http://www.perlmonks.org?node_id=609869


in reply to Re: Issue on covariance calculation
in thread Issue on covariance calculation

I disagree. You can do random access with ASCII so long as you write fixed length records, but it does take about 50% longer to do the conversion. It's the conversion that chews up most of the time, so pack() can be a substantial time saver.

In the test below, writing floats is 10 times faster than writhing ascii. The file is half the size, so that smaller size only accounts for a factor of 2 speed-up. A factor of 5 comes from not doing the ascii conversion.

use warnings; use strict; use Time::HiRes qw( gettimeofday tv_interval ); my $size = 1000; my $last_idx = $size-1; my $float_value; my $t0; $t0 = [gettimeofday]; open ASC, '>', "asciinum.txt" or die "can't open ascii file"; for my $row (0..$last_idx) { my @result = (); my $row_len = $row+1; $float_value = $row/10_000 + 1/100_000_000; for my $col (0..$last_idx) { #$float_value = $row/10_000 + $col/100_000_000; push @result, $float_value; } print ASC join(',', @result); } print STDERR "Ascii write took -- ",tv_interval ( $t0 ),"\n"; close ASC; $t0 = [gettimeofday]; open FLOAT, '>', "floatnum.txt" or die "can't open float file"; for my $row (0..$last_idx) { my @result = (); my $row_len = $row+1; $float_value = $row/10_000 + 1/100_000_000; for my $col (0..$last_idx) { #$float_value = $row/10_000 + $col/100_000_000; push @result, $float_value; } print FLOAT pack("F$row_len", @result); } print STDERR "Float write took -- ",tv_interval ( $t0 ),"\n"; close FLOAT; # Gave the following results on my fairly slow machine, with $size = 1 +000 # Ascii write took -- 15.078125 # Float write took -- 2.778736