The goal of Rpriori is to create association rules of type X=>Y by using the Apriori algorithm. For a more detailed demonstration of how the package works please consider the examples.R file in the examples folder. This README only covers the most important features.
You can install Rpriori from github with:
# install.packages("devtools")
devtools::install_github("TimToebrock/Rpriori")This is a basic example which shows you how to create association rules with Rpriori using the Groceries dataset.
Before we do that it might be useful to take a look at our data beforehand. Use the summary() function of Rpriori to do so:
summary(Groceries)
#>
#> Transaction database in binary sparse matrix representation
#> with 169 rows (items) and
#> 9835 columns (itemsets/transactions) and
#> a density of 0.0261 (sparsity: 0.9739)
#>
#> Most frequent items:
#> whole milk other vegetables rolls/buns soda yogurt
#> 2513 1903 1809 1715 1372
#> bottled water root vegetables tropical fruit
#> 1087 1072 1032
#>
#> Distribution of itemset length:
#>
#> 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
#> 2159 1643 1299 1005 855 645 545 438 350 246 182 117 78 77 55 46 29 14 14 9
#> 21 22 23 24 26 27 28 29 32
#> 11 4 6 1 1 1 1 3 1
#>
#> Min. 1st Qu. Median Mean 3rd Qu. Max.
#> 1.000 2.000 3.000 4.409 6.000 32.000We have 9835 transactions recorded with a total of 169 items and a density of 0.026. The average basket has 4 items in it.
Now we can try to find frequent items with the FindFrequentItemsets() function:
Frequent <- FindFrequentItemsets(Groceries, minsupport = 0.1)
show(Frequent)
#> Found 8 frequent itemset(s). Use print() to displayThere are 8 frequent itemsets that occur in one out of ten transactions. If you want to take look at the itemsets now you can use the print() function.
print(Frequent)
#> items support
#> 4 whole milk 0.2555160
#> 3 other vegetables 0.1934926
#> 6 rolls/buns 0.1839349
#> 8 soda 0.1743772
#> 5 yogurt 0.1395018
#> 7 bottled water 0.1105236
#> 2 root vegetables 0.1089985
#> 1 tropical fruit 0.1049314To create rules you need to supply AssociationRules() with a transactions database and a minimum support threshold. You can additionally set a minimum confidence threshold, the default value for minimal confidence is 0.
Rules <- AssociationRules(Groceries, minsupport = 0.01, minconfidence = 0.2)
show(Rules)
#> Found 231 rule(s). Use print() to displayWe found 231 rules Association rules with our specified paramters.
To get summary statistics on rules simply call summary()
summary(Rules)
#> Found 231 rule(s). Use print() to display
#>
#>
#> Summary of quality measures:
#> support confidence lift leverage
#> Min. :0.01007 Min. :0.2006 Min. :0.8991 Min. :-0.004495
#> 1st Qu.:0.01200 1st Qu.:0.2470 1st Qu.:1.4455 1st Qu.: 0.004608
#> Median :0.01474 Median :0.3180 Median :1.7278 Median : 0.006500
#> Mean :0.01903 Mean :0.3324 Mean :1.7924 Mean : 0.007261
#> 3rd Qu.:0.02222 3rd Qu.:0.4035 3rd Qu.:2.0781 3rd Qu.: 0.008461
#> Max. :0.07483 Max. :0.5862 Max. :3.2950 Max. : 0.026291If you want to take a look at the underlying data used in rule creation there are multiple ways. One way is to use the extract function:
Frequent <- extract(Rules)
summary(Frequent)
#>
#> Frequent itemsets in binary sparse matrix representation
#> with 88 rows (items) and
#> 333 columns (frequent itemsets)
#>
#> Most frequent items:
#> whole milk other vegetables yogurt rolls/buns root vegetables
#> 71 63 39 36 34
#> soda tropical fruit sausage
#> 28 26 17
#>
#>
#> Observed frequency in frequent itemsets:
#> whole milk other vegetables yogurt rolls/buns root vegetables
#> 0.2132 0.1892 0.1171 0.1081 0.1021
#> soda tropical fruit sausage
#> 0.0841 0.0781 0.0511
#>
#>
#> Distribution of itemset length:
#>
#> 1 2 3
#> 88 213 32
#>
#> Min. 1st Qu. Median Mean 3rd Qu. Max.
#> 1.000 1.000 2.000 1.832 2.000 3.000
#>
#>
#> Summary of the support measure:
#> Min. 1st Qu. Median Mean 3rd Qu. Max.
#> 0.01007 0.01190 0.01627 0.02507 0.02603 0.25552This extracts the frequent itemsets used to calculate association rules. You can also create a frequent itemmatrix directly:
Frequent2 <- FindFrequentItemsets(Groceries, 0.01)
Frequent2
#> Found 333 frequent itemset(s). Use print() to displaySince frequent itemset generation takes a lot longer than rule creation, it might be better to create a frequent item matrix first, and then use AssociationRules() to calculate rules.
fRules <- AssociationRules(Groceries, Frequent, minsupport = 0.03, minconfidence = 0.4)In this case AssociationRules won't need to recalculate the frequent item-sets if you do not lower the support threshold.
If you want to take a look at the transactions matrix used to calculate the frequent items you need to create a TAMatrix object first:
Transactions <- makeTAMatrix(Groceries)
summary(Transactions)
#>
#> Transaction database in binary sparse matrix representation
#> with 169 rows (items) and
#> 9835 columns (itemsets/transactions) and
#> a density of 0.0261 (sparsity: 0.9739)
#>
#> Most frequent items:
#> whole milk other vegetables rolls/buns soda yogurt
#> 2513 1903 1809 1715 1372
#> bottled water root vegetables tropical fruit
#> 1087 1072 1032
#>
#> Distribution of itemset length:
#>
#> 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
#> 2159 1643 1299 1005 855 645 545 438 350 246 182 117 78 77 55 46 29 14 14 9
#> 21 22 23 24 26 27 28 29 32
#> 11 4 6 1 1 1 1 3 1
#>
#> Min. 1st Qu. Median Mean 3rd Qu. Max.
#> 1.000 2.000 3.000 4.409 6.000 32.000All classes come with base plotting and ggplot2 methods. Both plot() and qplot() only need to be supplied a valid object to work, however qplot() is more flexible and can sometimes be supplied with additional arguments like color alpha.
plot(Transactions)
qplot(Transactions)
plot(Frequent)
qplot(Frequent, type = "scatter", col = "red", alpha = 0.1)
hist(Frequent)
qplot(Frequent, type = "hist")
plot(Rules)
qplot(Rules)
There is a set of convenience functions to access information about the rules quality directly.
support(Frequent)[1:5]
#> [1] 0.05897306 0.09395018 0.02602949 0.02582613 0.04290798
support(Rules)[1:5]
#> [1] 0.01647178 0.02053889 0.01921708 0.02694459 0.02989324
confidence(Rules)[1:5]
#> [1] 0.2793103 0.3482759 0.3258621 0.2867965 0.3181818
lift(Rules)[1:5]
#> [1] 1.443519 1.363029 1.771616 1.482209 1.245252
leverage(Rules)[1:5]
#> [1] 0.005060933 0.005470332 0.008369877 0.008765919 0.005887463In this example we will use the Epub dataset containing the download history of documents from the electronical publiation platform of the Vienna University of Economics and Business Administration.
If you want to find association rules with minimal support of 0.0009 and minimal confidence of 0.1 run:
rules <- AssociationRules(Epub, 0.0009, 0.1)If you use qplot you can see that there are some rules with extremely high lift values.
qplot(rules)
Now imagine you are only interested in rules with a lift value above 300:
rules_pruned <- prune(rules, Lift = 300)
print(rules_pruned)
#> lhs rhs Support Confidence Lift Leverage
#> 1 {doc_6e9} => {doc_6e7} 0.001271537 0.7142857 321.0000 0.001267575
#> 2 {doc_6e7} => {doc_6e9} 0.001271537 0.5714286 321.0000 0.001267575
#> 3 {doc_6e9} => {doc_6e8} 0.001207960 0.6785714 333.5391 0.001204338
#> 4 {doc_506} => {doc_507} 0.001207960 0.6551724 303.0943 0.001203974
#> 5 {doc_6e8} => {doc_6e9} 0.001207960 0.5937500 333.5391 0.001204338
#> 6 {doc_507} => {doc_506} 0.001207960 0.5588235 303.0943 0.001203974
#> 7 {doc_6e8, doc_6e9} => {doc_6e7} 0.001080806 0.8947368 402.0947 0.001078118
#> 8 {doc_6e7, doc_6e9} => {doc_6e8} 0.001080806 0.8500000 417.8016 0.001078219
#> 9 {doc_6e7, doc_6e8} => {doc_6e9} 0.001080806 0.8095238 454.7500 0.001078429Similarly, you can also prune by confidence and have a look at the rules with confidence of 0.75 or above:
print(prune(rules, Confidence = 0.75))
#> lhs rhs Support Confidence Lift Leverage
#> 1 {doc_6e8, doc_6e9} => {doc_6e7} 0.001080806 0.8947368 402.0947 0.001078118
#> 2 {doc_6e7, doc_6e9} => {doc_6e8} 0.001080806 0.8500000 417.8016 0.001078219
#> 3 {doc_6e7, doc_6e8} => {doc_6e9} 0.001080806 0.8095238 454.7500 0.001078429