carl-lib
1 Language
create-inputs
2 Grounding
ground
3 Embedding
embed
4 Covariate Detection
detect
5 Unit Table Construction
construct
6 Estimation
estimate
7.8

carl-lib

Moe Kayali

 (require carl-lib) package: carl-lib

The Causal Relational Learning (CaRL) framework is a tool for causal inference over relational data.

This software is currently in pre-alpha.

This is the reference implementation of the framework developed in the paper “Causal Relational Learning.”

1 Language

procedure

(create-inputs port)  inputs?

  port : input-port?
Read a file in the CaRL language and output the set of rules and queries it contains.

2 Grounding

procedure

(ground model conn)  (and/c graph? unweighted-graph?)

  model : (listof rule?)
  conn : connection?
Combine CaRL rules with a database instance to build a ground causal model.

3 Embedding

procedure

(embed gcm)  matrix?

  gcm : (listof pre-row?)
Apply embeddings to a semi-structured table to create flat table.

4 Covariate Detection

procedure

(detect model T Y)  (or/c list? #f)

  model : (listof rule?)
  T : any/c
  Y : any/c
Return a minimal set of sufficient covariates for adjustment of T on Y. #f if no such set found.

5 Unit Table Construction

procedure

(construct augmented-gcm query Z)  (listof pre-row?)

  augmented-gcm : (and/c graph? unweighted-graph?)
  query : c-query?
  Z : list?
Construct a unit table given an augmented GCM and a set of covariates.

6 Estimation

procedure

(estimate unit-table)  real?

  unit-table : matrix?
Estimate the average treatment effect (ATE) given a unit table.