The leader track 🤝 · for PMs, product & business leaders · 6 x 45 min · no code, ever
Thinking mode: what a forecast really is in plain English, the ladder of methods and when each fits, who on the team does what, and how you judge whether a forecast is any good - each session ends with the questions to ask your team.
Why forecast
Where forecasting drives real value - demand, staffing, cash-flow, web traffic - and the two-sided cost of getting it wrong.
📈 Leader 2 · easyThe shape of a series
Trend, seasonality, cycle, and noise - the four things every series is made of, with a visual for each.
🪜 Leader 3 · mediumThe methods, without math
The ladder in plain English: baselines to exponential smoothing to ARIMA to Prophet to ML and deep - and when each earns its place.
👥 Leader 4 · mediumThe team + the pipeline (R&R)
Who does what: the analyst profiling history, the data scientist modelling, engineering serving it, and the team monitoring drift.
🎯 Leader 5 · decidingMeasuring a forecast
MAPE, RMSE, and MASE, backtesting, "beat the naive baseline", and why every forecast needs a prediction interval.
🧭 Leader 6 · decidingStrategy & pitfalls
Horizon, leakage, over-fitting, cold-start, and the big one: knowing when NOT to forecast at all.
The builder track 🛠️ · for analysts, data scientists & engineers · 10 x 45 min · Python + a notebook
One running series, Lumen's daily web sessions, threaded through the demos: you implement each method in real Python and follow one forecast from the analyst's first plot to a monitored production model.
Forecasting foundations
The datetime index, resampling, the honest time-based split, and the seasonal-naive baseline every model must beat.
🧩 Builder 2 · easyDecomposition & baselines
Pull a series apart with STL, then ship the family of baselines - naive, drift, and seasonal-naive - the lines to beat.
〰️ Builder 3 · mediumExponential smoothing & ETS
SES, Holt's trend, and Holt-Winters seasonality in statsmodels - plus a live forecaster you drive in the browser.
📐 Builder 4 · mediumARIMA & SARIMA
The Box-Jenkins method: stationarity, differencing, ACF/PACF, model orders, and auto_arima on the Lumen series.
🔮 Builder 5 · mediumProphet
Decomposable trend, seasonality, and holidays - Facebook's analyst-friendly forecaster (Taylor & Letham 2018), with changepoints.
🌲 Builder 6 · hands-onML forecasting
Turn forecasting into regression: lag, rolling, and calendar features feeding gradient boosting - direct vs recursive.
🤖 Builder 7 · hardestDeep forecasting
Where the field is in 2026: DeepAR, N-BEATS, and Temporal Fusion Transformers - surveyed and cited, with an illustrative sketch.
📏 Builder 8 · mediumBacktesting & metrics
Rolling-origin cross-validation, MAPE, RMSE, and MASE, and prediction intervals with real coverage - the honest scoreboard.
🚀 Builder 9 · hands-onProductionizing
Retrain cadence, batch serving, the forecast API and dashboard, and the data-scientist-to-engineering handoff.
📡 Builder 10 · hardestMonitoring
Forecast-error tracking, data and concept drift, when to retrain, and the closed loop that keeps forecasts honest.
Choose your path 🗺️
Two doors, one course. Leaders get the concepts and the pipeline; builders implement every method in Python - and both meet at the live forecaster.
The knowledge map 🧠
Both tracks at a glance - hover a session to spotlight its concepts, click any node to jump in.