train-test-split: proper implementation

1 files changed, 48 insertions, 37 deletions

diff --git a/train-test-split b/train-test-split
index aa55534..6aa4796 100755
--- a/train-test-split
+++ b/train-test-split
@@ -4,53 +4,64 @@ require 'zipf'
 require 'optimist'
 
 conf = Optimist::options do
-  opt :foreign, "foreign file", :type => :string, :required => true
-  opt :english, "english file", :type => :string, :required => true
+  opt :source, "source file", :type => :string, :required => true
+  opt :target, "target file", :type => :string, :required => true
   opt :size, "one size", :type => :int, :required => true
   opt :repeat, "number of repetitions", :type => :int, :default => 1
-  opt :prefix, "prefix for output files", :type => :string
+  opt :prefix, "prefix for output files", :type => :string, :default => "split"
   opt :sets, "number of sets", :type => :int, :default => 1
 end
-fn = conf[:foreign]
-fn_ext = fn.split('.').last
-f = ReadFile.readlines fn
-en = conf[:english]
-en_ext = en.split('.').last
-e = ReadFile.readlines en
+
+source_filename = conf[:source]
+source_extension = source_filename.split('.').last
+source_lines = ReadFile.readlines source_filename
+
+target_filename = conf[:target]
+target_extension = target_filename.split('.').last
+target_lines = ReadFile.readlines target_filename
+
 size = conf[:size]
-nlines_f = `wc -l #{fn}`.split()[0].to_i
-nlines_e = `wc -l #{en}`.split()[0].to_i
-if nlines_f != nlines_e
-  STDERR.write "Unbalanced files (#{nlines_f} vs. #{nlines_e}), exiting!\n"
+
+if source_lines.size != target_lines.size
+  STDERR.write "Unbalanced files (#{source_lines.size} vs. #{target_lines.size}), exiting!\n"
   exit 1
 end
 
-prefix = conf[:prefix]
-a = (0..nlines_e-1).to_a
-i = 0
-conf[:repeat].times {
-  if conf[:repeat] == 1
-    infix = ""
-  else
-    infix = ".#{i}"
-  end
-  b = a.sample(size)
-  ax = a.reject{|j| b.include? j}
+index = (0..source_lines.size-1).to_a
+conf[:repeat].times { |i|
   `mkdir split_#{i}`
-  new_f = WriteFile.new "split_#{i}/#{prefix}.train#{infix}.#{fn_ext}"
-  new_e = WriteFile.new "split_#{i}/#{prefix}.train#{infix}.#{en_ext}"
-  ax.each { |j|
-    new_f.write f[j]
-    new_e.write e[j]
+
+  sampled = index.sample(size * conf[:sets])
+
+  test_strings_source = {}
+  test_strings_target = {}
+
+  conf[:sets].times { |s|
+    slice_start_index = (s-1) * size
+
+    source_file = WriteFile.new "split_#{i}/#{conf[:prefix]}.devtest.#{s}.#{source_extension}"
+    target_file = WriteFile.new "split_#{i}/#{conf[:prefix]}.devtest.#{s}.#{target_extension}"
+
+    sampled.slice(slice_start_index, size).each { |j|
+      source_file.write source_lines[j]
+      target_file.write target_lines[j]
+      test_strings_source[source_lines[j].downcase] = true
+      test_strings_target[target_lines[j].downcase] = true
+    }
+    source_file.close; target_file.close
   }
-  new_f.close; new_e.close
-  new_f = WriteFile.new "split_#{i}/#{prefix}.devtest#{infix}.#{fn_ext}"
-  new_e = WriteFile.new "split_#{i}/#{prefix}.devtest#{infix}.#{en_ext}"
-  b.each { |j|
-    new_f.write f[j]
-    new_e.write e[j]
+
+  filtered_index = index.reject{ |j| sampled.include? j }
+  source_file = WriteFile.new "split_#{i}/#{conf[:prefix]}.train.#{source_extension}"
+  target_file = WriteFile.new "split_#{i}/#{conf[:prefix]}.train.#{target_extension}"
+  filtered_index.each { |j|
+    if not test_strings_source.include? source_lines[j].downcase \
+        and not test_strings_target.include? target_lines[j]
+      source_file.write source_lines[j]
+      target_file.write target_lines[j]
+    end
   }
-  new_f.close; new_e.close
+  source_file.close; target_file.close
+
   i += 1
 }
-