| Title: | Causal Discovery from Discrete Data using Hidden Compact Representation |
|---|---|
| Description: | This code provides a method to fit the hidden compact representation model as well as to identify the causal direction on discrete data. We implement an effective solution to recover the above hidden compact representation under the likelihood framework. Please see the Causal Discovery from Discrete Data using Hidden Compact Representation from NIPS 2018 by Ruichu Cai, Jie Qiao, Kun Zhang, Zhenjie Zhang and Zhifeng Hao (2018) <https://nips.cc/Conferences/2018/Schedule?showEvent=11274> for a description of some of our methods. |
| Authors: | Jie Qiao [aut, cre], Ruichu Cai [ths, aut], Kun Zhang [ths, aut], Zhenjie Zhang [ths, aut], Zhifeng Hao [ths, aut] |
| Maintainer: | Jie Qiao <[email protected]> |
| License: | GPL (>= 2) |
| Version: | 0.1.1 |
| Built: | 2025-02-13 05:12:53 UTC |
| Source: | https://github.com/cran/HCR |
Causal Discovery from Discrete Data using Hidden Compact Representation.
HCR(X, Y, score_type = "bic", is_anm = FALSE, is_cyclic = FALSE, verbose = FALSE, max_iteration = 1000, ...)HCR(X, Y, score_type = "bic", is_anm = FALSE, is_cyclic = FALSE, verbose = FALSE, max_iteration = 1000, ...)
X |
The data of cause. |
Y |
The data of effect. |
score_type |
You can choose "bic","aic","aicc","log" as the type of score to fit the HCR model. Default: bic |
is_anm |
If is_anm=TRUE, it will enable a data preprocessing to adjust for the additive noise model. |
is_cyclic |
If is_anm=TRUE and is_cyclic=TRUE, it will enable a data preprocessing to adjust the cyclic additive noise model. |
verbose |
Show the score at each iteration. |
max_iteration |
The maximum iteration. |
... |
Other arguments passed on to methods. Not currently used. |
The fitted HCR model and its score.
library(data.table) set.seed(10) data=simuXY(sample_size=200) r1<-HCR(data$X,data$Y) r2<-HCR(data$Y,data$X) # The canonical hidden representation unique(r1$data[,c("X","Yp")]) # The recovery of hidden representation unique(data.frame(data$X,data$Yp))library(data.table) set.seed(10) data=simuXY(sample_size=200) r1<-HCR(data$X,data$Y) r2<-HCR(data$Y,data$X) # The canonical hidden representation unique(r1$data[,c("X","Yp")]) # The recovery of hidden representation unique(data.frame(data$X,data$Yp))
A fast implementation for fitting the HCR model. This implementation caches all intermediate results to speed up the greedy search. The basic idea is that if there are two categories need to be combined, for instance, X=1 and X=2 mapping to the same Y'=1, then the change of the score only depend on the frequency of the data where X=1 and X=2. Therefore, after combination, if the increment of the likelihood is greater than the penalty, then we will admit such combination.
HCR.fast(X, Y, score_type = "bic", ...)HCR.fast(X, Y, score_type = "bic", ...)
X |
The data of cause. |
Y |
The data of effect. |
score_type |
You can choose "bic","aic","aicc","log" as the type of score to fit the HCR model. Default: bic |
... |
Other arguments passed on to methods. Not currently used. |
The fitted HCR model and its score.
library(data.table) set.seed(1) data=simuXY(sample_size=2000) r1=HCR.fast(data$X,data$Y) r2=HCR.fast(data$Y,data$X) # The canonical hidden representation unique(r1$data[,c("X","Yp")]) # The recovery of hidden representation unique(data.frame(data$X,data$Yp))library(data.table) set.seed(1) data=simuXY(sample_size=2000) r1=HCR.fast(data$X,data$Y) r2=HCR.fast(data$Y,data$X) # The canonical hidden representation unique(r1$data[,c("X","Yp")]) # The recovery of hidden representation unique(data.frame(data$X,data$Yp))
Generate the X->Y pair HCR data
simuXY(sample_size = 2000, min_nx = 3, max_nx = 15, min_ny = 3, max_ny = 15, type = 0, distribution = "multinomial")simuXY(sample_size = 2000, min_nx = 3, max_nx = 15, min_ny = 3, max_ny = 15, type = 0, distribution = "multinomial")
sample_size |
Sample size |
min_nx |
The minimum value of |X| (Default: 3) |
max_nx |
The maximum value of |X| (Default: 15) |
min_ny |
The minimum value of |Y| (Default: 3) |
max_ny |
The maximum value of |Y| (Default: 15) |
type |
type=0: standard version, type=1: |X|=|Y|, type=2: |Y'|=|Y|, type=3: |X|=|Y'|, type=4: |X|=|Y'|=|Y| (Default: type=0) |
distribution |
The distribution of the cause X. The options are "multinomial","geom","hyper","nbinom","pois". Default: multinomial |
return the synthetic data
df=simuXY(sample_size=100,type=0) length(unique(df[,1])) length(unique(df[,2])) length(unique(df[,3])) df=simuXY(sample_size=100,type=1) length(unique(df[,1])) length(unique(df[,3])) df=simuXY(sample_size=100,type=2) length(unique(df[,2])) length(unique(df[,3])) df=simuXY(sample_size=100,type=3) length(unique(df[,1])) length(unique(df[,2])) df=simuXY(sample_size=100,type=4) length(unique(df[,1])) length(unique(df[,2])) length(unique(df[,3]))df=simuXY(sample_size=100,type=0) length(unique(df[,1])) length(unique(df[,2])) length(unique(df[,3])) df=simuXY(sample_size=100,type=1) length(unique(df[,1])) length(unique(df[,3])) df=simuXY(sample_size=100,type=2) length(unique(df[,2])) length(unique(df[,3])) df=simuXY(sample_size=100,type=3) length(unique(df[,1])) length(unique(df[,2])) df=simuXY(sample_size=100,type=4) length(unique(df[,1])) length(unique(df[,2])) length(unique(df[,3]))