library(readr) library(cluster) library(fpc) library(mclust) library(neuralnet) train10_1 = read_csv('../input/train/subj10_series1_data.csv') events10_1 = read_csv("../input/train/subj10_series1_events.csv") # collapsing event columns into one #events10_1$Results[events10_1$HandStart == 1] <- 1 #events10_1$Results[events10_1$FirstDigitTouch == 1] <- 2 #events10_1$Results[events10_1$BothStartLoadPhase == 1] <- 3 #events10_1$Results[events10_1$LiftOff == 1] <- 4 #events10_1$Results[events10_1$Replace == 1] <- 5 #events10_1$Results[events10_1$BothReleased == 1] <- 6 #HandStart, FirstDigitTouch, BothStartLoadPhase, LiftOff, Replace, BothReleased #table(is.na(events10_1$Results)) #temp$Results <- as.numeric(temp$Results) #events10_1$Results <- as.numeric(events10_1$Results) #summary(temp) #events10_1$Results <- temp$Results #hist(events10_1$Results) #summary(train10_1) # transforming id train10_1$id <- as.character(train10_1$id) train10_1$id <- sapply(train10_1$id, FUN=function(x) {strsplit(x, split='[_]')[[1]][3]}) train10_1$id <- as.factor(train10_1$id) events10_1$id <- as.character(events10_1$id) events10_1$id <- sapply(events10_1$id, FUN=function(x) {strsplit(x, split='[_]')[[1]][3]}) events10_1$id <- as.factor(events10_1$id) # checking n/a and null values #table(is.na(train10_1)) #summary(train10_1) #summary(events10_1) # 1/ #fit1 <- kmeans(train10_1, 6, iter.max = 10, nstart = 1, # algorithm = c("Hartigan-Wong"), trace=FALSE) #hist(fit1$cluster) #hist(events10_1$Results) #clusplot(train10_1, fit1$cluster, color=TRUE, shade=TRUE, # labels=2, lines=0) #plotcluster(train10_1, fit$cluster) # 2/ #fit2 <- kmeans(train10_1, 6) #clusplot(train10_1, fit2$cluster, color=TRUE, shade=TRUE, # labels=2, lines=0) # 3/ #fit3 <- Mclust(train10_1) #plot(fit3) # plot results #summary(fit3) # 4/ #fit4 <- kmeans(train10_1, 6, iter.max = 10, nstart = 1, # algorithm = c("Forgy"), trace=FALSE) #hist(fit4$cluster) #clusplot(train10_1, fit4$cluster, color=TRUE, shade=TRUE, # labels=2, lines=0) # 5/ #fit5 <- kmeans(train10_1, 6, iter.max = 10, nstart = 1, # algorithm = c("MacQueen"), trace=FALSE) #hist(fit5$cluster) #clusplot(train10_1, fit5$cluster, color=TRUE, shade=TRUE, # labels=2, lines=0) ############# #plot(train10_1, col = fit1$cluster) # ============================================================== #fit <- kmeans(train10_1, 6, iter.max = 10, nstart = 25, # algorithm = c("Hartigan-Wong"), trace=FALSE) #hist(fit$cluster) #clusplot(train10_1, fit$cluster, color=TRUE, shade=TRUE, # labels=2, lines=0) #summary(fit) #train10_1$Cluster <- fit$cluster train10_1$Results <- events10_1$Results #summary(train10_1) #table(train10_1$Results) #table(train10_1$Cluster) #plot(train10_1$Results) #plotcluster(train10_1, fit$cluster) #summary(train10_1) #summary(events10_1) #events10_1$HandStart[events10_1$HandStart == 0] <- NA #events10_1$FirstDigitTouch[events10_1$FirstDigitTouch == 0] <- NA #events10_1$BothStartLoadPhase[events10_1$BothStartLoadPhase == 0] <- NA #events10_1$LiftOff[events10_1$LiftOff == 0] <- NA #events10_1$Replace[events10_1$Replace == 0] <- NA #events10_1$BothReleased[events10_1$BothReleased == 0] <- NA #HandStart, FirstDigitTouch, BothStartLoadPhase, LiftOff, Replace, BothReleased # designing and executing # For Right Handed people # Fp1 designs the movement? # FC5, FC1, and C3 execute movement #plot(train10_1$Fp1, type="l", col="blue", # axes=F, ann=T, # cex.lab=0.8, lwd=2) #lines(train10_1$FC5, type="o", pch=22, col="red") #lines(train10_1$FC1, type="o", pch=23, col="magenta") #lines(train10_1$C3, type="o", pch=24, col="purple") #points(events10_1$HandStart, type="o", col="blue") #points(events10_1$FirstDigitTouch, type="o", col="red") #points(events10_1$BothStartLoadPhase, type="o", col="magenta") #points(events10_1$LiftOff, type="o", col="yellow") #points(events10_1$Replace, type="o", col="green") #points(events10_1$BothReleased, type="o", col="black") # OR F7, and F3 design the movement? #plot(train10_1$F7, type="l", col="blue", # axes=F, ann=T, # cex.lab=0.8, lwd=2) #lines(train10_1$F3, type="o", pch=21, col="yellow") #lines(train10_1$FC5, type="o", pch=22, col="red") #lines(train10_1$FC1, type="o", pch=23, col="magenta") #lines(train10_1$C3, type="o", pch=24, col="purple") #points(events10_1$HandStart, type="o", col="blue") #points(events10_1$FirstDigitTouch, type="o", col="red") #points(events10_1$BothStartLoadPhase, type="o", col="magenta") #points(events10_1$LiftOff, type="o", col="yellow") #points(events10_1$Replace, type="o", col="green") #points(events10_1$BothReleased, type="o", col="black") # hmm #plot(train10_1[2500:7717,]$F7, type="l", col="blue", # axes=F, ann=T, # cex.lab=0.8, lwd=2) #lines(train10_1[2500:7717,]$F3, type="o", pch=21, col="yellow") #lines(train10_1[2500:7717,]$FC5, type="o", pch=22, col="red") #lines(train10_1[2500:7717,]$FC1, type="o", pch=23, col="magenta") #lines(train10_1[2500:7717,]$C3, type="o", pch=24, col="purple") #points(events10_1[2500:7717,]$HandStart, type="o", col="blue") #points(events10_1[2500:7717,]$FirstDigitTouch, type="o", col="red") #points(events10_1[2500:7717,]$BothStartLoadPhase, type="o", col="magenta") #points(events10_1[2500:7717,]$LiftOff, type="o", col="yellow") #points(events10_1[2500:7717,]$Replace, type="o", col="green") #points(events10_1[2500:7717,]$BothReleased, type="o", col="black") #plot(events10_1$HandStart, type="o", col="blue", # axes=F, ann=T, # cex.lab=0.8, lwd=2) #lines(events10_1$FirstDigitTouch, type="o", pch=21, col="red") #lines(events10_1$BothStartLoadPhase, type="o", pch=22, col="magenta") #lines(events10_1$LiftOff, type="o", pch=23, col="yellow") #lines(events10_1$Replace, type="o", pch=24, col="green") #lines(events10_1$BothReleased, type="o", pch=25, col="black") #events10_1$HandStart[events10_1$HandStart == 0] <- NA #events10_1$FirstDigitTouch[events10_1$FirstDigitTouch == 0] <- NA #events10_1$BothStartLoadPhase[events10_1$BothStartLoadPhase == 0] <- NA #events10_1$LiftOff[events10_1$LiftOff == 0] <- NA #events10_1$Replace[events10_1$Replace == 0] <- NA #events10_1$BothReleased[events10_1$BothReleased == 0] <- NA train10_1$HandStart <- events10_1$HandStart train10_1$FirstDigitTouch <- events10_1$FirstDigitTouch train10_1$BothStartLoadPhase <- events10_1$BothStartLoadPhase train10_1$LiftOff <- events10_1$LiftOff train10_1$Replace <- events10_1$Replace train10_1$BothReleased <- events10_1$BothReleased # Neural network #fit <- neuralnet(HandStart + FirstDigitTouch + BothStartLoadPhase + LiftOff + Replace + BothReleased ~ Fp1 + Fp2 + # F7 + F3 + F8 + F4 + FC1 + FC2 + FC5 + FC6 + Fz + T7 + T8 + C3 + Cz + C4 + TP9 + TP10 + CP1 + CP2 + CP5 + CP6 + # P7 + P8 + P3 + P4 + Pz + PO9 + PO10 + O1 + O2 + Oz, train10_1, hidden = c(4, 8), lifesign = "minimal", # linear.output = FALSE, threshold = 0.1, rep = 1) #fit <- neuralnet(HandStart + FirstDigitTouch + BothStartLoadPhase + LiftOff + Replace + BothReleased ~ Fp1 + # F7 + F3 + FC1 + FC5 + T7 + C3 + TP9 + CP1 + CP5 + # P7 + P3 + PO9 + O1, train10_1, hidden = 4, lifesign = "minimal", # linear.output = FALSE, threshold = 0.1, rep = 1, err.fct = 'ce') train <- train10_1[0:246852,] temp <- train10_1[246853:262387,] test <- c(temp$Fp1, temp$F7, temp$F3, temp$FC1, temp$FC5, temp$T7, temp$C3, temp$TP9, temp$CP1, temp$CP5, temp$P7, temp$P3, temp$PO9, temp$O1) fit <- neuralnet(HandStart + FirstDigitTouch ~ Fp1 + F7 + F3 + FC1 + FC5 + T7 + C3 + TP9 + CP1 + CP5 + P7 + P3 + PO9 + O1, data=train, algorithm="backprop", hidden=c(14, 8, 4), learningrate=0.1, linear.output=FALSE, lifesign = "maximum") summary(fit) coefficients(fit) table(fit$response) table(fit$net) plot(fit, rep = "best") fit$results <- compute(fit, test)#$as.numeric(fit.result) table(pred$fit.result) summary(pred) pred$fit.result #test$HandStartPred <- prediction(fit, train10_1[246853:262387,]) #predict(fit, train10_1[246853:262387,]) #test[246853:262387,]$Test[test$HandStartPred == test$HandStart] <- 1 #sum(test)