DOCSEARCHbyAlgolia
Submit Docs

CausalImpact

repository·master·Indexed 23 days ago

https://github.com/google/causalimpact

An R package that uses Bayesian structural time-series models to estimate the causal effect of an intervention on a time series. It performs causal inference by modeling a counterfactual outcome using control time series when randomized experiments are unavailable.

Tokens
370
Snippets
2
Records
3
Agent score
34%

What's inside CausalImpact

  1. What is CausalImpact and how does it work?

    master

    CausalImpact is an R package used for causal inference on time series data. It estimates the causal effect of a designed intervention (e.g., an advertising campaign) by using a Bayesian structural time-series model.

    How it works

    The package estimates how a response metric would have evolved after an intervention if the intervention had not occurred (the counterfactual). It does this by modeling the outcome time series using a set of control time series.

    Key Assumptions

    For valid conclusions, the following assumptions must hold:

    1. Control Series Integrity: The control time series used in the model must not have been affected by the intervention.
    2. Stable Relationship: The relationship between the treated series and the control series must remain stable during the post-intervention period.

  3. Use the Python implementation of CausalImpact

    master

    The official CausalImpact repository is an R package. For Python users, it is recommended to use the TFP CausalImpact implementation, which is built on TensorFlow Probability. While designed to produce results close to the R package, some differences may occur due to the nature of randomized algorithms.

    https://github.com/google/tfp-causalimpact