在R中分类数据集

Question

I am having an issue with categorizing a dataset.

The dataset is an matrix in which the rows are observations, and the columns are the features. Each features value is between 0 - 1. The dataset is used for training purposes, and since the method I am going to use is vary sensitive to small variation, the dataset has to formatted to not be sensitive.

My idea was that instead of providing the raw data i want to bin the feature values into bins according to their numeric value, and provide the middle value of the bin as the training data for the training.

Ex. bins being (1-2,2-3,3-4,4-5,5-6,6-7,7-8,8-9,9-10)

dataset #original dataset
>         [,1] [,2] [,3] [,4] [,5]
[1,]    8.1    5.3   10    4.4    4.6
[2,]    5.2   10    3.2    9.3    3.5
[3,]    7.3    1.6    9    8.9    8.4
[4,]    6.4    2.8    8    6.5    9.3
[5,]   10    4.3    2.2    1.1    5.3

transformed_dataset #binned dataset


>         [,1] [,2] [,3] [,4] [,5]
[1,]    8.5    5.5   9.5   4.5    4.5
[2,]    5.5   9.5   3.5    9.5   3.5
[3,]    7.5    1.5   8.5    8.5    8.5
[4,]    6.5    2.5   8.5    6.5    9.5
[5,]    9.5    4.5  2.5    1.5    5.5

I am not sure on how i should bin the data like this, and give it as input for naiveBayes from the library("lattice") . I know that signif is capable of rounding the value to a giving number of digits, and thus "bining it", but i can't actually determine the number of bins.

Binning seems a way to improve the classification. But how to provide it as a input, that I am not certain of.

Updata about the data.frame

I think i forgot to mention it, but the data is stored in a data.frame , and the way i access the data is by $data. the data.frame all provide labels for each observation which can be accessed by $labels.

Answer 1

Hm. You may have some trouble with the data types here, because matrix class does not work well with factors - and the binning intervals are best described by factors.

In order to do the binning, you can use the cut function from the base R installation, such as:

> data <- c(1,2,4,1,5,3,3,5,2,2,5,5,5,7,8,9,5,3,2,6,8,9,3,1)
> breaks <- c(0, 3, 6, 9)
> cut(data, breaks=breaks)
 [1] (0,3] (0,3] (3,6] (0,3] (3,6] (0,3] (0,3] (3,6] (0,3] (0,3] (3,6] (3,6]
 [13] (3,6] (6,9] (6,9] (6,9] (3,6] (0,3] (0,3] (3,6] (6,9] (6,9] (0,3] (0,3]
 Levels: (0,3] (3,6] (6,9]

Or, using the left interval:

> cut(data, breaks=breaks, right=FALSE)
 [1] [0,3) [0,3) [3,6) [0,3) [3,6) [3,6) [3,6) [3,6) [0,3) [0,3) [3,6) [3,6)
[13] [3,6) [6,9) [6,9) <NA>  [3,6) [3,6) [0,3) [6,9) [6,9) <NA>  [3,6) [0,3)
Levels: [0,3) [3,6) [6,9)

Notice that the breaks you provide should cover the entire dataset, or else you will get some NA s.

Answer 2

A simple solution could be like this:

d <- matrix(c(8.1, 5.3, 10, 4.4, 4.6,
              5.2, 10, 3.2, 9.3, 3.5,
              7.3, 1.6, 9, 8.9, 8.4,
              6.4, 2.8, 8, 6.5, 9.3,
              10, 4.3, 2.2, 1.1, 5.3), nrow=5, ncol=5, byrow=TRUE)

d <- as.data.frame(apply(d, 2, function(column) {
  as.factor(round(column+0.5)-0.5)
}))

Leading to results:

> d
   V1  V2  V3  V4  V5
1 8.5 5.5 9.5 4.5 4.5
2 5.5 9.5 3.5 9.5 3.5
3 7.5 1.5 9.5 8.5 8.5
4 6.5 2.5 7.5 6.5 9.5
5 9.5 4.5 2.5 1.5 5.5

After the transformation, the columns of your dataset are factors, meaning that naiveBayes will not treat them as numeric but as categorical variables.

> class(d[,1])
[1] "factor"
> levels(d[,1])
[1] "5.5" "6.5" "7.5" "8.5" "9.5"

Note that the trick of adding and removing 0.5 will fail if you have any value equal to 0 - it will assign it to level "0" instead of "0.5". You could solve it adding this line to the function:

column[which(column == 0)] <- 0.5

Hope it helps.