install.packages(vegan)
library(vegan)

getwd()
setwd("c:/DANKA HLAVNY ADRESAR/Vyuka/Bi5980 Biodiverzita/Biodiverzita vyuka 2016")

pilatkoviti <- read.csv("05_2016_Data Pilatkoviti.csv",sep = ";" ,dec = ",", header = T, row.names = 1)
pilatkoviti

# nazvy riadkov su povodny prvy stlpec 

pilatkoviti.bray <- vegdist(pilatkoviti, "bray", diag = T, upper = T) 
pilatkoviti.bray

?hclust
hclust(pilatkoviti.bray, method = "complete")

pilatkoviti.bray.complete <- hclust(pilatkoviti.bray, method = "complete") 

plot(pilatkoviti.bray.complete)
plot(pilatkoviti.bray.complete, main = "Bray-Curtis, Complete linkage")
plot(pilatkoviti.bray.complete, labels = rownames(pilatkoviti.bray), hang = -1, main = "Bray-Curtis, Complete linkage")
plot(pilatkoviti.bray.complete, labels = rownames(pilatkoviti.bray), hang = -1, axes = FALSE, main = "Bray-Curtis, Complete linkage")
plot(pilatkoviti.bray.complete, labels = rownames(pilatkoviti.bray), hang = -1, axes = FALSE, ann = FALSE)

pilatkoviti.bray.ward <- hclust(pilatkoviti.bray, method = "ward.D2")  # ... ale tu pozor, lebo Ward sa nemá používať s Bray-Curtis
# Ward method should not be combined with distance measures, which are not strictly metric, which is e.g. popular Bray-Curtis distance

plot(pilatkoviti.bray.ward, labels = rownames(pilatkoviti.bray), hang = -1, main = "Bray-Curtis, Ward")
plot(pilatkoviti.bray.ward, labels = rownames(pilatkoviti.bray), hang = -1, axes = FALSE, ann = FALSE)


--------------------------------------------------------------------------------------------------------------

# oznaceni shluku v dendrogramu, budem tam chciet oznacit 2 shkuky, alebo 3 shluky

plot (pilatkoviti.bray.complete, labels = rownames(pilatkoviti.bray), hang = -1, axes = FALSE, main = 'Complete linkage')

rect.hclust (pilatkoviti.bray.complete, k = 2)
rect.hclust (pilatkoviti.bray.complete, k = 3, border = "blue")


prislusnost.3shluky <- cutree (pilatkoviti.bray.complete, k = 3)   # vrati prislusnost objektu ke shlukum 
prislusnost.3shluky

--------------------------------------------------------------------------------------------------------------

# nakreslime 3 dendrogramy na jeden obrazok - single linkage, average linkage, complete linkage

pilatkoviti.bray.single <- hclust(pilatkoviti.bray, method = "single") 
pilatkoviti.bray.average <- hclust(pilatkoviti.bray, method = "average") 
pilatkoviti.bray.complete <- hclust(pilatkoviti.bray, method = "complete") 

par (mfrow = c (1,3))   # nakresli vsetky dendrogramy do jedneho obrazku

plot (pilatkoviti.bray.single, main = 'Single linkage')
plot (pilatkoviti.bray.average, main = 'Average linkage')
plot (pilatkoviti.bray.complete, main = 'Complete linkage')

# alebo ak chceme mat "nozicky" od zaciatku, tak takto:

par (mfrow = c (1,3))
plot (pilatkoviti.bray.single, labels = rownames(pilatkoviti.bray), hang = -1, axes = FALSE, main = 'Single linkage')
plot (pilatkoviti.bray.average, labels = rownames(pilatkoviti.bray), hang = -1, axes = FALSE, main = 'Average linkage')
plot (pilatkoviti.bray.complete, labels = rownames(pilatkoviti.bray), hang = -1, axes = FALSE, main = 'Complete linkage')


# aby nechcelo kreslit vsetky dalsie obrazky po troch, tak ... par(mfrow = c(1,1))

--------------------------------------------------------------------------------------------------------------
POZNAMKA:

agnes (library cluster) ... obsahuje 6 shlukovacich algoritmu, nektere z nich nejsou obsazeny ve funkci 'hclust'.
pak pri kresleni dendrogramu je vhodne nejdrive transformovat objekt z agnes jako objekt z hclust (pouzit as.hclust)

--------------------------------------------------------------------------------------------------------------
POZNAMKA:
nehierarchicke shlukovani - metoda k-means

cluster.kmeans <- kmeans (pilatkoviti, centers = 3)
cluster.kmeans$cluster

------------------------------------ K-means clustering - KORYSE ---------------------------------------------

setwd("c:/DANKA HLAVNY ADRESAR/Vyuka/Bi5980 Biodiverzita/Biodiverzita vyuka 2016")
koryse <- read.csv("06_2016_Data_Plankton Koryse.csv",sep = ";" ,dec = ",", header = T, row.names = 1)
koryse

koryse.kmeans3 <- kmeans (koryse, centers = 3)
koryse.kmeans3
koryse.kmeans3$cluster
koryse.kmeans3$withinss

# mozne zopakovat k-means clustering pro pocet shluku 2, 3, 4, 5, ... a urcit si sucet vnutroskupinovych sum ctvercu

withinss2 <- sum(koryse.kmeans2$withinss)
withinss3 <- sum(koryse.kmeans3$withinss)
withinss4 <- sum(koryse.kmeans4$withinss)
withinss5 <- sum(koryse.kmeans5$withinss)

koryse.kmeans1 <- kmeans (koryse, centers = 1)
koryse.kmeans1$withinss

withinss <- koryse.kmeans1$withinss
withinss

# rozptyl pri jednom shluku
# je to to same, jako rozptyl sloupcu vynasobeny poctem radku snizeny o jednu, tj.: (nrow(koryse)-1)*sum(apply(koryse,2,var)) 

          
  for (i in 2:15) withinss[i] <- sum(kmeans(koryse, 
                                       centers=i)$withinss)
plot(1:15, withinss, type="b", xlab="Number of Clusters", ylab="Within groups sum of squares")

# z tohoto grafu - scree plot - muzeme odhadnout "spravny pocet shluku", tj. pocet pri "ohybu".