Cours DFGSM2 by VLooten

init

looten — 2019-01-09 13:33:16 UTC

install.packages("caret")
install.packages("e1071")
install.packages("randomForest")

library(caret)
library(e1071)
library(randomForest)
library(dplyr)
library(tidyr)

data(iris)
dtt <- iris

Tirage au sort

looten — 2019-01-09 13:40:08 UTC

# Echantillon train
set.seed(1987)
train_idx <- sample(rownames(dtt),100,F)
dtt_train <- dtt[which(rownames(dtt) %in% train_idx),]
# Echantillon test
dtt_test <- dtt[which(!(rownames(dtt) %in% train_idx)),]

GLM

looten — 2019-01-09 13:51:14 UTC

dtt_train$Y <- as.numeric(dtt_train$Species=="setosa")
colnames(dtt_train)
model_glm <- glm(Y~Sepal.Length + Sepal.Width + Petal.Length+ Petal.Width
, data=dtt_train, family="binomial")
preds_glm <- predict(model_glm, newdata = dtt_test, type = "response")

Trouvez les définitions de :

looten — 2019-01-09 13:56:51 UTC

Recall (rappel)
Precision
F-measure / F1-score

fonction

looten — 2019-01-09 14:10:29 UTC

evaluate <- function(TP,FP,TN,FN){
precision <- TP/(TP+FP)
recall <- TP/(TP+FN)
fscore <- 2*(precision*recall)/(precision+recall)
return(list(precision=precision,
recall=recall,
fscore=fscore))
}

evaluate(TP=19,FP=0,TN=31,FN=0)

looten — 2019-01-09 14:11:16 UTC

preds_setosa <- preds_glm>0.5
table(preds_setosa, dtt_test$Species)

looten — 2019-01-09 14:30:56 UTC

library(e1071)
dtt_train$Y <- as.factor(dtt_train$Y)

model_svm <- svm(Y~Sepal.Length + Sepal.Width + Petal.Length+ Petal.Width,
data=dtt_train,probability=T)
preds_svm <- predict(model_svm,newdata = dtt_test,
probability=T)

# attr(preds_svm,"probabilities")[,2]
preds_setosa2 <- preds_svm
table(preds_setosa2, dtt_test$Species)

Explicabilité

looten — 2019-01-09 14:43:04 UTC

https://www.ncbi.nlm.nih.gov/pubmed/27723811

looten — 2019-01-09 15:09:24 UTC

install.packages("rpart")
install.packages("partykit")
library(rpart)
library("partykit")
model_arbre <- rpart(Species~Sepal.Length + Sepal.Width + Petal.Length+ Petal.Width,
data=dtt_train)
plot(as.party(model_arbre))

looten — 2019-01-09 15:19:48 UTC

model_rf <- randomForest(Y~Sepal.Length + Sepal.Width + Petal.Length+ Petal.Width,
data=dtt_train)
preds_rf <- predict(model_rf,newdata = dtt_test,
probability=T)
preds_setosa3 <- preds_rf
table(preds_setosa3, dtt_test$Species)

looten — 2019-01-09 15:24:53 UTC

varImpPlot(model_rf)