ml-project-writeup.Rmd

---
title: "Assignment Write up Human activity prediction"
author: "Leo Lin"
date: "2014年9月21日"
output: html_document
---

This project is aimed to fit a Model, which captures the way of the persons are doing weight lifting. The model should can predict which class user belong to it (A,B,C,D or E), through the data from sensors.

Note: Notice that both of training data and testing data has lots of nearZeroData, so first of all, the NA and near Zero column should be removed.

Steps to build the model:

1. Load the training datat from .csv file.
2. Use nearZeroVar to remove near zero column.
3. Count the NA of each column, remove the column which has too much NA value.
4. Try to use Random Forest to fit the model

```{r}
library(caret);library(kernlab)
library(lattice)
library(ggplot2)

# load data
trainData <- read.csv("pml-training.csv",na.strings=c("NA",""))

#cleanup data
nsv <- nearZeroVar(trainData, saveMetrics = TRUE)

cleanTrainData <- trainData[,-nsv$nzv]

#cleanup NA which's left.
NAs <- apply(cleanTrainData,2,function(x) {sum(is.na(x))}) 

cleanTrainData <- cleanTrainData[,which(NAs == 0)]

trainIndex <- createDataPartition(y = cleanTrainData$classe, p=0.5,list=FALSE) 
trainingData <- cleanTrainData[trainIndex,]
vtestingData <- cleanTrainData[-trainIndex,]

#remove no used column
removeIndex <- grep("X|user_name|timestamp|new_window",names(trainingData))
trainingData<- trainingData[,-removeIndex]
vtestingData<- vtestingData[,-removeIndex]

#use parallel processing
set.seed(1235)
library(doMC)
registerDoMC(cores = 4)
modFit <- train(trainingData$classe ~.,data = trainingData,method="rf", parallel=TRUE)


#Predict
predictions <- predict(modFit, newdata = vtestingData)
```
5. Using cross vailidation data to estimate the predict error.
```{r}

#Estimate cross validataion error

1-(sum(predictions==vtestingData$classe)/dim(vtestingData)[1])
```

Now we can use the modFit to predict pml-testing.csv data.

```{r}
#predict pml-testing
predictData <- read.csv("pml-testing.csv",na.strings=c("NA",""))

nsv <- nearZeroVar(predictData, saveMetrics = TRUE)
cleanPredictData <- predictData[,-nsv$nzv]

NAs <- apply(cleanPredictData,2,function(x) {sum(is.na(x))}) 
cleanPredictData <- cleanPredictData[,which(NAs == 0)]

removeIndex <- grep("user_name|timestamp|new_window",names(cleanPredictData))
predictingData <- cleanPredictData[,-removeIndex]


predictions1 <- predict(modFit, newdata = predictingData)

data.frame(predictData$user_name, predictions1)
```



```{r}
#assignment Submission


pml_write_files = function(x){
  n = length(x)
  for(i in 1:n){
    filename = paste0("problem_id_",i,".txt")
    write.table(x[i],file=filename,quote=FALSE,row.names=FALSE,col.names=FALSE)
  }
}

pml_write_files(predictions1)

```

plots:

```{r, echo=FALSE}

```