int main(int argc, char* argv[])
{
if(argc == 1)
{
cout << "You must pass some parameters. Run \"./KNN help\" to take some tips." << endl;
}
if(argc == 2 && strcmp(argv[1],"help") == 0)
{
cout << "This is a implemantation of KNN algorithm. Case of handwritten numbers." << endl << endl;
cout << "[USAGE]: ./KNN [K] [dataTraineFile] [dataTestFile]" << endl;
}
if(argc == 4)
{
int k = atoi(argv[1]);
string dataTraineFile = argv[2];
string dataTestFile = argv[3];
KNN knn = KNN(k);
double precision = knn.analysisData(dataTraineFile, dataTestFile);
cout << "Precision of this test data: " << precision << "%" << endl << endl;
cout << "You can view details of this execution in result.txt" << endl;
}
return 0;
}
int main() {
cout << "!!!DataTeam project start!!!" << endl;
// int totalTrainRecords = 878049;
int totalTrainRecords = 100000;
int totalTestRecords = 884261;
string predictFieldName = "Category";
string trainFileName = string("../DataTeam/files/train.csv");
string testFileName = string("../DataTeam/files/test.csv");
string trainNewFileName = string("../DataTeam/files/trainNew.csv");
string submissionFileName = string("../DataTeam/files/submission.csv");
// Preparar el archivo para que funcione al pasarlo al NaiveBayes
FileManager fileManager = FileManager(totalTrainRecords, trainFileName, trainNewFileName);
fileManager.process();
// Pasarle el archivo con los datos de los registros como numericos
// NaiveBayes naiveBayes = NaiveBayes(predictFieldName, fileManager.getSetterData());
// naiveBayes.train(totalTrainRecords, trainNewFileName);
// naiveBayes.test(totalTestRecords, testFileName, submissionFileName);
KNN knn = KNN(50, fileManager.getSetterData());
knn.aplicarKNN(trainNewFileName,testFileName,submissionFileName);
cout << "Finish run app" << endl;
return 0;
}
// Tests the default constructor
TEST(KNN, Constructor) {
KNN knn;
//Check the type matches
EXPECT_TRUE( knn.getClassifierType() == KNN::getId() );
//Check the module is not trained
EXPECT_TRUE( !knn.getTrained() );
}
double predict(Database* db, int roadIndex, const Item& item)
{
double jc = (double)db->getJamCount(roadIndex);
double var = cluster->getVar(db, roadIndex);
//double A = 480;
//double B = 520;
double A = 780;
double B = 820;
if ( midAvg == 0 )
{
midAvg = new KNN(db, LOWER_LIMIT, UPPER_LIMIT, cluster);
midAvg->initialize(db);
}
if ( smallAvg == 0 )
{
smallAvg = new SmallAverage(db, cluster, LOWER_LIMIT);
smallAvg->initialize(db);
}
/* if ( jc >= 30 && var >= 25.0 )
{
return midAvg->predict(db, roadIndex, item);
}*/
if ( jc > B )
{
return midAvg->predict(db, roadIndex, item);
}
// return 50.0;
if ( jc < A )
{
return smallAvg->predict(db, roadIndex, item);
}
double mid = midAvg->predict(db, roadIndex, item);
double small = smallAvg->predict(db, roadIndex, item);
double K = (jc - A)/(B - A);
double res = small*(1.0 - K) + mid*K;
return res;
}
void iris()
{
vector<vector<double>> examples;
vector<string> target;
string data_file_name = "iris.data";
ifstream input(data_file_name.c_str());
string line, tar;
double a, b, c ,d;
while(getline(input, line))
{
istringstream iss(line);
vector<double> v;
iss>>a>>b>>c>>d>>tar;
examples.push_back({a, b, c, d});
target.push_back(tar);
}
input.close();
vector<vector<double>> test, train;
vector<string> tar_test, tar_train;
for(int i = 0; i < (int) examples.size(); ++i)
{
if(i%50 >= 40) test.push_back(examples[i]), tar_test.push_back(target[i]);
else train.push_back(examples[i]), tar_train.push_back(target[i]);
}
KNN knn;
knn.set_data(train, tar_train, 3);
for(int i = 0; i < (int)test.size(); ++i)
{
auto ans = knn.classify(test[i]);
cout<<'<';
for(auto& j:test[i]) cout<<j<<", ";
cout<<tar_test[i]<<">\t the answer is "<<ans<<endl;
}
}
// Tests the learning algorithm on a basic dataset
TEST(KNN, TrainBasicDataset) {
KNN knn;
//Check the module is not trained
EXPECT_TRUE( !knn.getTrained() );
//Generate a basic dataset
const UINT numSamples = 1000;
const UINT numClasses = 10;
const UINT numDimensions = 10;
ClassificationData::generateGaussDataset( "gauss_data.csv", numSamples, numClasses, numDimensions, 10, 1 );
ClassificationData trainingData;
EXPECT_TRUE( trainingData.load( "gauss_data.csv" ) );
ClassificationData testData = trainingData.split( 50 );
//Train the classifier
EXPECT_TRUE( knn.train( trainingData ) );
EXPECT_TRUE( knn.getTrained() );
EXPECT_TRUE( knn.print() );
for(UINT i=0; i<testData.getNumSamples(); i++){
EXPECT_TRUE( knn.predict( testData[i].getSample() ) );
}
EXPECT_TRUE( knn.save( "knn_model.grt" ) );
knn.clear();
EXPECT_TRUE( !knn.getTrained() );
EXPECT_TRUE( knn.load( "knn_model.grt" ) );
EXPECT_TRUE( knn.getTrained() );
for(UINT i=0; i<testData.getNumSamples(); i++){
EXPECT_TRUE( knn.predict( testData[i].getSample() ) );
}
}
void incrementallyAddPoints(vertex** v, int n, vertex* start) {
int numRounds = Exp::getNumRounds();
numRounds = numRounds <= 0 ? 100 : numRounds;
// various structures needed for each parallel insertion
//int maxR = (int) (n/100) + 1; // maximum number to try in parallel
int maxR = (int) (n/numRounds) + 1; // maximum number to try in parallel
Qs *qqs = newA(Qs,maxR);
Qs **qs = newA(Qs*,maxR);
for (int i=0; i < maxR; i++) {
qs[i] = new (&qqs[i]) Qs;
}
simplex *t = newA(simplex,maxR);
bool *flags = newA(bool,maxR);
vertex** h = newA(vertex*,maxR);
// create a point location structure
typedef kNearestNeighbor<vertex,1> KNN;
KNN knn = KNN(&start, 1);
int multiplier = 8; // when to regenerate
int nextNN = multiplier;
int top=n; int rounds = 0; int failed = 0;
// process all vertices starting just below top
while(top > 0) {
// every once in a while create a new point location
// structure using all points inserted so far
if ((n-top)>=nextNN && (n-top) < n/multiplier) {
knn.del();
knn = KNN(v+top, n-top);
nextNN = nextNN*multiplier;
}
// determine how many vertices to try in parallel
//int cnt = 1 + (n-top)/100; // 100 is pulled out of a hat
int cnt = 1 + (n-top)/numRounds;
cnt = (cnt > maxR) ? maxR : cnt;
cnt = (cnt > top) ? top : cnt;
vertex **vv = v+top-cnt;
// for trial vertices find containing triangle, determine cavity
// and reserve vertices on boundary of cavity
// parallel_for (int j = 0; j < cnt; j++) {
parallel_doall(int, j, 0, cnt) {
vertex *u = knn.nearest(vv[j]);
t[j] = find(vv[j],simplex(u->t,0));
reserveForInsert(vv[j],t[j],qs[j]);
} parallel_doall_end
// For trial vertices check if they own their boundary and
// update mesh if so. flags[i] is 1 if failed (need to retry)
// parallel_for (int j = 0; j < cnt; j++) {
parallel_doall(int, j, 0, cnt) {
flags[j] = insert(vv[j],t[j],qs[j]);
} parallel_doall_end
// Pack failed vertices back onto Q and successful
// ones up above (needed for point location structure)
int k = sequence::pack(vv,h,flags,cnt);
// parallel_for (int j = 0; j < cnt; j++) flags[j] = !flags[j];
parallel_doall(int, j, 0, cnt) { flags[j] = !flags[j]; } parallel_doall_end
sequence::pack(vv,h+k,flags,cnt);
// parallel_for (int j = 0; j < cnt; j++) vv[j] = h[j];
parallel_doall(int, j, 0, cnt) { vv[j] = h[j]; } parallel_doall_end
failed += k;
top = top-cnt+k; // adjust top, accounting for failed vertices
rounds++;
}
knn.del();
free(qqs); free(qs); free(t); free(flags); free(h);
cout << "n=" << n << " Total retries=" << failed
<< " Total rounds=" << rounds << endl;
}
