# model 2: filmy zamiast użytkowników
# if DEBUG == 1
#X11()
# endif
library(reshape2)

friends = c()
#długość wektora (sqrt(x1^2 + x2^2 + ...))
norm_vec <- function(x) sqrt(sum(x^2, na.rm=TRUE))

#podobieństwo kosinusowe
cosine <- function(user1, user2) {
  top = sum(user2 * user1, na.rm=TRUE)
  bot = norm_vec(user1) * norm_vec(user2)
  return (top/bot)
}

# user_n szukany użytkownik
# film_n szukany film
# nr = ilość najlepszych podobieństw do analizy
recommend <- function(user, film, nr, mx) {
  userrow = mx[rownames(mx) == user, ]
  tryCatch(
  {
  has_film = mx[!is.na(mx[,film]), ]
  has_film = has_film[rownames(has_film) != user, ]

  t = apply(has_film, 1, cosine, userrow)
  ord = sort(t, decreasing=TRUE, index.return = TRUE) $ix
  sorted = has_film[ord, ][1:min(nr, nrow(has_film)), ]
  friends[length(friends) + 1] <<- min(nr, nrow(has_film))
  return (sorted)
  }, error = function(e) {
        }
  )
return(NA)
}

pred_avg <- function(mx, user, film) {
    user = toString(user)
    film = toString(film)
    tmp = recommend(user, film, 20, mx)
    if (is.na(tmp)) {
        return(mean(mx[row.names(mx) == user, ], na.rm = TRUE))
        #return(3.5)
    }
    row_means = rowMeans(tmp, na.rm = TRUE)
    user_mean = data.frame(mean(mx[row.names(mx) == user, ], na.rm = TRUE))
    colnames(user_mean) = list("avg")
    regression_table = data.frame(row_means, tmp[, colnames(tmp) == film])
    colnames(regression_table) = list("avg", "rating")
    regression = lm(rating ~ avg, data=regression_table)
    plot(regression)
    return (predict(regression, user_mean))
}

pred_2 <- function(mx, user, film) {
  user = toString(user)
  film = toString(film)
  tmp = recommend(user, film, 20, mx)
  if (is.na(tmp)) {
      return(mean(mx[row.names(mx) == user, ], na.rm = TRUE))
      #return(3.5)
  }
  row_means = rowMeans(tmp, na.rm = TRUE)
  user_mean = data.frame(mean(mx[row.names(mx) == user, ], na.rm = TRUE))
  colnames(user_mean) = list("avg")
  regression_table = data.frame(row_means, tmp[, colnames(tmp) == film])
  colnames(regression_table) = list("avg", "rating")
  regression = lm(rating ~ avg, data=regression_table)
  print(c(regression_table["avg"]))
  print(c(regression_table["rating"]))
  # if DEBUG == 1
  #plot(unlist(regression_table["avg"]), unlist(regression_table["rating"]), xlim=c(2,5), ylim=c(2,5))
  #abline(regression, lwd=2)
  #Sys.sleep(1)
  # endif
  return (predict(regression, user_mean))
}

prog <- function() {
  set.seed(123)
  MyData <- read.csv(file="~/Desktop/mow-dane/ratings.csv", header=TRUE, sep=",")
  MyData <- MyData[1:3]

  test_indexes = sample(nrow(MyData), size = nrow(MyData)/1000)
  TestData <- MyData[test_indexes, ]
  MyData <- MyData[-test_indexes,]

  MyMatrix = acast(MyData, userId ~ movieId , value.var="rating")
  TestMatrix = acast(TestData, userId ~ movieId , value.var="rating")

  vect = c()
  vect_real = c()
  for(i in 1:nrow(TestData)) {
    row <- matrix(TestData[i,])
    predicted_val = pred_2(MyMatrix, row[1], row[2])
    vect[i] <- predicted_val
    vect_real[i] <- row[3]
    print(i)
  }
  vect_real = unlist(vect_real)
  print(vect)
  print(vect_real)

  diffs = abs(vect - vect_real)
  print(diffs)

  avg = sum(diffs)/length(vect)
  print(avg)

  pred2 = round(vect*2)/2
  matched = which(pred2 == vect_real)
  print(matched)

  print(friends)
  avg_friends = sum(friends)/length(friends)
  print(avg_friends)
}

prog()