Final leaderboard and Day 2 handoff

Day 2: deep learning, foundation models, uncertainty

Recap of the Day 1 leaderboard
Deep learning for global modelling: N-BEATS, N-HiTS, Temporal Fusion Transformers (TFT) vs gradient-boosted decision trees (GBDTs)
ML vs neural networks: data transformations, features, loss functions
Labs: default training, hyperparameter optimisation, exogenous variables
Foundation models: zero-shot vs fine-tuned, covariates, context length, inference costs
Labs: to fine-tune or not to tune?; hierarchy and forecast coherence
Uncertainty: proper scoring rules, conformal prediction
Lab: model selection and ensembling with win rates, Elo, Bradley–Terry rankings
Final standings across ML, neural networks, foundation models

Tools: StatsForecast, MLForecast, NeuralForecast, TimeCopilot, Chronos-2, TabPFN-TS; reusable notebooks, a GitHub repository, an evaluation harness, a Certificate of Completion.

As a researcher and educator in quantitative finance, I expect to see more clearly how these model families, carried over to the trading domain, can be applied in a supporting role to generate or verify trading signals and to enrich the backtesting process.

Apart from this teaser, I plan three more posts:

Local vs Global Forecasting Models: One Model per Series, or One for All?
Foundation Models for Time Series: What They Are and What to Check Before Relying on Them
a review with my impressions after the bootcamp

Full details and registration: https://www.eventbrite.co.uk/e/time-series-bootcamp-forecasting-with-ml-deep-learning-foundation-models-tickets-2003717845013?aff=Ozimek&discount=OZI40

This link includes a 40% discount (code OZI40), applied automatically at checkout.

Disclosure: I received a complimentary ticket to this event from Packt Publishing. The views above are my own.