final_project_code.Rmd

---
title: "final_project"
author: "Dylan Rose"
date: "September 24, 2015"
output: html_document
---

```{r,echo=FALSE,message=FALSE}
setwd("~/Dropbox/extra_courses/jhu_practical_machine_learning")
rm(list=ls())
training_data<-read.csv("./pml-training.csv",na.strings=c("#DIV/0!","NA"))
testing_data<-read.csv("./pml-testing.csv",na.strings=c("#DIV/0!","NA"))

```

Assume we've got the training and test data loaded into data frames called "training_data" and "testing_data" already. All "NA"s, "#DIV/0!" and "" records have been labeled as missing values. I've also loaded some key libraries (caret, dplyr,ggplot). 
```{r,echo=FALSE,message=FALSE}
library(caret)
library(ggplot2)
library(dplyr)
library(plyr)
library(e1071)
```

Most of these rows have significant amounts of missing data, so we can probably afford to dump them. 
```{r}
#Drop any columns containing missing values
training_data <- training_data[,colSums(is.na(training_data)) == 0]
testing_data <- testing_data[,colSums(is.na(testing_data))==0]

#Also some dump some probably extraneous factor variables
training_data <- select(training_data,-(X:num_window))
testing_data <- select(testing_data,-(X:num_window))
```

Next up is a 10-fold cross-validated SVM with a radial kernel for the classification. Because the classes are not the same size, each class will be fitted with a weight proportional to its percentage of the total data.

We'll also run this model on a training sub-set and test it on the validation set as a final sanity check.

```{r}
partition<-createDataPartition(y=training_data$classe,p=0.7,list=FALSE)
training_sub<-training_data[partition,]
validation_sub<-training_data[-partition,]
wts<-table(training_sub$classe)/100
svm_model<-svm(classe~.,data=training_sub,cross=10,class.weights=wts)
svm_model_predictions<-predict(svm_model,validation_sub)
confusionMatrix(svm_model_predictions,validation_sub$classe)
```

This classifier is ~99% accurate, so we'll stop here and use it to generate the files for uploading to the submission portion of the assignment:

```{r}
answers<-predict(svm_model,testing_data)

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(answers)
```