1 / 16

K-Means Lab

K-Means Lab. Start with 1D Data. # input data: https://cse.sc.edu/~rose/587/CSV/state_income.csv # 1) use wget to copy the dataset to the vm # 2) use GUI import from Text ( b ase ) to import state_income.csv #Sort data tmp = sort(state_income$V2 ) #Visualize data plot( tmp )

vcyr
Télécharger la présentation

K-Means Lab

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. K-Means Lab

  2. Start with 1D Data # input data: https://cse.sc.edu/~rose/587/CSV/state_income.csv # 1) use wget to copy the dataset to the vm # 2) use GUI import from Text (base) to import state_income.csv #Sort data tmp = sort(state_income$V2) #Visualize data plot(tmp) # create 4 clusters kmeans=kmeans(tmp,4,15) #Visualize the cluster centers points(kmeans$centers, col = 1:4, pch=20)

  3. Higher Dimensional Data # input data: https://cse.sc.edu/~rose/587/CSV/iris.csv # 1) use wget to copy to vm # 2) use GUI import from Text (base) to import iris.csv # Lets start with 2 dimensions: visualize data plot(iris[,1:2]) # create 4 clusters kmeans=kmeans(iris[,1:2],4,15) #Visualize the cluster centers points(kmeans$centers, col = 1:4, pch=20)

  4. # examine kmeans object kmeans # note: clustering statistics • cluster sizes • cluster means • clustering vector • within cluster sum of squares

  5. Plot Results # Visualize clusters plot(iris[,1:2], col=kmeans$cluster) # Visualize the cluster centers points(kmeans$centers, col = 1:4, pch=20)

  6. Best K? # Is 4 the best number of clusters? # explore different number of clusters withinSumSqrs = numeric(20) for (k in 1:20) withinSumSqrs[k] = sum(kmeans(iris[,1:2],centers=k)$withinss) #Visualize within cluster sum of square plot(1:20, withinSumSqrs, type="b", xlab="# Clusters", ylab="Within sum of square")

  7. Best K? Elbow Method • Look for the “elbow” • We want a small k that has a low WSS value • What does a low WSS value indicate?

  8. Higher Dimensional Data # Lets consider 3 dimensions: visualize data plot(iris[,1:3]) # create 4 clusters kmeans=kmeans(iris[,1:3],4,15) # Visualize the clusters plot(iris[,1:3],col=kmeans$cluster)

  9. Higher Dimensional Data # Examine a single 2D projection plot(iris[,1:2], col=kmeans$cluster) #Visualize the cluster centers points(kmeans$centers, col = 1:4, pch=20) # Examine another single 2D projection plot(iris[,1:3], col=kmeans$cluster) #Visualize the cluster centers points(kmeans$centers[,c(1,3)], col = 1:4, pch=20)

  10. Best K? # Is 4 the best number of clusters? # explore different number of clusters withinSumSqrs = numeric(20) for (k in 1:20) withinSumSqrs[k] = sum(kmeans(iris[,1:3],centers=k)$withinss) #Visualize within cluster sum of square plot(1:20, withinSumSqrs, type="b", xlab="# Clusters", ylab="Within sum of square")

  11. Higher Dimensional Data # Finally consider 4 dimensions: visualize data plot(iris[,1:4]) # create 4 clusters kmeans=kmeans(iris[,1:4],4,15) # Visualize the clusters plot(iris[,1:4], col=kmeans$cluster)

  12. Normalization? # notice that the columns are not the same scale > summary(iris[,1:4]) Sepal.LengthSepal.WidthPetal.LengthPetal.Width Min. :4.300 Min. :2.000 Min. :1.000 Min. :0.100 1st Qu.:5.100 1st Qu.:2.800 1st Qu.:1.600 1st Qu.:0.300 Median :5.800 Median :3.000 Median :4.350 Median :1.300 Mean :5.843 Mean :3.054 Mean :3.759 Mean :1.199 3rd Qu.:6.400 3rd Qu.:3.300 3rd Qu.:5.100 3rd Qu.:1.800 Max. :7.900 Max. :4.400 Max. :6.900 Max. :2.500 #Which will have more influence: petal width or petal length?

  13. Normalization? # Let’s shift and scale the data to be in the range [0,1] #  each dimension should be able to exert equal influence # function to “shift” a vector # subtract the min value so that it ranges from 0 to ?? myShift = function(x) { x - min(x, na.rm=TRUE)} # apply this function to one column myShift(iris[,3]) # apply this function to multiple columns as.data.frame(lapply(iris[,1:4], myShift)) #verify with summary statistics summary(as.data.frame(lapply(iris[,1:4], myShift)))

  14. Normalization? # Still need to scale data # function to “scale” a vector myScale = function(x) { max(x,na.rm=TRUE) - min(x,na.rm=TRUE)} # apply this function to one column myScale(iris[,3]) # apply this function to multiple columns as.data.frame(lapply(iris[,1:4], myScale))

  15. Normalization? # Put it all together # myShift= function(x) { x - min(x, na.rm=TRUE)} # myScale = function(x) { max(x,na.rm=TRUE) - min(x,na.rm=TRUE)} myNorm= function(x) { myShift(x)/myScale(x) } # apply this function to a column myNorm(iris[,3]) # apply this function to multiple columns tmp = as.data.frame(lapply(iris[,1:4], myNorm)) #verify with summary statistics summary(as.data.frame(lapply(iris[,1:4], myNorm)))

  16. Higher Dimensional Data # visualize “normalized” data plot(tmp[,1:4]) # try from 1 to 20 clusters for (k in 1:20) withinSumSqrs[k] = sum(kmeans(tmp[,1:4],centers=k)$withinss) #Visualize within cluster sum of square plot(1:20, withinSumSqrs, type="b", xlab="# Clusters", ylab="Within sum of square")

More Related