testCSVfiles.Rmd

---
title: "Test clustering on Csv files"
author: "Kevin"
date: "12 août 2019"
output:
  pdf_document: default
  html_document: default
  always_allow_html: yes
---

```{r setup, include=FALSE}
rm(list=ls())
knitr::opts_chunk$set(echo = TRUE)
library(trackeR)

source("Functions/k-means.R")
source("Functions/troncons.R")
source("RL/MDP.R")
```

12/08/19
# Une session
## Import données
```{r, include=TRUE}
run1 <- read.csv('../Data/DataCSV/25060097/2530279693.csv')
# run1 <- data.frame(run1)

p <- dim(run1)[2]

names(run1)
```

## Decoupage en troncons
```{r, include=TRUE}
# i <- 7
# run <- df.runs[df.runs$session == i, ]
runCleaned <- preProcessDataForSectionCSV(run1)
troncons <- sliceSession(runCleaned)


print(troncons)
n <- dim(troncons)[1]
p <- dim(troncons)[2]

indexTroncons <- 1:n
plot(indexTroncons, troncons$speed, main = paste("speed each section for session"), type = "p", pch = 20)
```

## k-means sur une course
```{r, include=TRUE}
set.seed(123)
# sans distance
indexTroncons <- 1:n
tronconsScaled <- scale(troncons[,-1]) 
kBest <- bestK(tronconsScaled, n-1)
km1 <- kmeans(tronconsScaled, 3)
plot(indexTroncons, troncons$speed, main = paste("speed each section - KMEANS"), type = "p", pch = 20, col = km1$cluster)
```



# Plus de courses
```{r, include=TRUE}
fp <- list.files("../Data/DataCSV/25060097", full.names = TRUE)

sessions <- list()
for(k in 1:length(fp)){
  if(substr(fp[k],nchar(fp[k])-2, nchar(fp[k])) == "csv"){
    dftemp <- read.csv(fp[k])
    sessions[[k]] <- dftemp
  }
}

print(sessions[[2]])

nbSessions <- length(sessions)
print(paste("nb of running sessions =", nbSessions))

df <- sessions[[74]]


runCleaned <- preProcessDataForSectionCSV(df)
troncons <- sliceSession(runCleaned)

```

## kmeans sans distance
```{r, include=TRUE}
k.avg <- 0
nbSessionsClean <- nbSessions

set.seed(1)

for (i in 1:nbSessions) {
  # print(i)
  run <- sessions[[i]]
  runCleaned <- preProcessDataForSectionCSV(run)
  troncons <- sliceSession(runCleaned)
  
  # Remove "distance" variable
  troncons <- troncons[, -1]
  
  n <- dim(troncons)[1]
  p <- dim(troncons)[2]
  
  troncons.norm <- sapply(troncons, scale)
  row.names(troncons.norm) <- row.names(troncons) 
  
  # A changer
  if (n == 2) {
    nbSessionsClean <- nbSessionsClean - 1
    next
  }

  kBest <- bestK(troncons.norm, n-1)
  if (kBest == 0) { 
    nbSessionsClean <- nbSessionsClean - 1
    kBest <- 1
  }
  k.avg <- k.avg + kBest 
  km <- kmeans(troncons.norm, kBest)
  
  indexTroncons <- 1:n
  # plot(indexTroncons, data.frame(troncons.norm)$speed, main = paste("speed each section for session", i), type = "p", pch = 20, col = km$cluster)
}

print(paste("nb of running sessions =", nbSessionsClean))
k.avg <- k.avg/nbSessionsClean
print(paste("k moyen =", round(k.avg, 2)))
```

21/08/19
## kmeans avec distance
```{r, include=TRUE}
k.avg <- 0
nbSessionsClean <- nbSessions
set.seed(1)

for (i in 1:nbSessions) {
  # print(i)
  run <- sessions[[i]]
  runCleaned <- preProcessDataForSectionCSV(run)
  troncons <- sliceSession(runCleaned)
  
  # Remove "distance" variable
  # troncons <- troncons[, -1]
  
  n <- dim(troncons)[1]
  p <- dim(troncons)[2]
  
  troncons.norm <- sapply(troncons, scale)
  row.names(troncons.norm) <- row.names(troncons) 
  
  # A changer
  if (n == 2) {
    nbSessionsClean <- nbSessionsClean - 1
    next
  }

  kBest <- bestK(troncons.norm, n-1)
  if (kBest == 0) { 
    nbSessionsClean <- nbSessionsClean - 1
    kBest <- 1
  }
  k.avg <- k.avg + kBest 
  km <- kmeans(troncons.norm, kBest)
  
  indexTroncons <- 1:n
  # plot(indexTroncons, data.frame(troncons.norm)$speed, main = paste("speed each section for session", i), type = "p", pch = 20, col = km$cluster)
}

print(paste("nb of running sessions =", nbSessionsClean))
k.avg <- k.avg/nbSessionsClean
print(paste("k moyen =", round(k.avg, 2)))
```



## kmeans with k = 3 par session
```{r, include=TRUE}
k <- 4
nbSessionsClean <- nbSessions

set.seed(1)

cluster1 <- list()

for (i in 1:nbSessions) {
  print(i)
  run <- sessions[[i]]
  runCleaned <- preProcessDataForSectionCSV(run)
  troncons <- sliceSession(runCleaned)
  
  # Remove "distance" variable
  troncons <- troncons[, -1]
  
  n <- dim(troncons)[1]
  p <- dim(troncons)[2]
  
  troncons.norm <- sapply(troncons, scale)
  row.names(troncons.norm) <- row.names(troncons) 
  
  # A changer
  if (n <= k) {
    nbSessionsClean <- nbSessionsClean - 1
    next
  }

  km <- kmeans(troncons.norm, k)
  
  indexTroncons <- 1:n
  # plot(indexTroncons, troncons$speed, main = paste("speed each section for session", i), type = "p", pch = 20, col = km$cluster)
  
  cluster1[[i]] <- km$cluster
}

print(paste("nb of running sessions =", nbSessionsClean))
```

20/08/19
## kmeans sur toutes les sessions combinées
```{r, include=TRUE}
k <- 4
nbSessionsClean <- nbSessions

set.seed(1)

allSessions <- data.frame()
for (i in 1:nbSessions) {
  run <- sessions[[i]]
  runCleaned <- preProcessDataForSectionCSV(run)
  troncons <- sliceSession(runCleaned)
  
  troncons <- troncons[, -1]
  
  troncons.norm <- sapply(troncons, scale)
  row.names(troncons.norm) <- row.names(troncons) 
  
  allSessions <- rbind(allSessions, cbind(session = rep(i, dim(troncons)[1]), troncons.norm))
}

print(allSessions)
  
km <- kmeans(allSessions[,-1], k)
print(km$centers)

for (i in 1:nbSessions) {
  # plot(allSessions[allSessions$session == i, ]$speed, main = paste("speed each section for session", i), type = "p", pch = 20, col = km$cluster[allSessions$session == i], xlab = "troncon", ylab = "speed")
}

cluster2 <- list()
for (i in 1:nbSessions) {
  cluster2[[i]] <- km$cluster[allSessions$session == i]
}
```

## Comparaison : kmeans sur tout VS kmeans par session
```{r, include=TRUE}
for (i in 1:nbSessions) {
  par(mfrow = c(2, 1))
  plot(allSessions[allSessions$session == i, ]$speed, main = paste("all session", i), type = "p", pch = 20, col = cluster1[[i]], xlab = "troncon", ylab = "speed")
  plot(allSessions[allSessions$session == i, ]$speed, main = paste("kmeans par session", i), type = "p", pch = 20, col = cluster2[[i]], xlab = "troncon", ylab = "speed")
  
}
```



21/08/19
# k-means sur une course avec/sans distance normalisé ou non
```{r, include=TRUE}
set.seed(123)
# sans distance
indexTroncons <- 1:n
tronconsScaled <- scale(troncons[,-1]) 
kBest <- bestK(tronconsScaled, n-1)
km1 <- kmeans(tronconsScaled, 3)
plot(indexTroncons, troncons$speed, main = paste("speed each section - KMEANS"), type = "p", pch = 20, col = km1$cluster)

# avec distance normalisée
indexTroncons <- 1:n
tronconsScaled <- scale(troncons) 
kBest <- bestK(tronconsScaled, n-1)
km2 <- kmeans(tronconsScaled, 3)
plot(indexTroncons, troncons$speed, main = paste("speed each section - KMEANS"), type = "p", pch = 20, col = km2$cluster)

# avec distance non normalisée
indexTroncons <- 1:n
tronconsScaled <- cbind(troncons[, 1], scale(troncons[,-1]))
kBest <- bestK(tronconsScaled, n-1)
km3 <- kmeans(tronconsScaled, 3)
plot(indexTroncons, troncons$speed, main = paste("speed each section - KMEANS"), type = "p", pch = 20, col = km3$cluster)

print(km1$centers)
print(km2$centers)
print(km3$centers)

```



# Kmeans avec k = 3 par session
```{r, include=TRUE}
k <- 3
nbSessionsClean <- nbSessions

set.seed(1)

cluster1WithDistance <- list()

for (i in 1:nbSessions) {
  # print(i)
  run <- sessions[[i]]
  runCleaned <- preProcessDataForSectionCSV(run)
  troncons <- sliceSession(runCleaned)
  
  # Remove "distance" variable
  # troncons <- troncons[, -1]
  
  n <- dim(troncons)[1]
  p <- dim(troncons)[2]
  
  troncons.norm <- sapply(troncons, scale)
  row.names(troncons.norm) <- row.names(troncons) 
  
  # A changer
  if (n == 2) {
    nbSessionsClean <- nbSessionsClean - 1
    next
  }

  km <- kmeans(troncons.norm, k)
  
  indexTroncons <- 1:n
  plot(indexTroncons, troncons$speed, main = paste("speed each section for session", i), type = "p", pch = 20, col = km$cluster)
  
  cluster1WithDistance[[i]] <- km$cluster
}

print(paste("nb of running sessions =", nbSessionsClean))
```

20/08/19
## kmeans sur toutes les sessions combinées
```{r, include=TRUE}
k <- 3
nbSessionsClean <- nbSessions

set.seed(1)

allSessions <- data.frame()
for (i in 1:nbSessions) {
  run <- sessions[[i]]
  runCleaned <- preProcessDataForSectionCSV(run)
  troncons <- sliceSession(runCleaned)
  
  # troncons <- troncons[, -1]
  
  troncons.norm <- sapply(troncons, scale)
  row.names(troncons.norm) <- row.names(troncons) 
  
  allSessions <- rbind(allSessions, cbind(session = rep(i, dim(troncons)[1]), troncons.norm))
}

print(allSessions)
  
km <- kmeans(allSessions[,-1], k)
print(km$centers)

for (i in 1:nbSessions) {
  plot(allSessions[allSessions$session == i, ]$speed, main = paste("speed each section for session", i), type = "p", pch = 20, col = km$cluster[allSessions$session == i], xlab = "troncon", ylab = "speed")
}

cluster2WithDistance <- list()
for (i in 1:nbSessions) {
  cluster2WithDistance[[i]] <- km$cluster[allSessions$session == i]
}
```

## Comparaison : kmeans sur tout VS kmeans par session
```{r, include=TRUE}
for (i in 1:nbSessions) {
  par(mfrow = c(2, 1))
  plot(allSessions[allSessions$session == i, ]$speed, main = paste("all session", i), type = "p", pch = 20, col = cluster1WithDistance[[i]], xlab = "troncon", ylab = "speed")
  plot(allSessions[allSessions$session == i, ]$speed, main = paste("kmeans par session", i), type = "p", pch = 20, col = cluster2WithDistance[[i]], xlab = "troncon", ylab = "speed")
}
```


# Comparaison : kmeans avec distance et sans + kmeans sur tout VS chaque session
```{r, include=TRUE}
for (i in 1:nbSessions) {
  par(mfrow = c(2, 1))
  plot(allSessions[allSessions$session == i, ]$distance, main = paste("kmeans par session avec distance", i), type = "p", pch = 20, col = cluster1WithDistance[[i]], xlab = "troncon", ylab = "distance")
  plot(allSessions[allSessions$session == i, ]$distance, main = paste("kmeans par session sans distance", i), type = "p", pch = 20, col = cluster1[[i]], xlab = "troncon", ylab = "distance")
}
```



22/08/19
# Regroupement des courses de mêmes distances

## Histogramme des distances des running sessions
```{r, include=TRUE}
sizesSessions <- vector(mode = "numeric", length = nbSessions)
for (i in 1:nbSessions) {
  sizesSessions[i] <- nrow(allSessions[allSessions$session == i, ])
}

hist(sizesSessions, breaks = 20, col = 6)
print(sizesSessions)
```

## Les courses de 10km - kmeans par session
```{r, include=TRUE}
indices10km <- which(sizesSessions == 10)

k <- 3
nbSessions10km <- length(indices10km)
set.seed(1)

clusters10kmSessions_perSession <- list()

for (i in 1:nbSessions10km) {
  run <- sessions[[indices10km[i]]]
  runCleaned <- preProcessDataForSectionCSV(run)
  troncons <- sliceSession(runCleaned)
  
  n <- dim(troncons)[1]
  p <- dim(troncons)[2]
  
  troncons.norm <- sapply(troncons, scale)
  row.names(troncons.norm) <- row.names(troncons) 

  km <- kmeans(troncons.norm, k)

  indexTroncons <- 1:n
  plot(indexTroncons, troncons$speed, main = paste("speed each section for session", indices10km[i]), type = "p", pch = 20, col = km$cluster)
  
  clusters10kmSessions_perSession[[indices10km[i]]] <- km$cluster
}
```


## Les courses de 10km - kmeans avec les sessions combinées
```{r, include=TRUE}
k <- 3

set.seed(1)

all10kmSessions <- data.frame()
for (i in 1:nbSessions10km) {
  run <- sessions[[indices10km[i]]]
  runCleaned <- preProcessDataForSectionCSV(run)
  troncons <- sliceSession(runCleaned)
  
  troncons.norm <- sapply(troncons, scale)
  row.names(troncons.norm) <- row.names(troncons) 
  
  all10kmSessions <- rbind(all10kmSessions, cbind(session = rep(indices10km[i], dim(troncons)[1]), troncons.norm))
}

print(all10kmSessions)
  
km <- kmeans(all10kmSessions[,-1], k)
print(km$centers)

for (i in 1:nbSessions10km) {
  plot(all10kmSessions[all10kmSessions$session == indices10km[i], ]$speed, main = paste("speed each section for session", indices10km[i]), type = "p", pch = 20, col = km$cluster[all10kmSessions$session == indices10km[i]], xlab = "troncon", ylab = "speed")
}

clusters10kmSessions_allSession <- list()
for (i in 1:nbSessions10km) {
  clusters10kmSessions_allSession[[indices10km[i]]] <- km$cluster[all10kmSessions$session == indices10km[i]]
}
```


## Comparaison pour sessions de 10km : kmeans sur tout VS kmeans par session
```{r, include=TRUE}
for (i in 1:nbSessions10km) {
  par(mfrow = c(2, 1))
  plot(all10kmSessions[all10kmSessions$session == indices10km[i], ]$speed, main = paste("kmeans sur toutes sessions", indices10km[i]), type = "p", pch = 20, col = clusters10kmSessions_allSession[[indices10km[i]]], xlab = "troncon", ylab = "speed")
  
  plot(all10kmSessions[all10kmSessions$session == indices10km[i], ]$speed, main = paste("kmeans par session", indices10km[i]), type = "p", pch = 20, col = clusters10kmSessions_perSession[[indices10km[i]]], xlab = "troncon", ylab = "speed")
}
```


## Comparaison pour sessions de 10km : kmeans avec toutes les sessions confondues VS groupe de 10km
```{r, include=TRUE}
for (i in 1:nbSessions10km) {
  par(mfrow = c(2, 1))
  plot(all10kmSessions[all10kmSessions$session == indices10km[i], ]$speed, main = paste("seance 10km - kmeans groupe 10km", indices10km[i]), type = "p", pch = 20, col = clusters10kmSessions_allSession[[indices10km[i]]], xlab = "troncon", ylab = "speed")
  
  plot(allSessions[allSessions$session == indices10km[i], ]$speed, main = paste("seance 10km - kmeans avec toutes les courses", indices10km[i]), type = "p", pch = 20, col = cluster2WithDistance[[indices10km[i]]], xlab = "troncon", ylab = "speed")
}
```




27/08/19
## Histogramme des distances des running sessions
```{r, include=TRUE}
sizesSessions <- vector(mode = "numeric", length = nbSessions)
for (i in 1:nbSessions) {
  sizesSessions[i] <- nrow(allSessions[allSessions$session == i, ])
}

hist(sizesSessions, breaks = 20, col = 6)
print(sizesSessions)

for (i in 1:25) {
  print(paste("size =", i))
  print(sum(sizesSessions == i))
}
```

Les sessions de 5 km, 10 km, 6, 11 .

## Les courses de 5km - kmeans par session
```{r, include=TRUE}
indices5km <- which(sizesSessions == 5)

k <- 3
nbSessions5km <- length(indices5km)
set.seed(1)

clusters5kmSessions_perSession <- list()

for (i in 1:nbSessions5km) {
  run <- sessions[[indices5km[i]]]
  runCleaned <- preProcessDataForSectionCSV(run)
  troncons <- sliceSession(runCleaned)
  
  n <- dim(troncons)[1]
  p <- dim(troncons)[2]
  
  troncons.norm <- sapply(troncons, scale)
  row.names(troncons.norm) <- row.names(troncons) 

  km <- kmeans(troncons.norm, k)

  indexTroncons <- 1:n
  plot(indexTroncons, troncons$speed, main = paste("speed each section for session", indices5km[i]), type = "p", pch = 20, col = km$cluster)
  
  clusters5kmSessions_perSession[[indices5km[i]]] <- km$cluster
}
```


## Les courses de 5km - kmeans avec les sessions combinées
```{r, include=TRUE}
k <- 3

set.seed(1)

all5kmSessions <- data.frame()
for (i in 1:nbSessions5km) {
  run <- sessions[[indices5km[i]]]
  runCleaned <- preProcessDataForSectionCSV(run)
  troncons <- sliceSession(runCleaned)
  
  troncons.norm <- sapply(troncons, scale)
  row.names(troncons.norm) <- row.names(troncons) 
  
  all5kmSessions <- rbind(all5kmSessions, cbind(session = rep(indices5km[i], dim(troncons)[1]), troncons.norm))
}

print(all5kmSessions)
  
km <- kmeans(all5kmSessions[,-1], k)
print(km$centers)

for (i in 1:nbSessions5km) {
  plot(all5kmSessions[all5kmSessions$session == indices5km[i], ]$speed, main = paste("speed each section for session", indices5km[i]), type = "p", pch = 20, col = km$cluster[all5kmSessions$session == indices5km[i]], xlab = "troncon", ylab = "speed")
}

clusters5kmSessions_allSession <- list()
for (i in 1:nbSessions5km) {
  clusters5kmSessions_allSession[[indices5km[i]]] <- km$cluster[all5kmSessions$session == indices5km[i]]
}
```


## Comparaison pour sessions de 5km : kmeans sur tout VS kmeans par session
```{r, include=TRUE}
for (i in 1:nbSessions5km) {
  par(mfrow = c(2, 1))
  plot(all5kmSessions[all5kmSessions$session == indices5km[i], ]$speed, main = paste("kmeans sur toutes sessions", indices5km[i]), type = "p", pch = 20, col = clusters5kmSessions_allSession[[indices5km[i]]], xlab = "troncon", ylab = "speed")
  
  plot(all5kmSessions[all5kmSessions$session == indices5km[i], ]$speed, main = paste("kmeans par session", indices5km[i]), type = "p", pch = 20, col = clusters5kmSessions_perSession[[indices5km[i]]], xlab = "troncon", ylab = "speed")
}
```


## Comparaison pour sessions de 5km : kmeans avec toutes les sessions confondues VS groupe de 5km
```{r, include=TRUE}
for (i in 1:nbSessions5km) {
  par(mfrow = c(2, 1))
  plot(all5kmSessions[all5kmSessions$session == indices5km[i], ]$speed, main = paste("seance 5km - kmeans groupe 5km", indices5km[i]), type = "p", pch = 20, col = clusters5kmSessions_allSession[[indices5km[i]]], xlab = "troncon", ylab = "speed")
  
  plot(allSessions[allSessions$session == indices5km[i], ]$speed, main = paste("seance 5km - kmeans avec toutes les courses", indices5km[i]), type = "p", pch = 20, col = cluster2WithDistance[[indices5km[i]]], xlab = "troncon", ylab = "speed")
}
```


# Recherche des actions pour MDP sur groupe de 10km, 5km
```{r, include=TRUE}
histPace <- function(indicesSessions, sessions) {
  pace <- c()
  for(i in 1:length(indicesSessions)){
    if(is.numeric(sessions[[indicesSessions[i]]]$pace)){
      pace <- rbind(pace,sessions[[indicesSessions[i]]]$pace)
    }
  }
  # length(as.vector(pace))
  hist(pace, breaks=100, col=1, xlim = c(100, 500))
}

histPace(indices10km, sessions)
histPace(indices5km, sessions)
```