Course

Forecasting Systems: Scale, Shift, Trust

Bartosz Szabłowski

Data scientist specializing in forecast

This course is popular

4 people enrolled last week.

Build a strong global forecasting model in Python

A global model can improve the average forecast while still failing on whole groups of series. Differences in scale, sparse demand, and changing patterns make it difficult to tell which changes actually help.

We will work through one forecasting case from the initial data review to backtesting, baselines, model comparison, and a global machine learning model. Most of the modeling work will focus on that global model: how to prepare its inputs, scale the series, train it, and investigate its errors. Local and foundation models will provide useful points of comparison.

During four live workshops, I will build and revise the solution in Python, explain the decisions, and discuss the results with you. You will keep the course code and reference materials. Two further sessions give us time to discuss participants’ forecasting problems and review their approaches.

Enrollment is now open for the first cohort, running December 1–18, 2026. You can review the syllabus and session schedule below and reserve your place by enrolling.

What you’ll learn

Build and evaluate a global forecasting model, from data preparation and baselines to error analysis and uncertainty.

  • Identify gaps, sparse demand, differences in scale, and changes in trend or seasonality.

  • Decide which groups of series need separate checks and where learning across series may help.

  • Design backtests using only information available at each forecast date.

  • Compare models against the same baselines and inspect bias and errors across series, groups, and horizons.

  • Evaluate the three approaches on the same data and backtests.

  • Use Forecast Value Added, computing cost, and maintenance needs to decide which approach to develop further.

  • Build SOTA features. Test whether external variables improve forecasts using only information available at forecast time.

  • Test target transformations and ts scaling for multi-horizon forecasting. Diagnose errors for sparse series and short histories.

  • Evaluate model changes against the same baselines and assess prediction intervals through coverage and width.

Learn directly from Bartosz

Bartosz Szabłowski

Bartosz Szabłowski

Contact

Forecasting practitioner and VN2 competition winner

See all products from Bartosz

Who this course is for

  • Data Scientists and ML Engineers building forecasting solutions across many time series.

  • Forecasting Leads and senior practitioners who review models, set evaluation standards, and guide forecasting development.

Prerequisites

  • Comfortable working with Python

    You can read and modify Python code and work with tabular data.

  • Working knowledge of machine learning

    You know regression, training, validation and test splits, overfitting, and evaluation metrics.

  • Basic experience with time series forecasting

    You have worked with forecast horizons, lags, seasonality, and forecast evaluation.

What's included

Bartosz Szabłowski

Live sessions

Learn directly from Bartosz Szabłowski in a real-time, interactive format.

Lifetime access

Go back to course content and recordings whenever you need to.

Community of peers

Discuss forecasting questions and compare approaches with other participants.

Certificate of completion

Share your new skills with your employer or on LinkedIn.

4 workshops and 2 discussion sessions

Four live workshops of two hours each, followed in week three by the Forecasting Roundtable and Forecasting Solutions Showcase.

Course code and a shared forecasting case

A Python reference repository that develops from backtests and baselines into a global forecasting model, with examples you can revisit after class.

Discussion of participants’ forecasting problems

Bring a modeling question to the Roundtable and prepare a short case for the Showcase. We will discuss the choices, results, and possible next steps.

Maven Guarantee

Your purchase is backed by the Maven Guarantee.

Course syllabus

Week 1

Dec 1—Dec 6

    Build backtests you can trust

    2 items

    Build a baseline worth beating

    2 items

Week 2

Dec 7—Dec 13

    Choose the right forecasting approach: local, global, or foundation

    2 items

    Build a robust global model that generalizes

    5 items

Schedule

Live sessions

4 hrs / week

Four workshops of two hours each during weeks one and two. Week three contains the Forecasting Roundtable and Forecasting Solutions Showcase. Exact session times will be confirmed before enrollment opens.

Preparation & applied work

1-2 hrs / week

Allow one to two hours each week to review examples and try the course code. There is no required coding project. For week three, prepare a short case or one slide for the Forecasting Solutions Showcase.

Frequently asked questions

Maven for Teams

Reimbursement

Get your company to pay

Everything L&D needs: email template, receipts, and certificate of completion.

Get reimbursed

Team discount

Learn with your teammates

Save 20%+ when 2 or more teammates enroll in the same cohort.

Save 20%+ with a team

Private cohort

Run a cohort for your org

A dedicated cohort with a custom schedule and curriculum, tailored to your team.

Book a private cohort

$997

USD

Nov 30—Dec 18
Enroll