arules.Rmd

---
title: "Association Rules"
output:
  pdf_document: default
  html_notebook: default
---

```{r}
library(arules)
library(arulesViz)
```

# Load the data

```{r}
data("Groceries")
```

# Explore and Visualize

```{r}
str(Groceries)
dim(Groceries)
summary(Groceries)
```


```{r}
inspect(Groceries[1:20,])
```

```{r}
# Which items contain the word fish?
colnames(Groceries)[grep("fish", colnames(Groceries))]

# Which items contain the word wine?
colnames(Groceries)[grep("wine", colnames(Groceries))]
```


```{r}
itemFrequencyPlot(Groceries,
                  type="relative",
                  topN=15, # can be changed to the number of interest
                  horiz=FALSE,
                  col='steelblue3',
                  xlab='',
                  main='Item frequency, relative')
```




# Simple contingency table

Sometimes, just looking at a simple contingency table could shed some light.

```{r}
tbl <- crossTable(Groceries)
tbl[1:13,1:13]
```


How many people buy canned bear and bottled beer at the same time? And other products.

```{r}
tbl['canned beer','bottled beer']
tbl['beef', 'red/blush wine']

# What do people buy most often with their wine?
sort(tbl[,'red/blush wine'], decreasing = TRUE)
```


# Build the association rules

```{r}
rules <- apriori(Groceries, parameter = list(supp = 0.001, conf = 0.1, target = "rules"))
summary(rules)
```

```{r}
# List the top 40 rules, ordered by lift
inspect(head(rules, n=40, by = "lift"))
```


```{r}
# Calculate another interestingness measure: conviction
quality(rules) <- cbind(quality(rules),
                  conviction = interestMeasure(rules, measure="conviction", transactions = Groceries))
```

```{r}
# List the top 40 rules, ordered by conviction
inspect(head(rules, n=40, by = "conviction"))
```

```{r}

# First, just plot them all:
# Note that the windows() call makes a new window pop up, which is what we want for interactivity.
windows(); plot(rules, interactive = T)

subrules <- rules[quality(rules)$confidence > 0.3]
plot(subrules, method="matrix", measure="lift")
plot(subrules, method="matrix3D", measure="lift")
plot(subrules, method="graph", measure="lift")
```


## Example

```{r}
inspect(sort(subset(rules, 
                    subset=rhs %in% 'bottled beer' & confidence > .7),
                    by = 'lift',
                    decreasing = T))

```


## Example

Both “whole milk” and “yogurt” must be present and rule’s confidence must be higher than .9

```{r}
inspect(subset(rules, subset=items %ain% c("whole milk","yogurt") & confidence >.90))
```


## Example
“Bread” must be present in lhs: any type of “bread” – “white bread”, “brown bread” – both qualify. “Whole milk” must be present in rhs “as is”. confidence of the rule must be higher than .9


```{r}
inspect(subset(rules, subset= lhs %pin% "bread" & rhs %in% "whole milk" & confidence > .9))
```

# Making up some transaction data

For use in class, I wanted to make up some data to describe different situations.


Looking for a rule with high support, high conf, and high lift. E.g.:
- Support of {A} is 70%
- Support of {B} is 50%
- Support of {A,B} is 50%
- Confidence of {A}->{B} is 100% 
- Lift is therefore 2.0

Looking for a rule with high support, high conf, and low lift. I.e.:
- Support of {A} is 100%
- Support of {B} is 50%
- Support of {A,B} is 50%
- Confidence of {A}->{B} is 50% 
- Lift is therefore 1.0

```{r}
a_list <- list(
c("milk","bread","carrot"),
c("milk","bread"),
c("milk","bread","meat"),
c("milk","diamonds"),
c("milk","bread","meat","diamonds"),
c("milk", "meat"),
c("milk"),
c("bread"),
c("bread", "meat"),
c("milk","bread", "diamonds"),
c("milk","bread", "carrot"),
c("milk","bread"),
c("diamonds", "meat"),
c("diamonds", "meat"),
c("diamonds", "meat"),
c("diamonds", "meat"),
c("diamonds", "meat"),
c("diamonds", "meat"),
c("diamonds", "meat"),
c("diamonds", "meat")
)
## set transaction names
names(a_list) <- paste("Tr",c(1:5), sep = "")
## coerce into transactions
trans1 <- as(a_list, "transactions")
## analyze transactions
summary(trans1)
#image(trans1)
rules <- apriori(trans1, parameter = list(supp = 0.001, conf = 0.001, target = "rules"))
```


```{r}
inspect(sort(subset(rules,  
                    subset=lhs %in% 'bread' ),
                    by = 'lift',
                    decreasing = T))
```

Let's look at milk-> bread. Expect high, high, high
```{r}
inspect(sort(subset(rules, 
                    subset=lhs %in% 'milk' & rhs %in% 'bread'),
                    by = 'lift',
                    decreasing = T))
```

```{r}
inspect(sort(subset(rules, 
                    subset=lhs %in% 'diamonds' & confidence > .0),
                    by = 'lift',
                    decreasing = T))
```

```{r}
inspect(head(rules, n=42, by = "support"))
```

```{r}
inspect(head(rules, n=42, by = "confidence"))
```

```{r}
inspect(head(rules, n=42, by = "lift"))
```