Handling a feature with multiple categorical values for the same instance value

Question

I have data in the following form:

table 1
id, feature1, predict
1, xyz,yes
2, abc, yes

table2
id, feature2
1, class1
1, class2
1, class3
2, class2

I could perform a one many join and train on the resultant set- which is one way to go about it. But If I rather wanted to maintain the length of the resultant set equal length of table 1, what is the technique?

Answer 1

One possible approach is to perform an encoding, where each level of the feature2 corresponds to a new feature (column). This way you may describe the 1:N relation between the feature 1 and 2

Here a small example in R

> table1  <- data.frame(id = c(1,2),  feature1 = c("xyz","abc"), predict = c(T,T))
> table2  <- data.frame(id = c(1,1,1,2), feature2 = c("class1", "class2", "class3", "class2"))
> 
> ## encoding
> table(table2)
   feature2
id  class1 class2 class3
  1      1      1      1
  2      0      1      0

The new object contains the (now unique) id and setting of the feature2. You need only to merge (join) the result to the table1 (basically same task a DB join - which variance: inner, outer or full depends on your requirements).