| 1 | library(agghoo) |
| 2 | |
| 3 | standardCV <- function(data, target, task = NULL, gmodel = NULL, params = NULL, |
| 4 | loss = NULL, CV = list(type = "MC", V = 10, test_size = 0.2, shuffle = TRUE) |
| 5 | ) { |
| 6 | if (!is.null(task)) |
| 7 | task = match.arg(task, c("classification", "regression")) |
| 8 | if (is.character(gmodel)) |
| 9 | gmodel <- match.arg(gmodel, c("knn", "ppr", "rf", "tree")) |
| 10 | if (is.numeric(params) || is.character(params)) |
| 11 | params <- as.list(params) |
| 12 | if (is.null(task)) { |
| 13 | if (is.numeric(target)) |
| 14 | task = "regression" |
| 15 | else |
| 16 | task = "classification" |
| 17 | } |
| 18 | |
| 19 | if (is.null(loss)) { |
| 20 | loss <- function(y1, y2) { |
| 21 | if (task == "classification") { |
| 22 | if (is.null(dim(y1))) |
| 23 | mean(y1 != y2) |
| 24 | else { |
| 25 | if (!is.null(dim(y2))) |
| 26 | mean(rowSums(abs(y1 - y2))) |
| 27 | else { |
| 28 | y2 <- as.character(y2) |
| 29 | names <- colnames(y1) |
| 30 | positions <- list() |
| 31 | for (idx in seq_along(names)) |
| 32 | positions[[ names[idx] ]] <- idx |
| 33 | mean(vapply( |
| 34 | seq_along(y2), |
| 35 | function(idx) sum(abs(y1[idx,] - positions[[ y2[idx] ]])), |
| 36 | 0)) |
| 37 | } |
| 38 | } |
| 39 | } |
| 40 | else |
| 41 | mean(abs(y1 - y2)) |
| 42 | } |
| 43 | } |
| 44 | |
| 45 | n <- nrow(data) |
| 46 | shuffle_inds <- NULL |
| 47 | if (CV$type == "vfold" && CV$shuffle) |
| 48 | shuffle_inds <- sample(n, n) |
| 49 | get_testIndices <- function(v, shuffle_inds) { |
| 50 | if (CV$type == "vfold") { |
| 51 | first_index = round((v-1) * n / CV$V) + 1 |
| 52 | last_index = round(v * n / CV$V) |
| 53 | test_indices = first_index:last_index |
| 54 | if (!is.null(shuffle_inds)) |
| 55 | test_indices <- shuffle_inds[test_indices] |
| 56 | } |
| 57 | else |
| 58 | test_indices = sample(n, round(n * CV$test_size)) |
| 59 | test_indices |
| 60 | } |
| 61 | list_testinds <- list() |
| 62 | for (v in seq_len(CV$V)) |
| 63 | list_testinds[[v]] <- get_testIndices(v, shuffle_inds) |
| 64 | |
| 65 | gmodel <- agghoo::Model$new(data, target, task, gmodel, params) |
| 66 | best_error <- Inf |
| 67 | best_model <- NULL |
| 68 | for (p in seq_len(gmodel$nmodels)) { |
| 69 | error <- 0 |
| 70 | for (v in seq_len(CV$V)) { |
| 71 | testIdx <- list_testinds[[v]] |
| 72 | dataHO <- data[-testIdx,] |
| 73 | testX <- data[testIdx,] |
| 74 | targetHO <- target[-testIdx] |
| 75 | testY <- target[testIdx] |
| 76 | if (!is.matrix(dataHO) && !is.data.frame(dataHO)) |
| 77 | dataHO <- as.matrix(dataHO) |
| 78 | if (!is.matrix(testX) && !is.data.frame(testX)) |
| 79 | testX <- as.matrix(testX) |
| 80 | model_pred <- gmodel$get(dataHO, targetHO, p) |
| 81 | prediction <- model_pred(testX) |
| 82 | error <- error + loss(prediction, testY) |
| 83 | } |
| 84 | if (error <= best_error) { |
| 85 | newModel <- list(model=model_pred, param=gmodel$getParam(p)) |
| 86 | if (error == best_error) |
| 87 | best_model[[length(best_model)+1]] <- newModel |
| 88 | else { |
| 89 | best_model <- list(newModel) |
| 90 | best_error <- error |
| 91 | } |
| 92 | } |
| 93 | } |
| 94 | best_model[[ sample(length(best_model), 1) ]] |
| 95 | } |
| 96 | |
| 97 | compareToCV <- function(df, t_idx, task=NULL, rseed=-1, verbose=TRUE, ...) { |
| 98 | if (rseed >= 0) |
| 99 | set.seed(rseed) |
| 100 | if (is.null(task)) |
| 101 | task <- ifelse(is.numeric(df[,t_idx]), "regression", "classification") |
| 102 | n <- nrow(df) |
| 103 | test_indices <- sample( n, round(n / ifelse(n >= 500, 10, 5)) ) |
| 104 | data <- as.matrix(df[-test_indices,-t_idx]) |
| 105 | target <- df[-test_indices,t_idx] |
| 106 | test <- as.matrix(df[test_indices,-t_idx]) |
| 107 | a <- agghoo(data, target, task, ...) |
| 108 | a$fit() |
| 109 | if (verbose) { |
| 110 | print("Parameters:") |
| 111 | print(unlist(a$getParams())) |
| 112 | } |
| 113 | pa <- a$predict(test) |
| 114 | err_a <- ifelse(task == "classification", |
| 115 | mean(pa != df[test_indices,t_idx]), |
| 116 | mean(abs(pa - df[test_indices,t_idx]))) |
| 117 | if (verbose) |
| 118 | print(paste("error agghoo:", err_a)) |
| 119 | # Compare with standard cross-validation: |
| 120 | s <- standardCV(data, target, task, ...) |
| 121 | if (verbose) |
| 122 | print(paste( "Parameter:", s$param )) |
| 123 | ps <- s$model(test) |
| 124 | err_s <- ifelse(task == "classification", |
| 125 | mean(ps != df[test_indices,t_idx]), |
| 126 | mean(abs(ps - df[test_indices,t_idx]))) |
| 127 | if (verbose) |
| 128 | print(paste("error CV:", err_s)) |
| 129 | invisible(c(err_a, err_s)) |
| 130 | } |
| 131 | |
| 132 | library(parallel) |
| 133 | compareMulti <- function(df, t_idx, task = NULL, N = 100, nc = NA, ...) { |
| 134 | if (is.na(nc)) |
| 135 | nc <- detectCores() |
| 136 | compareOne <- function(n) { |
| 137 | print(n) |
| 138 | compareToCV(df, t_idx, task, n, verbose=FALSE, ...) |
| 139 | } |
| 140 | errors <- if (nc >= 2) { |
| 141 | mclapply(1:N, compareOne, mc.cores = nc) |
| 142 | } else { |
| 143 | lapply(1:N, compareOne) |
| 144 | } |
| 145 | print("error agghoo vs. cross-validation:") |
| 146 | Reduce('+', errors) / N |
| 147 | } |