huge-count.pl - Divide huge text into pieces and run count.pl separately on each (and then combine)
Runs count.pl efficiently on a huge data.
huge-count.pl [OPTIONS] DESTINATION [SOURCE]+
Input to huge-count.pl should be a -
item 2. Single flat directory containing multiple plain text files
A complete path to a writable directory to which huge-count.pl can write all intermediate and final output files. If DESTINATION does not exist, a new directory is created, otherwise, the current directory is simply used for writing the output files.
NOTE: If DESTINATION already exists and if the names of some of the existing files in DESTINATION clash with the names of the output files created by huge-count, these files will be over-written w/o prompting user.
This option should be specified when SOURCE is a single plain file. huge-count will divide the given SOURCE file into P (approximately) equal parts, will run count.pl separately on each part and will then recombine the bigram counts from all these intermediate result files into a single bigram output that shows bigram counts in SOURCE.
If SOURCE file contains M lines, each part created with --split P will contain approximately M/P lines. Value of P should be chosen such that count.pl can be efficiently run on any part containing M/P lines from SOURCE. As #words/line differ from files to files, it is recommended that P should be large enough so that each part will contain at most million words in total.
Specify a file containing Perl regular expressions that define the tokenization scheme for counting. This will be provided to count.pl's --token option.
Specify a file containing Perl regular expressions of non-token sequences that are removed prior to tokenization. This will be provided to the count.pl's --nontoken option.
Specify a file of Perl regex/s containing the list of stop words to be omitted from the output BIGRAMS. Stop list can be used in two modes -
AND mode declared with '@stop.mode = AND' on the 1st line of the STOPFILE
OR mode declared using '@stop.mode = OR' on the 1st line of the STOPFILE.
In AND mode, bigrams whose both constituent words are stop words are removed while, in OR mode, bigrams whose either or both constituent words are stopwords are removed from the output.
Tokens appearing within W positions from each other (with at most W-2 intervening words) will form bigrams. Same as count.pl's --window option.
Bigrams with counts less than L in the entire SOURCE data are removed from the sample. The counts of the removed bigrams are not counted in any marginal totals. This has same effect as count.pl's --remove option.
Bigrams with counts less than F in the entire SOURCE are not displayed. The counts of the skipped bigrams ARE counted in the marginal totals. In other words, --frequency in huge-count.pl has same effect as the count.pl's --frequency option.
Switches ON the --newLine option in count.pl. This will prevent bigrams from spanning across the lines.
Displays this message.
Displays the version information.
# create output dir if(!-e DESTINATION) then mkdir DESTINATION;
Split SOURCE into P smaller files (as specified by --split P). These files are created in the DESTINATION directory and their names are formatted as SOURCE1, SOURCE2, ... SOURCEP.
Run count.pl on each of the P smaller files. The count outputs are also created in DESTINATION and their names are formatted as SOURCE1.bigrams, SOURCE2.bigrams, .... SOURCEP.bigrams.
count.pl is run on each file present in the SOURCE directory. All files in SOURCE are treated as the data files. If SOURCE contains sub-directories, these are simply skipped. Intermediate bigram outputs are written in DESTINATION.
If #arg > 2, all arguments specified after the first argument are considered as the SOURCE file names. count.pl is separately run on each of the SOURCE files specified by argv, argv, ... argv[n] (skipping argv which should be DESTINATION). Intermediate results are created in DESTINATION.
Files specified in the list of SOURCE should be relatively small sized plain files with #words < 1,000,000.
In summary, a large datafile can be provided to huge-count in the form of
a. A single plain file (along with --split P)
b. A directory containing several plain files
c. Multiple plain files directly specified as command line arguments
In all these cases, count.pl is separately run on SOURCE files or parts of SOURCE file and intermediate results are written in DESTINATION dir.
Intermediate count results created in STEP 2 are recombined in a pair-wise fashion such that for P separate count output files, C1, C2, C3 ... , CP,
C1 and C2 are first recombined and result is written to huge-count.output
Counts from each of the C3, C4, ... CP are then combined (added) to huge-count.output and each time while recombining, always the smaller of the two files is loaded.
After all files are recombined, the resultant huge-count.output is then sorted in the descending order of the bigram counts. If --remove is specified, bigrams with counts less than the specified value of --remove, in the final huge-count.output file are removed from the sample and their counts are deleted from the marginal totals. If --frequency is selected, bigrams with scores less than the specified value are simply skipped from output.
After huge-count finishes successfully, DESTINATION will contain -
huge-count.pl doesn't consider bigrams at file boundaries. In other words, the result of count.pl and huge-count.pl on the same data file will differ if --newLine is not used, in that, huge-count.pl runs count.pl on multiple files separately and thus looses the track of the bigrams on file boundaries. With --window not specified, there will be loss of one bigram at each file boundary while its W bigrams with --window W.
Functionality of huge-count is same as count only if --newLine is used and all files start and end on sentence boundaries. In other words, there should not be any sentence breaks at the start or end of any file given to huge-count.
Amruta Purandare, Ted Pedersen. University of Minnesota at Duluth.
Copyright (c) 2004,
Amruta Purandare, University of Minnesota, Duluth. firstname.lastname@example.org
Ted Pedersen, University of Minnesota, Duluth. email@example.com
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program; if not, write to
The Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.