aboutsummaryrefslogtreecommitdiff
path: root/src/c/quikr.c
diff options
context:
space:
mode:
authorCalvin <calvin@EESI>2013-06-11 10:24:46 -0400
committerCalvin <calvin@EESI>2013-06-11 10:24:46 -0400
commit9aafbe618e8ba48c8afbed30497b2ad7e3378b8a (patch)
tree72b3dd9aed689a01d1a24bb5705f823adb38cfef /src/c/quikr.c
parenta07e692d4ea69466a26c5cbdb3176046b71d761e (diff)
use longs instead of ints, fix a mistake in the --help
Diffstat (limited to 'src/c/quikr.c')
-rw-r--r--src/c/quikr.c14
1 files changed, 7 insertions, 7 deletions
diff --git a/src/c/quikr.c b/src/c/quikr.c
index 17d6480..f588c5f 100644
--- a/src/c/quikr.c
+++ b/src/c/quikr.c
@@ -12,7 +12,7 @@
#include "quikr_functions.h"
#define sensing_matrix(i,j) (sensing_matrix[width*i + j])
-#define USAGE "Usage:\n\tmultifasta_to_otu [OPTION...] - Calculate estimated frequencies of bacteria in a sample.\n\nOptions:\n\n-i, --input\n\tthe sample's fasta file of NGS READS (fasta format)\n\n-f, --sensing-fasta\n\tlocation of the fasta file database used to create the sensing matrix (fasta format)\n\n-s, --sensing-matrix\n\t location of the sensing matrix. (trained from quikr_train)\n\n-k, --kmer\n\tspecify what size of kmer to use. (default value is 6)\n\n-l, --lambda\n\tlambda value to use. (default value is 10000)\n\n-o, --output\n\tOTU_FRACTION_PRESENT a vector representing the percentage of database sequence's presence in sample. (csv output)\n\n-v, --verbose\n\tverbose mode.\n\n-d, --debug\n\tdebug mode, read manpage for more details"
+#define USAGE "Usage:\n\tquikr [OPTION...] - Calculate estimated frequencies of bacteria in a sample.\n\nOptions:\n\n-i, --input\n\tthe sample's fasta file of NGS READS (fasta format)\n\n-f, --sensing-fasta\n\tlocation of the fasta file database used to create the sensing matrix (fasta format)\n\n-s, --sensing-matrix\n\t location of the sensing matrix. (trained from quikr_train)\n\n-k, --kmer\n\tspecify what size of kmer to use. (default value is 6)\n\n-l, --lambda\n\tlambda value to use. (default value is 10000)\n\n-o, --output\n\tOTU_FRACTION_PRESENT a vector representing the percentage of database sequence's presence in sample. (csv output)\n\n-v, --verbose\n\tverbose mode.\n\n-d, --debug\n\tdebug mode, read manpage for more details"
int main(int argc, char **argv) {
@@ -24,8 +24,8 @@ int main(int argc, char **argv) {
char *sensing_fasta_filename = NULL;
char *output_filename = NULL;
- int x = 0;
- int y = 0;
+ long x = 0;
+ long y = 0;
int verbose = 0;
int debug = 0;
@@ -33,8 +33,8 @@ int main(int argc, char **argv) {
int lambda = 10000;
int kmer = 6;
- int width = 0;
- int sequences = 0;
+ long width = 0;
+ long sequences = 0;
while (1) {
static struct option long_options[] = {
@@ -144,7 +144,7 @@ int main(int argc, char **argv) {
}
if(verbose) {
- printf("width: %d\nsequences %d\n", width, sequences);
+ printf("width: %ld\nsequences %ld\n", width, sequences);
}
double *sensing_matrix = load_sensing_matrix(sensing_matrix_filename, sequences, width);
@@ -157,7 +157,7 @@ int main(int argc, char **argv) {
}
}
- for(x= 0; x < sequences; x++) {
+ for(x = 0; x < sequences; x++) {
sensing_matrix(x, 0) = 1.0;
}
// normalize our count_matrix