library(seqinr)
library(protr)
library(party) # never actually used
library(BioSeqClass)
library(randomForest)
library(lme4)
library(caret)
library(adabag)
library(tree)
library(rpart)
library(DMwR)
library(e1071)
### "Refresh" seed - not really necessary:
set.seed(0)
### Read in ready-to-go data:
DATA = read.csv("C:/Users/jdmunyon/Desktop/Dropbox/Senior_Project/data_and_code/50_50.csv", header = TRUE, row.names = 1)
### Set a seed for reproducability (since random sampling is coming up within the cross-validation):
set.seed(69)
### Determine optimal SVM parameters using ten-fold cross-validation
tune = tune.svm(x = DATA[,-1], y = DATA[,1], gamma = 2^(seq(from = -5, to = 15, by = 2)), cost = 2^(seq(from = -15, to = 3, by = 2)), tunecontrol = tune.control(sampling = "cross", cross = 10))
### Save the optimal parameters:
parameters = unlist(tune$best.parameters)
### Save a list of numbers, from 1 to the number of proteins in the dataset:
nums = 1:nrow(DATA)
### Create (3002) SVM models, each one using all but one left-out protein
### Then predict the left-out protein's location
### (this is jackknife validation):
svms = lapply(nums, function(z){
  a = svm(x = DATA[-z,-1], y = DATA[-z,1], cost = parameters["cost"], gamma = parameters["gamma"])
  b = predict(object = a, newdata = DATA[z,-1])
  return(prediction = b)
})
### Unlist svms (turn R object from a list to a vector):
svms = unlist(svms)
### Save actual locations of full dataset:
ACTUAL = DATA[,1]
### Calculate error rate of classification:
error_svms = 1 - sum(svms == ACTUAL)/nrow(DATA)
### Create confusion matrix and performance statistics for each location:
confusion_svms = confusionMatrix(data = svms, reference = ACTUAL)
### Save SVM results to machine
write.csv(as.data.frame.matrix(confusion_svms$table), file = "C:/Users/jdmunyon/Desktop/Dropbox/Senior_Project/data_and_code/results/50_50/svms_confusion.csv")
write.csv(as.data.frame.matrix(confusion_svms$byClass), file = "C:/Users/jdmunyon/Desktop/Dropbox/Senior_Project/data_and_code/results/50_50/svms_stats.csv")
write.table(error_svms, file = "C:/Users/jdmunyon/Desktop/Dropbox/Senior_Project/data_and_code/results/50_50/svms_stats.csv", append = TRUE)
### Save R workspace
save.image("C:/Users/jdmunyon/Desktop/Dropbox/Senior_Project/data_and_code/results/50_50/SVM_results_from_lab.RData")