Unable to use '%pin%' on 'lhs' about arules

[mhahsler]

I cannot reproduce the error on either the CRAN version or the development version.

library(arules)
data(Groceries)
rules <- apriori(Groceries, 
   parameter=list(support=0.001, confidence=0.8))
x <- 'milk'
rules <- subset(rules, subset = (lhs %pin% as.character(x)) & lift > 1)
rules

set of 63 rules 

Please send me the data and the exact code that produces the error.

[gdbassett]

The data set is proprietary but I can probably provide you a rules object that fails. Would that work?

[mhahsler]

Yes, it would be great if you can modify my code from above so that it fails.

[gdbassett]

I tried duplicating with a dataset I could share and failed, however it fails consistently with the private data set. It failed under versions 1.4-1, 1.4-2, and 1.4-2-1. I can provide a failed rules object and will continue to try and replicate from scratch with a public dataset. Is 'standardGeneric' the appropriate class for 'lhs' (or possibly the rules file in general) or should it be something else?

[gdbassett]

interestingly if I remove the (lhs %pin% as.character(x)) & portion of the subset, I receive the error "object 'lift' not found".

[mhahsler]

I am pretty sure that this is your problem:

rules <- 10
subset(rules, subset = (lhs %pin% as.character(x)) & lift > 1)
Show Traceback

Rerun with Debug
Error in (function (classes, fdef, mtable)  : 
unable to find an inherited method for function ‘%pin%’ for signature ‘"standardGeneric", "character"’ 

In other words, rules does not contain rules and R unfortunately says this not in a very readable way.

[gdbassett]

It's possible. I'm honestly confused and suspect it could be a complication of how it is run. However when I remove the subset line and run I receive full output. Here is the plot output for plot(rules, method="grouped", measure="confidence")): https://s3.amazonaws.com/gabe1/arulesVis.png so I'm confident the rules file is some sort of rules file.

The execution path is private module -> function -> %do% -> function -> rmarkdown:render -> rmarkdown file -> sub-rmarkdown file (incorporated into the main file using knit_expand).

The rmarkdown::render() is called with envir=env.new(). Still, is it possible subset simply doesn't know how to deal with 'lift' and 'lhs' due to the way it is called?

[mhahsler]

This is strange. Does everything also look ok if you say inspect(rules)?

[gdbassett]

inspect(rules) looks normal:

##    lhs                                 rhs                                                       support confidence     lift
## 1  {}                               => {victim.employee_count.Unknown}                         0.2534307  0.2534307 1.000000
## 2  {}                               => {attribute.confidentiality.data.variety.Personal}       0.3153787  0.3153787 1.000000
## 3  {}                               => {attribute.confidentiality.data.variety.Unknown}        0.3827941  0.3827941 1.000000
## 4  {}                               => {actor.external.variety.Unknown}                        0.4337762  0.4337762 1.000000
## 5  {}                               => {asset.variety.Unknown}                                 0.4582470  0.4582470 1.000000
## 6  {}                               => {asset.assets.variety.Unknown}                          0.4582470  0.4582470 1.000000
## 7  {}                               => {actor.internal.variety.Unknown}                        0.4863627  0.4863627 1.000000
## 8  {}                               => {attribute.confidentiality.data_disclosure.Potentially} 0.5349928  0.5349928 1.000000
## 9  {}                               => {victim.employee_count.Large}                           0.7320986  0.7320986 1.000000
## 10 {}                               => {victim.industry2.92}                                   0.7357903  0.7357903 1.000000
## 11 {}                               => {victim.orgsize.Large}                                  0.7384539  0.7384539 1.000000
## 12 {asset.assets.variety.P - Other} => {pattern.Everything Else}                               0.1335846  0.8854016 6.396792
## 13 {asset.assets.variety.P - Other} => {action.social.variety.Phishing}                        0.1487095  0.9856494 6.607947
## 14 {asset.assets.variety.P - Other} => {action.social.vector.Email}                            0.1486939  0.9855461 6.515403
## 15 {asset.assets.variety.P - Other} => {asset.variety.Person}                                  0.1508746  1.0000000 6.497874
## 16 {asset.assets.variety.P - Other} => {attribute.integrity.variety.Alter behavior}            0.1508590  0.9998968 6.480151
## 17 {asset.assets.variety.P - Other} => {actor.external.variety.Unknown}                        0.1360769  0.9019203 2.079229
## 18 {asset.assets.variety.P - Other} => {victim.employee_count.Large}                           0.1265752  0.8389428 1.145942
## 19 {asset.assets.variety.P - Other} => {victim.industry2.92}                                   0.1270736  0.8422465 1.144683
## 20 {asset.assets.variety.P - Other} => {victim.orgsize.Large}                                  0.1267465  0.8400785 1.137618

[gdbassett]

The only odd thing is the 1<->1 mapping. I can't remember if that's an artifact of how I ran apriori or an anomaly.

Clearly I peg asset.assets.variety.P - Other to the LHS, but i'm surprised I don't see groups.

[mhahsler]

did you put the inspect right before the subset in the original code that creates the error (the markdown file)?

[gdbassett]

The inspect is immediately after the subset, however the subset is commented out. (It would get rid of the empty LHS's if it were working.)

[mhahsler]

This is really strange. Try to replace the subset with this:

 rules[lhs(rules) %pin% as.character(x) & quality(rules)$lift>1]

Also, do you load a package that overwrites %pin%?

[gdbassett]

It's possible that %pin% is being overwritten as I load a bunch of stuff, however I tried removing it and still received an error that 'lift' didn't exist. I also directly defined %pin% (%pin%` <- arules::`%pin% ) and still received the error.

Technically I used rules <- rules[lhs(rules) %pin% as.character(x) & quality(rules)$lift>1], but that worked. Received no errors and received the filtered output:

##   lhs                                 rhs                                            support confidence     lift
## 1 {asset.assets.variety.P - Other} => {pattern.Everything Else}                    0.1335846  0.8854016 6.396792
## 2 {asset.assets.variety.P - Other} => {action.social.variety.Phishing}             0.1487095  0.9856494 6.607947
## 3 {asset.assets.variety.P - Other} => {action.social.vector.Email}                 0.1486939  0.9855461 6.515403
## 4 {asset.assets.variety.P - Other} => {asset.variety.Person}                       0.1508746  1.0000000 6.497874
## 5 {asset.assets.variety.P - Other} => {attribute.integrity.variety.Alter behavior} 0.1508590  0.9998968 6.480151
## 6 {asset.assets.variety.P - Other} => {actor.external.variety.Unknown}             0.1360769  0.9019203 2.079229
## 7 {asset.assets.variety.P - Other} => {victim.employee_count.Large}                0.1265752  0.8389428 1.145942
## 8 {asset.assets.variety.P - Other} => {victim.industry2.92}                        0.1270736  0.8422465 1.144683
## 9 {asset.assets.variety.P - Other} => {victim.orgsize.Large}                       0.1267465  0.8400785 1.137618