Skip to content

Tutorials

Each tutorial makes one thing work from start to finish. It gives the screens or the code that you need. The code is not only an example. Each Python snippet runs in the test suite. The Apps Scripts run against substitutes for the services of Google.

  • Your first MMM in Guided mode

    Start with a spreadsheet of weekly spend and revenue. Get the ROI for each channel, a sanity scorecard and a budget scenario, without code. It takes about ten minutes.

    For analysts. Start

  • Connect Google Sheets

    An Apps Script writes the forecast of a model into a sheet, with its likely range. It refreshes the forecast each morning.

    For analysts and integrators. Start

  • Embed a chart in Notion, Slack or BI

    Make a read-only link to the chart of a model, for a doc, a dashboard or a thread. The link expires. You also get a PNG for tools that cannot show live content.

    For analysts. Start

  • Receive webhooks

    Make a FastAPI or Flask receiver. It verifies the signature of each delivery and ignores duplicates. It does an action when a model is ready or when a run fails.

    For developers. Start

  • Optimise a budget

    Ask a fitted marketing-mix model how to divide a budget across channels. Learn about bounds, groups, cautious objectives, the uplift, the probability that it helps, and the efficient frontier.

    For data scientists. Start

  • Measure the effect of a campaign v0.13

    Measure what a campaign, a launch or a geo-lift test changed, against a prediction without the change. Then give the result of a geo-lift test to the marketing mix model.

    For analysts and data scientists. Start

  • Compare runs in an experiment v0.14

    Log metrics, params and images from a fit. Put the runs of a study in an experiment, and compare 2 to 10 runs side by side.

    For data scientists. Start · Track metrics

  • Classify images v0.14

    Upload a zip of images, look at them in the gallery, hold out a test set for each class and fit a classifier. For large image sets, train a neural network on a collection.

    For data scientists. Start · Deep learning

  • A forecast that tests itself v0.5

    Make one workflow with unit-root tests, a differencing decision and an AR/MA order search. Add a rolling-origin backtest against naive forecasts, and governance. Run it with --watch and schedule it for Monday mornings.

    For data scientists. Start

  • Prove that a model does not fit noise v0.5

    Write a policy with null tests and validate a model version. Get the approval of a colleague and export the evidence pack. Then set the endpoint to refuse all versions that are not validated.

    For data scientists and reviewers. Start

  • Science models from pymc-challenges

    Fifteen case studies, from disease maps and networks to images, fields and directions. Each page gets the data, fits the model in the web app, reads the results, scores the model and runs a workflow recipe. It compares the numbers with the notebook.

    For data scientists. Start

  • Complex workflows

    Build model ladders, convergence gates, posterior predictive checks, governance checks, calibrate-then-design runs, plots in reports and sub-workflows in the workflow editor.

    For data scientists. Start

  • Explore your data, then choose a model v0.5

    Read the quality warnings, stationarity views and lead-lag views of the Data Explorer. Then answer some questions in Help me choose. It gives a shortlist with reasons.

    For analysts. Start · Choosing a model

Before you start

  • A platform to talk to. All tutorials work against a local stack: make install && make dev && make seed (Quickstart). The Google Sheets and public-embed tutorials need a deployment that the internet can get to. The reason is that Apps Script loads from the machines of Google, and embedded charts load from the machines of your viewers.
  • A fitted model. The Guided tutorial makes one. make seed loads a demo marketing-mix spec that you can fit from the web app.
  • An endpoint, for Sheets and other tools. To put a model live, use the Guided Share step or the Integrations hub.

For more guides about tasks, see Guides. For the ideas behind them, see Concepts. Use cases shows which tutorials apply to which problem.