{ "version": "https://jsonfeed.org/version/1.1", "title": "chem-bla-ics", "description": "Chemblaics (pronounced chem-bla-ics) is the science that uses open science and computers to solve problems in chemistry, biochemistry and related fields.", "home_page_url": "https://chem-bla-ics.linkedchemistry.info/", "feed_url": "https://chem-bla-ics.linkedchemistry.info/2007/05/30/weka-decision-trees-to-java-conversion.json", "icon": "https://chem-bla-ics.linkedchemistry.info/assets/images/chem-bla-ics_logo.png", "language": "en", "authors": [ { "name": "Egon Willighagen", "url": "https://orcid.org/0000-0001-7542-0286", "_orcid": "0000-0001-7542-0286" } ], "items": [ { "id": "https://doi.org/10.59350/bnwdb-pa057", "url": "https://chem-bla-ics.linkedchemistry.info/2007/05/30/weka-decision-trees-to-java-conversion.html", "title": "Weka Decision Trees to Java Conversion", "content_html": "

Some time ago I wrote a small Perl script to convert a decision tree created with Weka in the\nARFF format to Java source code, for use in the\nionization potential prediction\nin CDK. The advantage is that Weka is no longer used are runtime, and that there is no model that needs to be loaded and interpreted. Instead, it is simple Java code that does the work, much faster.

\n\n

This is the code:

\n\n
#!/usr/bin/perl\n#\n# Copyright 2007 (C) Egon Willighagen\n# License: GPL\n\nuse diagnostics;\nuse strict;\n\nmy $filename = $ARGV[0];\n\nprint \"double result = 0.0;\\n\";\nopen(INPUT, \"<$filename\");\nmy $level = 0;\nmy $prevLevel = -1;\nwhile (my $line = <INPUT>) {\n  $line =~ s/\\n//g;\n  $level = 0;\n  while ($line =~ /^\\|\\s*(.*)/) {\n    $level++;\n    $line = $1;\n  }\n  my $else = \"\";\n  if ($prevLevel == $level) {\n    $else = \"else \";\n  } elsif ($prevLevel < $level) {\n    # we increase one level at a time\n    for (my $i=0; $i<($level-1); $i++) { print \"  \"; };\n    print \"{\\n\";\n    $prevLevel = $level;\n  } else {\n    # this is a bit more tricky: we possibly need more than\n    # one end bracket\n    my $diff = $prevLevel - $level;\n    for (my $closes=0; $closes<$diff; $closes++) {\n      for (my $i=0; $i<($prevLevel-$closes-1); $i++) { print \"  \"; };\n      print \"}\\n\";\n    }\n    $prevLevel = $level;\n  }\n  if ($line =~ /:/) {\n    my ($if, $then) = split(\":\",$line);\n    for (my $i=0; $i<$level; $i++) { print \"  \"; };\n    # FIXME: java-fy $then\n    if ($then =~ /([\\d|_]*)\\s*\\(([^\\)]*)\\)/) {\n      my $result = $1;\n      my $stats = $2;\n      $result =~ s/_/\\./g;\n      print $else . \"if ($if) { result = $result; // $stats }\\n\";\n    } else {\n      print $else . \"if ($if) { result = $then; }\\n\";\n    }\n  } else {\n    for (my $i=0; $i<$level; $i++) { print \"  \"; };\n    print $else . \"if ($line)\\n\";\n  }\n}\n\n# OK, now add the rest of the closing brackets\nfor (my $closes=$prevLevel; $closes>0; $closes--) {\n  for (my $i=0; $i<($closes-1); $i++) { print \"  \"; };\n  print \"}\\n\";\n}\n
", "summary": "Some time ago I wrote a small Perl script to convert a decision tree created with Weka in the ARFF format to Java source code, for use in the ionization potential prediction in CDK. The advantage is that Weka is no longer used are runtime, and that there is no model that needs to be loaded and interpreted. Instead, it is simple Java code that does the work, much faster.", "date_published": "2007-05-30T00:10:00+00:00", "date_modified": "2025-02-15T00:00:00+00:00", "tags": ["java","cheminf","cdk"], "authors": [ { "name": "Egon Willighagen", "url": "https://orcid.org/0000-0001-7542-0286" } ] } ] }