matrix * copy_matrix_subset(matrix *m, struct slInt * desiredRows, struct slInt * desiredCols)
{
int i = 0, j = 0;
//if desiredRows or desiredCols is NULL that means grab them all.
struct slInt *rows = NULL, *cols = NULL;
if(desiredRows == NULL)
rows = list_indices(m->rows);
else
rows = desiredRows;
if(desiredCols == NULL)
cols = list_indices(m->cols);
else
cols = desiredCols;
struct slInt * curr_row = rows;
struct slInt * curr_col = cols;
int subsetRows = count_indices(rows);
int subsetCols = count_indices(cols);
matrix * result = init_matrix(subsetRows, subsetCols);
//if the source matrix has labels, make space for them in the result matrix
if(m->labels)
{
//copy labels
i = 0;
while(curr_row != NULL)
{
safef(result->rowLabels[i++],MAX_LABEL, "%s", m->rowLabels[curr_row->val]);
curr_row = curr_row->next;
}
curr_row = rows;
i = 0;
while(curr_col != NULL)
{
safef(result->colLabels[i++],MAX_LABEL,"%s", m->colLabels[curr_col->val]);
curr_col = curr_col->next;
}
curr_col = cols;
result->labels=1;
}
//run over your selected rows and cols, copying from source matrix
i=0;i=0;
while(curr_row != NULL)
{
while(curr_col != NULL)
{
result->graph[i][j++] = m->graph[curr_row->val][curr_col->val];
curr_col = curr_col->next;
}
i++;
j = 0;
curr_col = cols;
curr_row = curr_row->next;
}
if(desiredRows == NULL)
slFreeList(&rows);
if(desiredCols == NULL)
slFreeList(&cols);
return result;
}
int main(int argc, char **argv)
{
st_manager holder;
/*options */
char c;
int disp_f = 0; /* shows patterns in coloured text */
int num_f = 0; /* displays the number of occurences */
int ind_f = 0; /* lists all the indices[from 0] of occurence of pattern */
if (argc < 3) {
print_usage(argv[0]);
return EXIT_FAILURE;
}
FILE *inp_file = fopen(argv[1], "r");
char *pattern = argv[2];
/* parse arguments */
while ((c = getopt(argc, argv, "dni")) != -1)
switch(c)
{
case 'd':
disp_f = 1;
break;
case 'n':
num_f = 1;
break;
case 'i':
ind_f = 1;
break;
default:
break;
}
/*
FILE *inp_file = fopen(argv[1], "r");
char *pattern = argv[2];
*/
holder = pat_search(inp_file, pattern);
if (num_f) { printf("occurences: %d \n", holder.num_occur); }
if (disp_f) {
//rewind file as it has already been read
fseek(inp_file, 0, SEEK_SET);
display_pattern(inp_file, pattern, holder);
}
if (ind_f) {
list_indices(holder);
}
fclose(inp_file);
return EXIT_SUCCESS;
}
int main(int argc, char *argv[])
{
if(argc < 2)
usage();
char * configFile = argv[1];
struct hash * config = raReadSingle(configFile);
matrix * data = NULL;
if(sameString("bioInt", hashMustFindVal(config, "inputType")))
{
char * profile = hashMustFindVal(config, "profile"); //TODO: make this optional so if it's not set it defaults
char * db = hashMustFindVal(config, "db");
struct sqlConnection *conn = hAllocConnProfile(profile, db);
char *tableName = hashMustFindVal(config, "tableName");
data = bioInt_fill_matrix(conn, tableName);
hFreeConn(&conn);
}
else if(sameString("flatfiles", hashMustFindVal(config, "inputType")))
{
char * dataFilepath = hashMustFindVal(config, "dataFilepath");
FILE * dataFile;
dataFile = fopen(dataFilepath, "r");
if(dataFile == NULL)
errAbort("ERROR: Couldn't open the file \"%s\"\n", dataFilepath);
data = f_fill_matrix(dataFile, 1);
fclose(dataFile);
}
else
errAbort("Unsupported input type");
int itemsInFold = -1;
int folds = foldsCountFromConfig(config);
char * cv = hashMustFindVal(config, "crossValidation");
if(sameString("k-fold", cv))
itemsInFold = floor(data->cols/folds);
else if(sameString("loo", cv))
itemsInFold = 1;
if(folds == -1 || itemsInFold == -1)
errAbort("Couldn't assign folds or itemsInFold\n");
if(hashFindVal(config, "excludeList"))
{
matrix * trimmedData = filterColumnsByExcludeList(config, data);
free_matrix(data);
data = trimmedData;
}
struct slInt *list = list_indices(data->cols);
int split, splits = splitsCountFromConfig(config);
matrix * foldReports = NULL;
for(split = 1; split <= splits; split++)
{
struct slInt *shuffledList = seeded_shuffle_indices(list, split);
matrix * tmp = reportFolds(config, data, folds,itemsInFold, shuffledList);
if(split == 1)
{
foldReports = copy_matrix(tmp);
}
else
{
matrix * tmp2 = append_matrices(foldReports, tmp, 2);
free_matrix(foldReports);
foldReports = copy_matrix(tmp2);
}
slFreeList(&shuffledList);
}
fprint_discreteMatrix(stdout, foldReports);
free_matrix(foldReports);
freeHash(&config);
slFreeList(&list);
return 0;
}
/**** WEKA specific functions *********/
matrix * WEKApopulateAccuracyMatrix(struct hash * config, int split, int fold)
{
char * trainingDir = hashMustFindVal(config, "trainingDir");
char * validationDir = hashMustFindVal(config, "validationDir");
char * modelDir = hashMustFindVal(config, "modelDir");
char filename[256];
//cat togetehr the training and validation KH values and record which were used to train
safef(filename, sizeof(filename), "%s/split%02d/fold%02d/data.arff", trainingDir, split, fold);
matrix * trMetadata = WEKAtoMetadataMatrix(filename);
safef(filename, sizeof(filename), "%s/split%02d/fold%02d/data.arff", validationDir, split, fold);
matrix * valMetadata = WEKAtoMetadataMatrix(filename);
matrix * metadata = append_matrices(trMetadata, valMetadata, 1);
struct slInt * trainingList = list_indices(trMetadata->cols);
//create a labeled matrix for results to be stored in
matrix * result = init_matrix(2, metadata->cols);
safef(result->rowLabels[0], MAX_LABEL, "trainingAccuracies");
safef(result->rowLabels[1], MAX_LABEL, "testingAccuracies");
copy_matrix_labels(result, metadata, 2,2);
result->labels=1;
//read the results from file
safef(filename, sizeof(filename), "%s/split%02d/fold%02d/weka.training.results", modelDir, split, fold);
FILE * fp = fopen(filename, "r");
if(fp == NULL)
errAbort("Couldn't open %s for reading.", filename);
//advance the cursor to where data starts
char * line;
while( (line = readLine(fp)) && line != NULL)
{
if(strstr(line, "inst#") != NULL)
break;
}
//read each result and save to results matrix
int i;
for(i = 0; i < trMetadata->cols && (line = readLine(fp)) != NULL; i++)
{
if(strstr(line, ":?") == NULL)
{
if(strstr(line, " + ") == NULL)
result->graph[0][i] = 1;
else
result->graph[0][i] = 0;
}
}
safef(filename, sizeof(filename), "%s/split%02d/fold%02d/weka.validation.results", modelDir, split, fold);
fp = fopen(filename, "r");
if(fp == NULL)
errAbort("Couldn't open %s for reading.", filename);
//advance the cursor to where data starts
while( (line = readLine(fp)) && line != NULL)
{
if(strstr(line, "inst#") != NULL)
break;
}
//read each result and save to results matrix
for(i = i; i < result->cols && (line = readLine(fp)) != NULL; i++)
{
if(strstr(line, ":?") == NULL)
{
if(strstr(line, " + ") == NULL)
result->graph[1][i] = 1;
else
result->graph[1][i] = 0;
}
}
free_matrix(trMetadata);
free_matrix(valMetadata);
free_matrix(metadata);
slFreeList(&trainingList);
return result;
}
matrix * SVMgetSampleAccuracy(struct hash *config)
/*Read all the folds and calculate training and testing accuracies from best models*/
{
char * trainingDir = hashMustFindVal(config, "trainingDir");
char * validationDir = hashMustFindVal(config, "validationDir");
char * modelDir = hashMustFindVal(config, "modelDir");
int fold, folds = foldsCountFromDataDir(config);
int split, splits = splitsCountFromDataDir(config);
matrix * accuracies = NULL;
char filename[1024];
for(split = 1; split <= splits; split++)
{
for(fold = 1; fold <= folds; fold++)
{
//cat togetehr the training and validation KH values and record which were used to train
safef(filename, sizeof(filename), "%s/split%02d/fold%02d/data.svm", trainingDir, split, fold);
matrix * trMetadata = SVMtoMetadataMatrix(filename);
safef(filename, sizeof(filename), "%s/split%02d/fold%02d/data.svm", validationDir, split, fold);
matrix * valMetadata = SVMtoMetadataMatrix(filename);
struct slInt * trainingList = list_indices(trMetadata->cols);
matrix * metadata = append_matrices(trMetadata, valMetadata, 1);
//cat together the guesses from SVM
safef(filename, sizeof(filename), "%s/split%02d/fold%02d/svm.training.results", modelDir, split, fold);
matrix * trainingPred = SVMpopulatePredictionsMatrix(filename);
safef(trainingPred->rowLabels[0],MAX_LABEL, "Prediction");
copy_matrix_labels(trainingPred, trMetadata, 2,2);
trainingPred->labels=1;
safef(filename, sizeof(filename), "%s/split%02d/fold%02d/svm.validation.results", modelDir, split, fold);
matrix * testingPred = SVMpopulatePredictionsMatrix(filename);
safef(testingPred->rowLabels[0], MAX_LABEL, "Prediction");
copy_matrix_labels(testingPred, valMetadata, 2,2);
testingPred->labels=1;
matrix * predictions = append_matrices(trainingPred, testingPred, 1);
//get accuracies
matrix * accuraciesInFold = SVMpopulateAccuracyMatrix(predictions, metadata, trainingList);
//add the accuracies to the running totals
if(split == 1 && fold == 1)
accuracies = copy_matrix(accuraciesInFold);
else
add_matrices_by_colLabel(accuracies, accuraciesInFold);
//clean up
free_matrix(trMetadata);
free_matrix(valMetadata);
free_matrix(metadata);
free_matrix(trainingPred);
free_matrix(testingPred);
free_matrix(predictions);
free_matrix(accuraciesInFold);
slFreeList(&trainingList);
}
}
//normalize accuracies over number of splits and folds
int i;
for(i = 0; i < accuracies->cols; i++)
{
if(accuracies->graph[0][i] != NULL_FLAG)
accuracies->graph[0][i] = (accuracies->graph[0][i] / ((folds-1) * splits));
if(accuracies->graph[1][i] != NULL_FLAG)
accuracies->graph[1][i] = (accuracies->graph[1][i] / (1 * splits));
}
return accuracies;
}
matrix * NMFgetSampleAccuracy(struct hash *config)
/*Read all the folds and calculate training and testing accuracies from best models*/
{
char * trainingDir = hashMustFindVal(config, "trainingDir");
char * validationDir = hashMustFindVal(config, "validationDir");
char * modelDir = hashMustFindVal(config, "modelDir");
int fold, folds = foldsCountFromDataDir(config);
int split, splits = splitsCountFromDataDir(config);
matrix * accuracies = NULL;
char filename[256];
FILE * fp;
for(split = 1; split <= splits; split++)
{
for(fold = 1; fold <= folds; fold++)
{
//cat togetehr the training and validation KH values and record which were used to train
safef(filename, sizeof(filename), "%s/split%02d/fold%02d/metadata.tab", trainingDir, split, fold);
fp = fopen(filename, "r");
if(fp == NULL)
errAbort("Couldn't open file %s\n", filename);
matrix * trMetadata = f_fill_matrix(fp, 1);
fclose(fp);
safef(filename, sizeof(filename), "%s/split%02d/fold%02d/metadata.tab", validationDir, split, fold);
fp = fopen(filename, "r");
if(fp == NULL)
errAbort("Couldn't open file %s\n", filename);
matrix * valMetadata = f_fill_matrix(fp, 1);
fclose(fp);
struct slInt * trainingList = list_indices(trMetadata->cols);
matrix * metadata = append_matrices(trMetadata, valMetadata, 1);
safef(filename, sizeof(filename), "%s/split%02d/fold%02d/NMFpredictor.training.results", modelDir, split, fold);
fp = fopen(filename , "r");
if(!fp)
errAbort("Couldn't open training results file %s", filename);
matrix * trainingPred = f_fill_matrix(fp, 1);
fclose(fp);
safef(filename, sizeof(filename), "%s/split%02d/fold%02d/NMFpredictor.validation.results", modelDir, split, fold);
fp = fopen(filename , "r");
if(!fp)
errAbort("Couldn't open validation results file %s", filename);
matrix * valPred = f_fill_matrix(fp, 1);
fclose(fp);
//calc the accuracy by sample
matrix * predictions = append_matrices(trainingPred, valPred, 1);
matrix * accuraciesInFold = NMFpopulateAccuracyMatrix(predictions, metadata, trainingList);
//add the accuracies to the running totals
if(split == 1 && fold == 1)
accuracies = copy_matrix(accuraciesInFold);
else
add_matrices_by_colLabel(accuracies, accuraciesInFold);
//clean up
free_matrix(trainingPred);
free_matrix(valPred);
free_matrix(predictions);
free_matrix(trMetadata);
free_matrix(valMetadata);
free_matrix(metadata);
free_matrix(accuraciesInFold);
}
}
//normalize accuracies over number of splits and folds
int i;
for(i = 0; i < accuracies->cols; i++)
{
if(accuracies->graph[0][i] != NULL_FLAG)
accuracies->graph[0][i] = (accuracies->graph[0][i] / ((folds-1) * splits));
if(accuracies->graph[1][i] != NULL_FLAG)
accuracies->graph[1][i] = (accuracies->graph[1][i] / (1 * splits));
}
return accuracies;
}
