library(tm)
library(lda)
library(topicmodels)
library(ROCR)
library(parallel)
### console --------------------------------------------------------------------
options(max.print = 100)
### load data ------------------------------------------------------------------
data.corpus <- readRDS("./saved/corpus.rds")
data <- data.frame(
meta(data.corpus, "label", type="indexed"),
meta(data.corpus, "file",  type="indexed")
)
data.dtm <- readRDS("./saved/dtmatrix.rds")
data.dtm <- dtm2ldaformat(data.dtm, omit_empty = FALSE)
### training & testing datasets ------------------------------------------------
trainset.files <- c(
"T6_2013_Queensland_Floods"   # 8732
, "T6_2013_Alberta_Floods"      # 9069
, "T26_2013_Manila_floods"      # 1167
, "T26_2013_Sardinia_floods"    #  180
, "T26_2012_Philipinnes_floods" # 1318
)
testset.files <- c(
"T26_2013_Colorado_floods"    # 1664
)
trainset.idx <- which(data$file %in% trainset.files)
testset.idx  <- which(data$file %in% testset.files)
cat(sprintf("train/test: %d / %d (%.1f%% / %.1f%%)",
length(trainset.idx), length(testset.idx),
length(trainset.idx)/(length(trainset.idx)+length(testset.idx))*100,
length(testset.idx) /(length(trainset.idx)+length(testset.idx))*100
))
### sLDA -----------------------------------------------------------------------
# trainset.size <- seq(1000, 20000, 1000)
# trainset.size <- c( seq(1000, 6800, 200), seq(9200, 20000, 200) )
trainset.size <- seq(1000, 20000, 200)
trainset.size <- rep(trainset.size, 20)
model.quality <- mclapply(trainset.size, mc.cores = 3, FUN = function(iter.size) {
# choose raindom records for the trainset of demanding size
# iter.size <- 1000
iter.idx <- sample(trainset.idx, iter.size, replace = FALSE)
# train and test model -------------------------------------------------------
topics.cnt <- 100
model <- slda.em(
documents = data.dtm$documents[iter.idx],
K = topics.cnt,
vocab = data.dtm$vocab,
num.e.iterations = 10,
num.m.iterations =  4,
alpha = 1.0, eta = 0.1,
annotations = data$label[iter.idx],
params = sample(c(0,1), topics.cnt, replace = TRUE),
variance = 0.25,
logistic = FALSE,    # default
lambda = 1.0,
regularise = FALSE,  # default
method="sLDA",
trace = 0L           # default
)
predict <- slda.predict(
documents = data.dtm$documents[testset.idx],
topics = model$topics,
model = model$model,
alpha = 1.0,
eta = 0.1,
num.iterations = 1000,   # default
average.iterations = 50, # default
trace = 0L               # default 0L
)
# accuracy metrics -----------------------------------------------------------
rocr.pred <- prediction(
predictions = predict,
labels      = data$label[testset.idx]
)
rocr.perf    <- performance(rocr.pred, measure = "acc", x.measure = "cutoff")
cutoff.index <- which.max(rocr.perf@y.values[[1]])
return(c(
"prec" = performance(rocr.pred, measure = "prec")@y.values[[1]][cutoff.index],
"rec"  = performance(rocr.pred, measure = "rec" )@y.values[[1]][cutoff.index],
"auc"  = performance(rocr.pred, measure = "auc" )@y.values[[1]]
))
})
model.quality <- do.call(rbind, model.quality)
model.quality <- as.data.frame.matrix(model.quality)
model.quality$size <- trainset.size
if (file.exists("./saved/training_size_slda.csv") == TRUE) {
result <- read.csv("./saved/training_size_slda.csv", header = TRUE)
result <- rbind(result, model.quality)
} else {
result <- model.quality
}
write.csv(result, file = "./saved/training_size_slda.csv", row.names = FALSE)
### ----------------------------------------------------------------------------
rm(
data, data.corpus, data.dtm,
testset.idx, testset.files, trainset.idx, trainset.files, trainset.size,
model.quality, result
)
### plot sLDA ------------------------------------------------------------------
result <- read.csv("./saved/training_size_slda.csv", header = TRUE)
p <- ggplot(
data = melt(result, id.vars = "size", measure.vars = c("prec", "rec", "auc")),
mapping = aes_string(x = "size", y = "value")
)
p <- p + stat_summary(fun.data = "median_hilow", geom = "line",   aes_string(linetype = "variable"))
p <- p + stat_summary(fun.data = "median_hilow", geom = "ribbon", aes_string(fill = "variable"), alpha = 0.1, show.legend = FALSE)
p <- p + theme_bw(base_size = 11)
p <- p + scale_x_continuous(breaks = seq(from=0, to=20000, by=2000))
p <- p + labs(title = "sLDA model", x = "training dataset size", y = "model evaluation")
p <- p + scale_fill_grey(name = "Performance", start = 0.2, end = 0.2)
p <- p + scale_linetype_discrete(name = "Performance",
breaks = c("prec",      "rec",    "auc"),
labels = c("Precision", "Recall", "AUC"))
p + theme(legend.position=c(.9, .2))
source('./04_01_trainset_size.R')
library(ggplot2)
library(reshape2)
library(Cairo)
### console --------------------------------------------------------------------
options(max.print = 100)
### load data ------------------------------------------------------------------
data.corpus <- readRDS("./saved/corpus.rds")
data <- data.frame(
meta(data.corpus, "label", type="indexed"),
meta(data.corpus, "file",  type="indexed")
)
data.dtm <- readRDS("./saved/dtmatrix.rds")
library(ggplot2)
library(reshape2)
library(Cairo)
### console --------------------------------------------------------------------
options(max.print = 100)
### plot SVM -------------------------------------------------------------------
result <- read.csv("./saved/training_size_svm.csv", header = TRUE)
p <- ggplot(
data = melt(result, id.vars = "size", measure.vars = c("prec", "rec", "auc")),
mapping = aes_string(x = "size", y = "value")
)
p <- p + stat_summary(fun.data = "median_hilow", geom = "line",   aes_string(linetype = "variable"))
p <- p + stat_summary(fun.data = "median_hilow", geom = "ribbon", aes_string(fill = "variable"), alpha = 0.1, show.legend = FALSE)
p <- p + theme_bw(base_size = 11)
p <- p + scale_x_continuous(breaks = seq(from=1000, to=20000, by=2000))
p <- p + labs(title = "SVM model", x = "training dataset size", y = "model evaluation")
p <- p + scale_fill_grey(name = "Performance", start = 0.2, end = 0.2)
p <- p + scale_linetype_discrete(name = "Performance",
breaks = c("prec",      "rec",    "auc"),
labels = c("Precision", "Recall", "AUC"))
p + theme(legend.position=c(.9, .2))
### plot sLDA ------------------------------------------------------------------
result <- read.csv("./saved/training_size_slda.csv", header = TRUE)
p <- ggplot(
data = melt(result, id.vars = "size", measure.vars = c("prec", "rec", "auc")),
mapping = aes_string(x = "size", y = "value")
)
p <- p + stat_summary(fun.data = "median_hilow", geom = "line",   aes_string(linetype = "variable"))
p <- p + stat_summary(fun.data = "median_hilow", geom = "ribbon", aes_string(fill = "variable"), alpha = 0.1, show.legend = FALSE)
p <- p + theme_bw(base_size = 11)
p <- p + scale_x_continuous(breaks = seq(from=0, to=20000, by=2000))
p <- p + labs(title = "sLDA model", x = "training dataset size", y = "model evaluation")
p <- p + scale_fill_grey(name = "Performance", start = 0.2, end = 0.2)
p <- p + scale_linetype_discrete(name = "Performance",
breaks = c("prec",      "rec",    "auc"),
labels = c("Precision", "Recall", "AUC"))
p + theme(legend.position=c(.9, .2))
### plot SVM & sLDA ------------------------------------------------------------
result1 <- read.csv("./saved/training_size_svm.csv",  header = TRUE)
result1 <- melt(result1, id.vars = "size", measure.vars = c("auc"))
result1$model <- "svm"
result2 <- read.csv("./saved/training_size_slda.csv", header = TRUE)
result2 <- melt(result2, id.vars = "size", measure.vars = c("auc"))
result2$model <- "slda"
result <- rbind(result1, result2)
p <- ggplot(
data = result,
mapping = aes_string(x = "size", y = "value", group = "model")
)
p <- p + stat_summary(fun.data = "median_hilow", geom = "line",   aes_string(linetype = "model"))
p <- p + stat_summary(fun.data = "median_hilow", geom = "ribbon", aes_string(fill = "model"), alpha = 0.1, show.legend = FALSE)
p <- p + theme_bw(base_size = 11)
p <- p + scale_x_continuous(breaks = seq(from=1000, to=20000, by=2000))
p <- p + labs(
title = "SVM vs sLDA",
x = "training dataset size",
y = "model evaluation"
)
p <- p + scale_fill_grey(name = "Performance", start = 0.2, end = 0.2)
p <- p + scale_linetype_discrete(name = "Performance",
breaks = c("svm", "slda"),
labels = c("SVM", "sLDA"))
p + theme(legend.position = c(.9, .2))
result <- read.csv("./saved/training_size_slda.csv", header = TRUE)
p <- ggplot(
data = melt(result, id.vars = "size", measure.vars = c("prec", "rec")),
mapping = aes_string(x = "size", y = "value")
)
p <- p + stat_summary(fun.data = "median_hilow", geom = "line",   aes_string(linetype = "variable"))
p <- p + stat_summary(fun.data = "median_hilow", geom = "ribbon", aes_string(fill = "variable"), alpha = 0.1, show.legend = FALSE)
p <- p + theme_bw(base_size = 11)
# p <- p + scale_x_continuous(breaks = seq(from=0, to=20000, by=2000))
p <- p + scale_x_continuous(breaks = c(1000, 5000, 10000, 15000, 20000), expand = c(0, 0))
p <- p + scale_y_continuous(limits = c(0.6, 0.9))
p <- p + labs(x = "training dataset size", y = "model evaluation")
p <- p + scale_fill_grey(start = 0.2, end = 0.2)
p <- p + scale_linetype_discrete(name = "",
breaks = c("prec",      "rec"),
labels = c("Precision", "Recall"))
p <- p + theme(legend.position=c(.8, .2))
p
pdf(
file = "./tmp/trainset-size.pdf",
width = 3.03, height = 2.5#, units = "in"
)
print(p)
dev.off()
rm(result, result1, result2, p)
### dataset --------------------------------------------------------------------
# clean and format raw data
source('./01_01_CrisisLexT26_prepare.R', echo=FALSE)
source('./01_02_CrisisLexT6_prepare.R',  echo=FALSE)
source('./01_03_AIDR2015Q2_prepare.R',   echo=FALSE)
# balance classes
source('./01_04_balance.R', echo=FALSE)
### corpora --------------------------------------------------------------------
source('./02_01_corpora.R', echo=FALSE)
source('./03_01_svm.R') # SVM
source('./03_02_lda.R') # sLDA
# the best one is SVM, train on full dataset
source('./03_03_svm_full.R')
source('./04_02_trainset_size_svm.R') # SVM
source('./04_03_trainset_size_lda.R') # sLDA
# plot
source('./04_01_trainset_size.R')
source('./04_01_trainset_size.R')
library(ggplot2)
library(reshape2)
library(Cairo)
### console --------------------------------------------------------------------
options(max.print = 100)
### plot SVM -------------------------------------------------------------------
result <- read.csv("./saved/training_size_svm.csv", header = TRUE)
p <- ggplot(
data = melt(result, id.vars = "size", measure.vars = c("prec", "rec", "auc")),
mapping = aes_string(x = "size", y = "value")
)
p <- p + stat_summary(fun.data = "median_hilow", geom = "line",   aes_string(linetype = "variable"))
p <- p + stat_summary(fun.data = "median_hilow", geom = "ribbon", aes_string(fill = "variable"), alpha = 0.1, show.legend = FALSE)
p <- p + theme_bw(base_size = 11)
p <- p + scale_x_continuous(breaks = seq(from=1000, to=20000, by=2000))
p <- p + labs(title = "SVM model", x = "training dataset size", y = "model evaluation")
p <- p + scale_fill_grey(name = "Performance", start = 0.2, end = 0.2)
p <- p + scale_linetype_discrete(name = "Performance",
breaks = c("prec",      "rec",    "auc"),
labels = c("Precision", "Recall", "AUC"))
p + theme(legend.position=c(.9, .2))
source('./04_01_trainset_size.R')
print(p)
library(ggplot2)
library(reshape2)
library(Cairo)
### console --------------------------------------------------------------------
options(max.print = 100)
### plot SVM -------------------------------------------------------------------
result <- read.csv("./saved/training_size_svm.csv", header = TRUE)
p <- ggplot(
data = melt(result, id.vars = "size", measure.vars = c("prec", "rec", "auc")),
mapping = aes_string(x = "size", y = "value")
)
p <- p + stat_summary(fun.data = "median_hilow", geom = "line",   aes_string(linetype = "variable"))
p <- p + stat_summary(fun.data = "median_hilow", geom = "ribbon", aes_string(fill = "variable"), alpha = 0.1, show.legend = FALSE)
p <- p + theme_bw(base_size = 11)
p <- p + scale_x_continuous(breaks = seq(from=1000, to=20000, by=2000))
p <- p + labs(title = "SVM model", x = "training dataset size", y = "model evaluation")
p <- p + scale_fill_grey(name = "Performance", start = 0.2, end = 0.2)
p <- p + scale_linetype_discrete(name = "Performance",
breaks = c("prec",      "rec",    "auc"),
labels = c("Precision", "Recall", "AUC"))
p + theme(legend.position=c(.9, .2))
### plot sLDA ------------------------------------------------------------------
result <- read.csv("./saved/training_size_slda.csv", header = TRUE)
p <- ggplot(
data = melt(result, id.vars = "size", measure.vars = c("prec", "rec", "auc")),
mapping = aes_string(x = "size", y = "value")
)
p <- p + stat_summary(fun.data = "median_hilow", geom = "line",   aes_string(linetype = "variable"))
p <- p + stat_summary(fun.data = "median_hilow", geom = "ribbon", aes_string(fill = "variable"), alpha = 0.1, show.legend = FALSE)
p <- p + theme_bw(base_size = 11)
p <- p + scale_x_continuous(breaks = seq(from=0, to=20000, by=2000))
p <- p + labs(title = "sLDA model", x = "training dataset size", y = "model evaluation")
p <- p + scale_fill_grey(name = "Performance", start = 0.2, end = 0.2)
p <- p + scale_linetype_discrete(name = "Performance",
breaks = c("prec",      "rec",    "auc"),
labels = c("Precision", "Recall", "AUC"))
p + theme(legend.position=c(.9, .2))
print(p)
print(p)
library(ggplot2)
library(reshape2)
library(Cairo)
### console --------------------------------------------------------------------
options(max.print = 100)
### plot SVM -------------------------------------------------------------------
result <- read.csv("./saved/training_size_svm.csv", header = TRUE)
p <- ggplot(
data = melt(result, id.vars = "size", measure.vars = c("prec", "rec", "auc")),
mapping = aes_string(x = "size", y = "value")
)
p <- p + stat_summary(fun.data = "median_hilow", geom = "line",   aes_string(linetype = "variable"))
p <- p + stat_summary(fun.data = "median_hilow", geom = "ribbon", aes_string(fill = "variable"), alpha = 0.1, show.legend = FALSE)
p <- p + theme_bw(base_size = 11)
p <- p + scale_x_continuous(breaks = seq(from=1000, to=20000, by=2000))
p <- p + labs(title = "SVM model", x = "training dataset size", y = "model evaluation")
p <- p + scale_fill_grey(name = "Performance", start = 0.2, end = 0.2)
p <- p + scale_linetype_discrete(name = "Performance",
breaks = c("prec",      "rec",    "auc"),
labels = c("Precision", "Recall", "AUC"))
p + theme(legend.position=c(.9, .2))
### plot sLDA ------------------------------------------------------------------
result <- read.csv("./saved/training_size_slda.csv", header = TRUE)
p <- ggplot(
data = melt(result, id.vars = "size", measure.vars = c("prec", "rec", "auc")),
mapping = aes_string(x = "size", y = "value")
)
p <- p + stat_summary(fun.data = "median_hilow", geom = "line",   aes_string(linetype = "variable"))
p <- p + stat_summary(fun.data = "median_hilow", geom = "ribbon", aes_string(fill = "variable"), alpha = 0.1, show.legend = FALSE)
p <- p + theme_bw(base_size = 11)
p <- p + scale_x_continuous(breaks = seq(from=0, to=20000, by=2000))
p <- p + labs(title = "sLDA model", x = "training dataset size", y = "model evaluation")
p <- p + scale_fill_grey(name = "Performance", start = 0.2, end = 0.2)
p <- p + scale_linetype_discrete(name = "Performance",
breaks = c("prec",      "rec",    "auc"),
labels = c("Precision", "Recall", "AUC"))
p <- p + theme(legend.position=c(.9, .2))
print(p)
library(ggplot2)
library(reshape2)
library(Cairo)
### console --------------------------------------------------------------------
options(max.print = 100)
### plot SVM -------------------------------------------------------------------
result <- read.csv("./saved/training_size_svm.csv", header = TRUE)
p <- ggplot(
data = melt(result, id.vars = "size", measure.vars = c("prec", "rec", "auc")),
mapping = aes_string(x = "size", y = "value")
)
p <- p + stat_summary(fun.data = "median_hilow", geom = "line",   aes_string(linetype = "variable"))
p <- p + stat_summary(fun.data = "median_hilow", geom = "ribbon", aes_string(fill = "variable"), alpha = 0.1, show.legend = FALSE)
p <- p + theme_bw(base_size = 11)
p <- p + scale_x_continuous(breaks = seq(from=1000, to=20000, by=2000))
p <- p + labs(title = "SVM model", x = "training dataset size", y = "model evaluation")
p <- p + scale_fill_grey(name = "Performance", start = 0.2, end = 0.2)
p <- p + scale_linetype_discrete(name = "Performance",
breaks = c("prec",      "rec",    "auc"),
labels = c("Precision", "Recall", "AUC"))
p <- p + theme(legend.position=c(.9, .2))
print(p)
print("SVM model / training-ste size")
print(p)
result <- read.csv("./saved/training_size_slda.csv", header = TRUE)
p <- ggplot(
data = melt(result, id.vars = "size", measure.vars = c("prec", "rec", "auc")),
mapping = aes_string(x = "size", y = "value")
)
p <- p + stat_summary(fun.data = "median_hilow", geom = "line",   aes_string(linetype = "variable"))
p <- p + stat_summary(fun.data = "median_hilow", geom = "ribbon", aes_string(fill = "variable"), alpha = 0.1, show.legend = FALSE)
p <- p + theme_bw(base_size = 11)
p <- p + scale_x_continuous(breaks = seq(from=0, to=20000, by=2000))
p <- p + labs(title = "sLDA model", x = "training dataset size", y = "model evaluation")
p <- p + scale_fill_grey(name = "Performance", start = 0.2, end = 0.2)
p <- p + scale_linetype_discrete(name = "Performance",
breaks = c("prec",      "rec",    "auc"),
labels = c("Precision", "Recall", "AUC"))
p <- p + theme(legend.position=c(.9, .2))
print("sLDA model / training-ste size")
print(p)
result <- read.csv("./saved/training_size_slda.csv", header = TRUE)
p <- ggplot(
data = melt(result, id.vars = "size", measure.vars = c("prec", "rec", "auc")),
mapping = aes_string(x = "size", y = "value")
)
p <- p + stat_summary(fun.data = "median_hilow", geom = "line",   aes_string(linetype = "variable"))
p <- p + stat_summary(fun.data = "median_hilow", geom = "ribbon", aes_string(fill = "variable"), alpha = 0.1, show.legend = FALSE)
p <- p + theme_bw(base_size = 11)
p <- p + scale_x_continuous(breaks = seq(from=0, to=20000, by=2000))
p <- p + labs(title = "sLDA model", x = "training dataset size", y = "model evaluation")
p <- p + scale_fill_grey(name = "Performance", start = 0.2, end = 0.2)
p <- p + scale_linetype_discrete(name = "Performance",
breaks = c("prec",      "rec",    "auc"),
labels = c("Precision", "Recall", "AUC"))
p <- p + theme(legend.position=c(.9, .2))
print("sLDA model / training-ste size")
print(p)
result1 <- read.csv("./saved/training_size_svm.csv",  header = TRUE)
result1 <- melt(result1, id.vars = "size", measure.vars = c("auc"))
result1$model <- "svm"
result2 <- read.csv("./saved/training_size_slda.csv", header = TRUE)
result2 <- melt(result2, id.vars = "size", measure.vars = c("auc"))
result2$model <- "slda"
result <- rbind(result1, result2)
p <- ggplot(
data = result,
mapping = aes_string(x = "size", y = "value", group = "model")
)
p <- p + stat_summary(fun.data = "median_hilow", geom = "line",   aes_string(linetype = "model"))
p <- p + stat_summary(fun.data = "median_hilow", geom = "ribbon", aes_string(fill = "model"), alpha = 0.1, show.legend = FALSE)
p <- p + theme_bw(base_size = 11)
p <- p + scale_x_continuous(breaks = seq(from=1000, to=20000, by=2000))
p <- p + labs(
title = "SVM vs sLDA",
x = "training dataset size",
y = "model evaluation"
)
p <- p + scale_fill_grey(name = "Performance", start = 0.2, end = 0.2)
p <- p + scale_linetype_discrete(name = "Performance",
breaks = c("svm", "slda"),
labels = c("SVM", "sLDA"))
p + theme(legend.position = c(.9, .2))
print("AUC of SVM & sLDA")
print(p)
echo("AUC of SVM & sLDA")
print("AUC of SVM & sLDA")
print(p)
rm(result, result1, result2, p)
source('./04_01_trainset_size.R')
cat("PLOT: SVM model / training-ste size")
cat(p)
print(p)
source('./04_01_trainset_size.R')
cat("PLOT: AUC of SVM & sLDA\n")
source('./04_01_trainset_size.R')
rm(
container, cutoff.index, cutoff.value, data, data.corpus, data.dtm,
label_on, model, predict, rocr.perf, rocr.pred, test, testSet, testSet.files,
testSet.quality, trainSet
)
library(tm)
library(RTextTools)
library(ROCR)
library(ggplot2)
library(reshape2)
library(parallel)
### console --------------------------------------------------------------------
options(max.print = 100)
### load data ------------------------------------------------------------------
data.corpus <- readRDS("./saved/corpus.rds")
data <- data.frame(
meta(data.corpus, "label", type="indexed"),
meta(data.corpus, "file",  type="indexed")
)
data.dtm <- readRDS("./saved/dtmatrix.rds")
### training & testing datasets ------------------------------------------------
trainSet.files <- c(
"T6_2013_Queensland_Floods"   # 8732
, "T6_2013_Alberta_Floods"      # 9069
, "T26_2013_Manila_floods"      # 1167
, "T26_2013_Sardinia_floods"    #  180
, "T26_2012_Philipinnes_floods" # 1318
)
testSet.files <- c(
"T26_2013_Colorado_floods"    # 1664
)
trainSet <- which(data$file %in% trainSet.files)
testSet  <- which(data$file %in% testSet.files)
cat(sprintf("train/test: %d / %d (%.1f%% / %.1f%%)",
length(trainSet), length(testSet),
length(trainSet)/(length(trainSet)+length(testSet))*100,
length(testSet) /(length(trainSet)+length(testSet))*100
))
### SVM ------------------------------------------------------------------------
model.size <- seq(1000, 20000, 1000)
model.size <- rep(model.size, 3)
model.quality <- mclapply(model.size, mc.cores = 3, FUN = function(iter.size) {
# iter.size <- 1000
### train and test model -----------------------------------------------------
container <- create_container(
matrix = data.dtm,
labels = factor(data$label, levels = c(0,1), labels = c("off","on")),
trainSize = sort( sample(trainSet, size = iter.size, replace = FALSE) ),
testSize = testSet,
virgin = FALSE
)
model <- train_model(container, "SVM")
predict <- predict(model, container@classification_matrix, probability=TRUE)
label_on <- which(dimnames(attr(predict, "probabilities"))[[2]] == "on")
### accuracy metrics ---------------------------------------------------------
rocr.pred <- prediction(
predictions = as.vector( attr(predict, "probabilities")[, label_on] ),
labels      = data$label[testSet]
)
rocr.perf    <- performance(rocr.pred, measure = "acc", x.measure = "cutoff")
cutoff.index <- which.max(rocr.perf@y.values[[1]])
return(c(
"prec" = performance(rocr.pred, measure = "prec")@y.values[[1]][cutoff.index],
"rec"  = performance(rocr.pred, measure = "rec" )@y.values[[1]][cutoff.index],
"auc"  = performance(rocr.pred, measure = "auc" )@y.values[[1]]
))
})
model.quality <- do.call(rbind, model.quality)
model.quality <- as.data.frame.matrix(model.quality)
model.quality$size <- model.size
if (file.exists("./saved/training_size_svm.csv") == TRUE) {
result <- read.csv("./saved/training_size_svm.csv", header = TRUE)
result <- rbind(result, model.quality)
} else {
result <- model.quality
}
write.csv(result, file = "./saved/training_size_svm.csv", row.names = FALSE)
rm(
data, model.quality, result, data.corpus, data.dtm, model.size, testSet,
testSet.files, trainSet, trainSet.files
)
source('./04_01_trainset_size.R')
