R nearest neighbour (knn) demo.

knn

cat("\014")
library(class)

# I like boolean support.
no  <- 0
yes <- 1

# Entry to clasify
test <- c(
    "cat", # The correct class
    4,     # Number of legs
    no,    # has tail
    yes,   # has lungs
    yes    # has fur
)

test2 <- c(
  "fish", # The correct class
  0,      # Number of legs
  yes,    # has tail
  no,     # has lungs
  no      # has fur
)

# Merge them into a matrix, we'll test multiple.
alltests <- rbind(test, test2)

# Training data set
data <- rbind(
#   class        legs  tail    lungs   fur 
    c("human",   2,    no,     yes,    no),
    c("fish",    0,    yes,    no,     no),
    c("cat",     4,    yes,    yes,    yes),
    c("spider",  8,    no,     no,     yes),
    c("spider",  6,    no,     no,     no),   # rare spider
    c("wasp",    6,    no,     no,     yes),
    c("alien",   5,    no,     no,     yes),  # alien variant 1
    c("alien",   5,    yes,    no,     no)    # alien variant 2
) 


# Extract columns
class <- data[, 1]             # First column contains class names
train <- data[, 2:ncol(data)]  # Other columns, discriptive data


# (CNN) Self test, each pattern is used as input. If it cannot predict
# itself, it is dropped. E.g., when a cat pattern cannot predict
# it's a cat, it's dropped as a rule.
keep <- condense(train, class)

# Remove entries that do not change the outcome.
keep2 <- reduce.nn(train, keep, class)

# Nearest neighbour
print("Statistically speaking, you are a:")
knn(train[keep, , drop=FALSE], alltests[,2:length(test)], class[keep], k=1, TRUE)

if(length(keep2) > 0) {
  print("After reducing the training data, you are a:")
  knn(train[keep2, , drop=FALSE], alltests[,2:length(test)], class[keep2], k=1, TRUE)
} else {
  print("Reducing the training data resulted into an empty set:");
  keep2
}

print("Information provided:")
print(alltests)