High performance auto-correlation of fMRI data
- Intial version supports .1D | .1D.gz format as exported from afni's SUMA program.
- Uses multi-threaded BLAS to do the correlation
- Allows one to provide an upper RAM limit
From the help:
usage: pnicorr file.1D[.gz] [options]
-norm|-nonorm: normalize rows (or don't). default is to normalize.
-mem=MB: memory (MB)
Smaller means more file activity; computing is done
in stages. Default is the same as -mem=4000 (4G)
-iotype=1D|1Dgz|mat: 1D: same as input (SUMA ascii) [default for input]
1Dgz: same as input, gzipped
mat: matlab .mat file [default for output]
Really the input can be any matrix of values, where each row is a time-series to be correlated with all other rows. The format is:
[# comments]
[int int int ] fl.oat fl.oat ...
([] means optional.)
Note that only the metadata columns must be ints and must have no decimal. Data must have a decimal, even if it is just .0
. This is the quick and dirty way the file is parsed. This is the format saved by the 3dVol2Surf
program from afni. Other input formats will be supported asap.
Each row is a timeseries with its values separated by whitespace. The rows can be any timeseries data. No spatial information is used (nor available.)