Now to bring on the big guns

(the big guns are curve fitting and rsquared for defyning our model's equation)

~lross2k~

analysis.R

@@ -0,0 +1,46 @@
+# Function for plotting the sensor data
+.PlotSensors  <- function(tmp, file.name) {
+  png(file=file.name, width=1280, height=720)
+  par(mar=c(5.1, 4.8, 1.1, 8.1), xpd=TRUE)
+  colores = rainbow(500)
+  plot(tmp$Cycle, tmp[,6], type='l', col=colores[100], 
+       ylim=c(min(tmp[,-(1:5)]),max(tmp[,-(1:5)])), 
+       ylab = 'Sensores', xlab = 'Ciclos', lwd=4.0)
+  leg <- c(names(tmp)[6])
+  col <- c(colores[100])
+  for (x in 7:length(tmp)) {
+    lines(tmp$Cycle, tmp[,x], type='l', 
+          col=colores[x+(30*x)], lwd=4.0)
+    leg <- rbind(leg, c(names(tmp)[x]))
+    col <- rbind(col, c(colores[x+(30*x)]))
+  }
+  legend("topright", inset=c(-0.09,0), box.col = "brown",
+         bg ="yellow", box.lwd = 2, legend=leg, fill = col)
+  dev.off()
+}
+
+.TransformDecreasing <- function(col) {
+  if (col[length(col)] < col[1]) {
+    col <- rev(col)
+  }
+  return(col)
+}
+
+.FormatMotor <- function(df) {
+  df[,-c(1:5)] <- apply(df[,-c(1:5)], 2, .TransformDecreasing)
+  df[,-c(1:5)] <- apply(df[,-c(1:5)], 2, (function(col) 
+                        predict(loess(col~df$Cycle))))
+  df[,-c(1:5)] <- data.frame(apply(df[,-c(1:5)], 2, 
+                                   (function(col) col/max(col))))
+  return(df)
+}
+
+TestMotor <- function(df, motor, file.name) {
+  tmp <- data.frame(as.list(split(df, df$Motor)[260]))
+  names(tmp) <- names(Data_1)
+  tmp <- .FormatMotor(tmp)
+
+  write.csv2(tmp, paste(file.name, '.csv', sep = ''), row.names = FALSE)
+
+  .PlotSensors(tmp, paste(file.name, '.png', sep = ''))
+}

formatting.R

@@ -0,0 +1,37 @@
+FormatSplit <- function(data, index, col.name) {
+  res <- as.data.frame(tmp[1])
+  names(res) <- col.name
+  return(res)
+}
+
+.CalcInsignificance <- function(dataFrame, rowName) {
+  res <- data.frame(as.list(apply(dataFrame, 2, (function(col) 
+    sum(quantile(col))/(quantile(col)[5]*5)))))
+  row.names(res) <- c(rowName)
+  return(res)
+}
+
+.FilterInsignificance <- function(sig, criteria) {
+  res <- apply(sig, 2, (function(col) min(col) > criteria))
+  res <- sig[,- as.array(which(res == "FALSE", arr.ind = FALSE))]
+  return(res)
+}
+
+GetInsignificance <- function(Data_1, Data_2, Data_3, 
+                              Data_4, Data_5, Data_6,
+                              criteria) {
+  res <- .CalcInsignificance(Data_1[-c(1:5)], "Subset_1")
+  res <- rbind(res, .CalcInsignificance(Data_2[-c(1:5)], "Subset_2"))
+  res <- rbind(res, .CalcInsignificance(Data_3[-c(1:5)], "Subset_3"))
+  res <- rbind(res, .CalcInsignificance(Data_4[-c(1:5)], "Subset_4"))
+  res <- rbind(res, .CalcInsignificance(Data_5[-c(1:5)], "Subset_5"))
+  res <- rbind(res, .CalcInsignificance(Data_6[-c(1:5)], "Subset_6"))
+  res <- .FilterInsignificance(res, criteria)
+  return(res)
+}
+
+RemoveByInsignificance <- function(insig,df) {
+  res <- df[,-apply(as.array(names(insig)), 1, (function(name,df) 
+    grep(paste("^",name,"$",sep=""), colnames(df))), df)]
+  return(res)
+}

main.R

@@ -1,73 +1,28 @@
 Data <- read.csv("train_data.csv")
 col.name <- names(Data)
 
+source('formatting.R')
 tmp <- split(Data, cut(Data$Height, c(-5,5,15,24,30,40,45)))
-Data_1 <- as.data.frame(tmp[1])
-Data_2 <- as.data.frame(tmp[2])
-Data_3 <- as.data.frame(tmp[3])
-Data_4 <- as.data.frame(tmp[4])
-Data_5 <- as.data.frame(tmp[5])
-Data_6 <- as.data.frame(tmp[6])
+rm(Data)
+Data_1 <- FormatSplit(tmp, 1, col.name)
+Data_2 <- FormatSplit(tmp, 2, col.name)
+Data_3 <- FormatSplit(tmp, 3, col.name)
+Data_4 <- FormatSplit(tmp, 4, col.name)
+Data_5 <- FormatSplit(tmp, 5, col.name)
+Data_6 <- FormatSplit(tmp, 6, col.name)
 rm(tmp)
-names(Data_1) <- col.name
-names(Data_2) <- col.name
-names(Data_3) <- col.name
-names(Data_4) <- col.name
-names(Data_5) <- col.name
-names(Data_6) <- col.name
 
-GetInsignificance <- function(dataFrame, rowName) {
-  res <- data.frame(as.list(apply(dataFrame, 2, (function(col) sum(quantile(col))/(quantile(col)[5]*5)))))
-  row.names(res) <- c(rowName)
-  return(res)
-}
-
-insig <- GetInsignificance(Data_1[-c(1:5)], "Subset_1")
-insig <- rbind(insig, GetInsignificance(Data_2[-c(1:5)], "Subset_2"))
-insig <- rbind(insig, GetInsignificance(Data_3[-c(1:5)], "Subset_3"))
-insig <- rbind(insig, GetInsignificance(Data_4[-c(1:5)], "Subset_4"))
-insig <- rbind(insig, GetInsignificance(Data_5[-c(1:5)], "Subset_5"))
-insig <- rbind(insig, GetInsignificance(Data_6[-c(1:5)], "Subset_6"))
-
-FilterInsignificance <- function(sig, criteria) {
-  res <- apply(sig, 2, (function(col) min(col) > criteria))
-  res <- sig[,- as.array(which(res == "FALSE", arr.ind = FALSE))]
-  return(res)
-}
-
-insig <- FilterInsignificance(insig, 0.988)
-
-RemoveByInsignificance <- function(insig,df) {
-  res <- df[,-apply(as.array(names(insig)), 1, (function(name,df) 
-                    grep(paste("^",name,"$",sep=""), colnames(df))), df)]
-  return(res)
-}
+insig <- GetInsignificance(Data_1, Data_2, Data_3, 
+                           Data_4, Data_5, Data_6,
+                           0.988)
 
 Data_1 <- RemoveByInsignificance(insig, Data_1)
 Data_2 <- RemoveByInsignificance(insig, Data_2)
 Data_3 <- RemoveByInsignificance(insig, Data_3)
 Data_4 <- RemoveByInsignificance(insig, Data_4)
 Data_5 <- RemoveByInsignificance(insig, Data_5)
 Data_6 <- RemoveByInsignificance(insig, Data_6)
+rm(insig)
 
-TransformDecreasing <- function(col) {
-  if (col[length(col)] < col[1]) {
-    col <- rev(col)
-  }
-  return(col)
-}
-
-FormatMotor <- function(df) {
-  df[,-(1:5)] <- apply(df[,-(1:5)], 2, TransformDecreasing)
-  df[,-c(1:5)] <- apply(df[,-c(1:5)], 2, (function(col) 
-                         predict(loess(col~df$Cycle))))
-  df <- data.frame(apply(df, 2, (function(col) col/max(col))))
-  return(df)
-}
-
-tmp <- data.frame(as.list(split(Data_1, Data_1$Motor)[260]))
-names(tmp) <- names(Data_1)
-tmp <- FormatMotor(tmp)
-
-source('plotting.R')
-PlotSensors(tmp, 'motor260.png')
+source('analysis.R')
+TestMotor(Data_1, 1, 'Subset_1_Regulador_60_Motor_1')

proceso.Rmd

@@ -46,43 +46,23 @@ plot(Data100$Cycle, Data100$Height)
 ```
 
 ```{r}
-tmp <- split(Data100, cut(Data100$Height, c(-10,5,15,25,39,55)))
-Data100_1 <- as.data.frame(tmp[1])
-Data100_2 <- as.data.frame(tmp[2])
-Data100_3 <- as.data.frame(tmp[3])
-Data100_4 <- as.data.frame(tmp[4])
-Data100_5 <- as.data.frame(tmp[5])
-names(Data100_1) <- names(Data)
-names(Data100_2) <- names(Data)
-names(Data100_3) <- names(Data)
-names(Data100_4) <- names(Data)
-names(Data100_5) <- names(Data)
+splittedData <- split(Data, cut(Data$Height, c(-5,5,15,24,30,40,45)))
 ```
 
 Los datos de estos sub grupos presentan comportamiento homogéneo
 
 ```{r echo=FALSE}
-plot(Data100_5$Cycle, Data100_5$Height)
+plottingData <- data.frame(splittedData[6])
+names(plottingData) <- names(Data)
+plot(plottingData$Cycle, plottingData$Height)
 ```
 
 Se ha notado en todos los gráficos generados, que a partir del ciclo 200 hay una disminución en la densidad de datos, o eso aparenta, se realiza un análisis general de cada sub set, auque primero se debe determinar cuáles datos no vale la pena analizar, algunos fáciles de descartar corresponden al motor, ciclo y regulador, ya que son valores discretos con variaciones conocidas que de momento no nos interesan, por ejemplo
 
-```{r}
-summary(Data[,-c(3:4,6:39)])
-```
-
 Se separan los datos a mostrar por regiones, ya que son muchos sensores para mostrar todos en una sóla salida, es posible observar para los datos a 60%, que la altura es casi constante, con variaciones de 0,01 unidades, el número Mach presenta variaciones ligeramente mayores, con 0,022 unidades, ambos variando hacia el tercer cuartil y el sensor 1 es constante
 
-```{r}
-summary(Data60[,-c(1,2,5,9:39)])
-```
-
 El sensor 5 es constante y el 8 sólo tiene una mínima variación en el máximo, la cual no llega a ser significativa ni para el tercer cuartil
 
-```{r}
-summary(Data60[,-c(1:8,14:39)])
-```
-
 Para simplificar la discriminación de variables, se decide emplear únicamente los cuártiles, sumando todos estos y dividiendo según el valor del primer cuartíl multiplicado por 5, para obtener un número sencillo que si es 1, significa que la variable no es significativa para el análisis
 
 ```{r}
@@ -243,6 +223,7 @@ Separamos con base en los porcentajes del regulador
 tmp <- split(Data, Data$Throttle)
 Data60_1 <- as.data.frame(tmp[1])
 Data100 <- as.data.frame(tmp[2])
+rm(tmp)
 names(Data60_1) <- names(Data)
 names(Data100) <- names(Data)
 ```