deep-inference
Deep Learning for Individual Heterogeneity with Valid Inference
deep-inference fits structural economic models with neural networks and still returns
honest 95% confidence intervals. It implements the Farrell, Liang, and Misra (2021, 2025)
influence-function framework, and adds the RieszNet automatic-debiasing procedure of
Chernozhukov et al. (2022).
If you only read one page, read Overview. If you want to run something in the next five minutes, read Quick Start.
Documentation
- Overview
- Quick Start
- Loading Data & Pre-Estimation
- Models
- Estimation
- Inference
- Guide
- Theory
- Replications
- Simulation Studies
- Pass/Fail Thresholds
- Suite Overview
- Running the Eval Suite
- Eval 01: Parameter Recovery
- Eval 02: Autodiff vs Calculus
- Eval 03: Lambda Estimation
- Eval 04: Target Jacobian
- Eval 05: Influence Function Assembly
- Eval 06: Frequentist Coverage
- Eval 07: End-to-End Workflow
- Eval 09: Multinomial Logit
- Verification Against FLM2
- Eval Suite Scorecard
- References
- API Reference
- References