DoTime

A synthetic benchmark generator for interventional and counterfactual time series.

dotime samples multivariate temporal structural causal models (SCMs), applies interventions, and produces paired observational / interventional trajectories with exact counterfactual targets. It ships four frozen evaluation suites, reference baselines, and an evaluation harness.

Getting started

• Quick Start
  • 1. Generate paired observational / interventional data
  • 2. Generate a dataset
  • 3. Load a frozen benchmark
  • 4. Evaluate a baseline
  • Where to go next
• Frozen Benchmark Suites
  • Suites
  • Loader
  • Evaluation protocol
• Leaderboard
  • Metrics
  • Reference results (preliminary)
  • Submitting

Reference

• API Reference
  • Prior and SCM types
  • benchmarks
  • baselines
  • evaluation
• Troubleshooting
  • Install
  • Runtime
  • Benchmark cache

Install

pip install dotime              # core generator + suite loaders (CPU)
pip install 'dotime[baselines]' # classical / Bayesian baselines
pip install 'dotime[models]'    # the Do-Over-Time-PFN model
pip install 'dotime[all]'       # everything except dev tooling

Quickstart

from dotime import DoTime

prior = DoTime(seed=42)
X_obs, X_int, intervention, scm = prior.generate_pair(T=100)

See Quick Start for the benchmark-loading and evaluation walkthrough.

Indices

• Index

• Module Index