Skip to main content

> ML_LIBRARY // DOWHY_v1.0

DoWhy

PyWhy / Microsoft Research — Microsoft Research and PyWhy's end-to-end framework for causal inference and assumption refutation.

probabilistic-modellingv0.11.1MITqualified
Official DocumentationSource Repository

Model Training

Supported
Accelerators:
CPU
Distributed Training:No

Model Inference

Supported
Inference Accelerators:
CPU
Deployment Targets:server

What It Does

  • +Rigorous 4-step causal pipeline: Model (DAGs), Identify (backdoor/frontdoor criterion), Estimate, and Refute
  • +Automated causal refutation tests (placebo treatments, random unobserved confounders, dummy outcomes)
  • +Root cause analysis attributing system microservice anomalies to specific causal paths (GCM module)
  • +Seamless integration with EconML for heterogeneous treatment effect estimation

What It Does Not Do

  • -Train vision foundation models or speech recognition networks
  • -Replace transactional relational databases
  • -Execute in low-power mobile browser engines

>Suitable Work Types

  • Software incident root-cause attribution across distributed microservices
  • Estimating true customer uplift from promotional marketing campaigns without confounding bias
  • Validating medical intervention effects in observational non-randomized health studies

>Unsuitable Work Types

  • Pure predictive Kaggle competitions where confounding does not matter as long as correlation predicts the test set
  • High-frequency algorithmic trading microservices
Data Residency Implications

Runs strictly locally on corporate compute servers. Zero telemetry.

Security Considerations

MIT license with permissive commercial rights. PyWhy governance ensures enterprise safety.

Operational Profile & Known Limitations

Maturity:mature
Learning Curve:high
Ops Complexity:moderate
Cost Tier:free-oss
> Known Limitations:
  • Causal conclusions are only as valid as the underlying Directed Acyclic Graph (DAG) assumptions; omitted confounders violate backdoor identification.

Associated Incident Patterns (Incidentpedia)

Enforce safeguards and monitoring to guard against these documented real-world failure modes:

training-serving-skew

> Primary Evidence & Benchmark Citations

DoWhy Documentationofficial-docs • >=0.10.0, <=0.11.x
2026-09-25HIGH