frprep 0.0.1.prealpha

data/lib/shalmaneser/version.rb

@@ -0,0 +1,3 @@
+module Shalmaneser
+  VERSION = '0.0.1.prealpha'
+end

data/test/frprep/test_opt_parser.rb

@@ -0,0 +1,94 @@
+# -*- coding: utf-8 -*-
+
+require 'test/unit'
+require 'stringio' # for helper methods
+require 'frprep/opt_parser'
+
+include FrPrep
+
+class TestOptParser < Test::Unit::TestCase
+
+  def setup
+    @exp_file = 'test/frprep/data/prp_test.salsa'
+    @valid_opts = ['--expfile', @exp_file,
+                   '--help'
+                  ]
+  end
+
+  def test_public_methods
+    assert_respond_to(OptParser, :parse)
+  end
+
+  # It should return a FrPrepConfigData object.
+  def test_parse_method
+    input = ['-e', @exp_file]
+    return_value = OptParser.parse(input)
+    assert(return_value.instance_of?(FrPrepConfigData))
+  end
+
+  # It should reject the empty input and exit.
+  def test_empty_input
+    out, err = intercept_output do
+      assert_raises(SystemExit) { OptParser.parse([]) }
+    end
+    assert_match(/You have to provide some options./, err)
+  end
+
+  # It should accept correct options.
+  # Invalid options is the matter of OptionParser itself,
+  # do not test it here.
+  # We test only, that OP exits and does not raise an exception.
+  def test_accept_correct_options
+    # this options we should treat separately
+    @valid_opts.delete('--help')
+    assert_nothing_raised { OptParser.parse(@valid_opts) }
+
+    stdout, stderr = intercept_output do 
+      assert_raises(SystemExit) { OptParser.parse(['--invalid-option']) }
+    end
+
+    assert_match(/You have provided an invalid option:/, stderr)
+  end
+
+  # It should successfully exit with some options.
+  def test_successful_exit
+    quietly do
+      success_args = ['-h', '--help']
+      success_args.each do |arg|
+        assert_raises(SystemExit) { OptParser.parse(arg.split) }
+      end
+    end
+  end
+
+end
+################################################################################
+# It is a helper method, many testable units provide some verbose output
+# to stderr and/or stdout. It is usefull to suppress any kind of verbosity.
+def quietly(&b)
+  begin
+    orig_stderr = $stderr.clone
+    orig_stdout = $stdout.clone
+    $stderr.reopen(File.new('/dev/null', 'w'))
+    $stdout.reopen(File.new('/dev/null', 'w'))
+    b.call
+  ensure
+    $stderr.reopen(orig_stderr)
+    $stdout.reopen(orig_stdout)
+  end
+end
+
+# It is a helper method for handling stdout and stderr as strings.
+def intercept_output
+  orig_stdout = $stdout
+  orig_stderr = $stderr
+  $stdout = StringIO.new
+  $stderr = StringIO.new
+  
+  yield
+  
+  return $stdout.string, $stderr.string
+ensure
+  $stdout = orig_stdout
+  $stderr = orig_stderr
+end
+

data/test/functional/functional_test_helper.rb

@@ -0,0 +1,40 @@
+require 'erb'
+
+module FunctionalTestHelper
+  PREF = 'test/functional/sample_experiment_files'
+
+  PRP_TEST_FILE   = 'test/functional/sample_experiment_files/prp_test.salsa'
+  PRP_TEST_FILE_FRED_STD = "#{PREF}/prp_test.salsa.fred.standalone"
+  PRP_TEST_FILE_ROSY_STD = "#{PREF}/prp_test.salsa.rosy.standalone"
+  PRP_TRAIN_FILE  = 'test/functional/sample_experiment_files/prp_train.salsa'
+  PRP_TRAIN_FILE_FRED_STD  = "#{PREF}/prp_train.salsa.fred.standalone"
+  PRP_TRAIN_FILE_ROSY_STD  = "#{PREF}/prp_train.salsa.rosy.standalone"
+
+  FRED_TEST_FILE  = 'test/functional/sample_experiment_files/fred_test.salsa'
+  FRED_TRAIN_FILE = 'test/functional/sample_experiment_files/fred_train.salsa'
+  ROSY_TEST_FILE  = 'test/functional/sample_experiment_files/rosy_test.salsa'
+  ROSY_TRAIN_FILE = 'test/functional/sample_experiment_files/rosy_train.salsa'
+
+  # Run an external process for functional testing and check the return code.
+  # <system> returns <true> if the external code exposes no errors.
+  # <@msg> is defined for every test object.
+  def execute(cmd)
+    status = system(cmd)
+    assert(status, @msg)
+  end
+  
+  # Create a temporary exp file only for this test.
+  # Shalmaneser needs absolute paths, we provide them in exp files
+  # using templating.
+  def create_exp_file(file)
+    template = File.read("#{file}.erb")
+    text = ERB.new(template).result
+    File.open(file, 'w') do |f|
+      f.write(text)
+    end
+  end
+
+  def remove_exp_file(file)
+    File.delete(file)
+  end
+end

data/test/functional/sample_experiment_files/fred_test.salsa.erb

@@ -0,0 +1,122 @@
+experiment_ID = fred_test
+
+apply_to_all_known_targets = true
+
+enduser_mode = false
+
+verbose = true
+
+
+############################
+# Paths
+# - fred_directory: directory where Fred puts its internal data
+# - directory_output: 
+#   redirect system output of disambiguated text (in SalsaTigerXML)
+#   to another directory. 
+#   If you do not set anything here, output is to
+#   <fred_directory>/<experiment_ID>/output/stxml
+# - classifier_dir:
+#   Write trained classifiers to this directory.
+#   If you do not set this parameter, classifiers are written to
+#   <fred_directory>/<experiment_ID>/classifiers
+
+fred_directory = <%= File.expand_path('test/functional/output') %>
+classifier_dir = <%= File.expand_path('test/functional/input/fred/cls') %>
+# - preproc_descr_file_train / ...test
+#   where the experiment file for frprep is located
+#   (preprocessing for Fred and Rosy)
+#   for the preprocessing of the data used in this experiment
+#
+#   give one preprocessing file name for the training data
+#   and one for the test data
+#   (If you only ever use test data in this experiment, you only
+#   need to give preproc_descr_file_test, and vice versa for training data.)
+
+preproc_descr_file_test = <%= File.expand_path('test/functional/sample_experiment_files/prp_test.salsa.fred.standalone') %>
+
+#####################
+# noncontiguous input?
+# if so, set 'noncontiguous_input' to 'true' (default is 'false')
+# Also give the larger corpus from which the input sentences are:
+# - directory
+# - format: same possibilities as for frprep format
+# - encoding: same possibilities as for frprep encoding
+
+noncontiguous_input = false
+#larger_corpus_dir = 
+larger_corpus_format = SalsaTigerXML
+#larger_corpus_encoding = iso
+
+
+#################
+# Features
+
+# bag-of-words context, with given context size,
+# for example:
+ feature = context 50
+ feature = context 2
+#
+# (you can give more than one context feature line!)
+#
+# other possible features:
+# feature = syntax
+# feature = synsem
+#
+# syntax: grammatical functions
+# synsem: grammatical functions plus headwords
+
+#feature = context % %contextsize%
+feature = syntax
+
+# How to handle training data that is labeled
+# with multiple sense labels?
+# - binarize (default): This works only with binary classifiers. 
+#   When featurizing for the binary classifiers, consider an item 
+#   positive if its set of assigned labels includes the 
+#   label for this binary classifier.
+# - repeat: Repeat the instance, once for each
+#   sense label that has been assigned. (Basically, treat it
+#   as N instances with equal features but different labels.)
+# - join: join all the assigned senses into one combined sense
+#   and treat that as a separate sense to train on.
+# - keep: keep as multiple sense labels. (Note that this
+#   makes sense only for classifiers that can deal with
+#   multiple labels.)
+
+#handle_multilabel = binarize
+handle_multilabel = repeat
+
+# What to do with numerical features?
+# - keep: just leave as is
+# - repeat: for a feature with max. numerical value N,
+#   use N binary features
+# - bin: use a fixed number of bins, e.g. 5, then
+#   if feature value > 20: set all bins to 1,
+#   if feature value > 10: set the first four bins to 1, 
+#   etc.
+#   default: bin.
+#numerical_features = bin
+numerical_features = keep
+
+# Binary classifiers, or n-ary classifiers?
+# if binary classifiers, set 'binary_classifiers = true'
+# default is 'false'.
+binary_classifiers = false
+
+#################
+# Fred internal settings
+
+# what kind of classifier to use?
+#
+# format:
+# <classifier type> <path> <optionally another path>
+#
+# for maxent, give first the path where maxent resides, 
+# then <where_shalmaneser_resides>/program/tools/maxent
+classifier = maxent /opt/OpenNLP-maxent/2.4.0 <%= File.expand_path('tools/maxent/') %>
+
+
+# for binary classifiers, you can set the pseudolabel
+# on the 'negative' sense.
+# Default is 'NONE'      
+negsense = NONE

data/test/functional/sample_experiment_files/fred_train.salsa.erb

@@ -0,0 +1,135 @@
+# ID identifying this experiment and all its data
+# please do not use spaces inside the experiment ID
+experiment_ID = fred_train
+
+# targets:
+# if apply_to_all_known_targets is set to true,
+# disambiguate all words for which we have training data
+# when performing task "test" (i.e. applying trained classifiers)
+apply_to_all_known_targets = true
+
+# Enduser mode?
+# The idea is that the enduser will only _apply_
+# pre-trained classifiers. So in enduser mode many 
+# options are disallowed.
+enduser_mode = false
+
+
+# print warnings and 
+# give detailed progress reports
+verbose = true
+
+
+############################
+# Paths
+# - fred_directory: directory where Fred puts its internal data
+# - directory_output: 
+#   redirect system output of disambiguated text (in SalsaTigerXML)
+#   to another directory. 
+#   If you do not set anything here, output is to
+#   <fred_directory>/<experiment_ID>/output/stxml
+# - classifier_dir:
+#   Write trained classifiers to this directory.
+#   If you do not set this parameter, classifiers are written to
+#   <fred_directory>/<experiment_ID>/classifiers
+
+fred_directory = <%= File.expand_path('test/functional/output') %>
+
+# - preproc_descr_file_train / ...test
+#   where the experiment file for frprep is located
+#   (preprocessing for Fred and Rosy)
+#   for the preprocessing of the data used in this experiment
+#
+#   give one preprocessing file name for the training data
+#   and one for the test data
+#   (If you only ever use test data in this experiment, you only
+#   need to give preproc_descr_file_test, and vice versa for training data.)
+preproc_descr_file_train = <%= File.expand_path('test/functional/sample_experiment_files/prp_train.salsa.fred.standalone') %>
+
+
+
+#####################
+# noncontiguous input?
+# if so, set 'noncontiguous_input' to 'true' (default is 'false')
+# Also give the larger corpus from which the input sentences are:
+# - directory
+# - format: same possibilities as for frprep format
+# - encoding: same possibilities as for frprep encoding
+
+noncontiguous_input = false
+#larger_corpus_dir = 
+larger_corpus_format = SalsaTigerXML
+#larger_corpus_encoding = iso
+
+
+#################
+# Features
+
+# bag-of-words context, with given context size,
+# for example:
+ feature = context 50
+ feature = context 2
+#
+# (you can give more than one context feature line!)
+#
+# other possible features:
+# feature = syntax
+# feature = synsem
+#
+# syntax: grammatical functions
+# synsem: grammatical functions plus headwords
+
+#feature = context % %contextsize%
+feature = syntax
+
+# How to handle training data that is labeled
+# with multiple sense labels?
+# - binarize (default): This works only with binary classifiers. 
+#   When featurizing for the binary classifiers, consider an item 
+#   positive if its set of assigned labels includes the 
+#   label for this binary classifier.
+# - repeat: Repeat the instance, once for each
+#   sense label that has been assigned. (Basically, treat it
+#   as N instances with equal features but different labels.)
+# - join: join all the assigned senses into one combined sense
+#   and treat that as a separate sense to train on.
+# - keep: keep as multiple sense labels. (Note that this
+#   makes sense only for classifiers that can deal with
+#   multiple labels.)
+
+#handle_multilabel = binarize
+handle_multilabel = repeat
+
+# What to do with numerical features?
+# - keep: just leave as is
+# - repeat: for a feature with max. numerical value N,
+#   use N binary features
+# - bin: use a fixed number of bins, e.g. 5, then
+#   if feature value > 20: set all bins to 1,
+#   if feature value > 10: set the first four bins to 1, 
+#   etc.
+#   default: bin.
+#numerical_features = bin
+numerical_features = keep
+# Binary classifiers, or n-ary classifiers?
+# if binary classifiers, set 'binary_classifiers = true'
+# default is 'false'.
+binary_classifiers = false
+
+#################
+# Fred internal settings
+
+# what kind of classifier to use?
+#
+# format:
+# <classifier type> <path> <optionally another path>
+#
+# for maxent, give first the path where maxent resides, 
+# then <where_shalmaneser_resides>/program/tools/maxent
+classifier = maxent /opt/OpenNLP-maxent/2.4.0 <%= File.expand_path('tools/maxent/') %>
+
+
+# for binary classifiers, you can set the pseudolabel
+# on the 'negative' sense.
+# Default is 'NONE'      
+negsense = NONE

data/test/functional/sample_experiment_files/prp_test.salsa.erb

@@ -0,0 +1,138 @@
+#################################################
+# This is a sample experiment file
+# with explanations of all features 
+# that can be set for the frprep preprocessing system for Fred and Rosy.
+#
+# To start your own experiment,
+# replace all occurrences of 
+# %...% by values of your choice.
+#
+# Boolean features may be omitted and are false by default.
+#
+# Experiment file lines that start with '#'
+# are comments and are ignored. Empty lines are ignored as well. 
+
+########################
+# Experiment description
+#
+
+# ID identifying this experiment and all its data
+# please do not use spaces inside the experiment ID
+prep_experiment_ID = prp_test
+
+# YOUR INPUT DATA:
+# frprep accepts an input directory rather than an input file.
+# It will process all files in the directory directory_input
+# and write the results to directory_preprocessed.
+#
+# For input formats see the discussion of "format" below.
+directory_input = <%= File.expand_path('test/functional/input/frprep/test.salsa') %>
+directory_preprocessed = <%= File.expand_path('test/functional/output/frprep/test.salsa') %>
+
+##
+# Experimental data is described by the following parameters:
+# 
+# - language: en / de
+#    en for English or de for German
+#
+# - format:  SalsaTigerXML / FNXml / SalsaTab / BNC / Plain
+#
+#    Format of the input data, training/test set
+#    SalsaTigerXML:  Parsed data, English or German
+#    FNXml:          FrameNet Lexical Unit files in FrameNet XML format
+#    FNCorpusXML:    FrameNet files in the FrameNet corpus XML format
+#    SalsaTab:       tabular format (internal)
+#    BNC             BNC XML format, alternating words and POS tags
+#    Plain           Plain text, ONE SENTENCE PER LINE. 
+#
+#    Preprocessing transforms all data to SalsaTigerXML.
+#
+# - origin:  SalsaTiger / FrameNet / <not specified>
+#    This is the origin of the training/test data.
+#    SalsaTiger: data from the Tiger corpus, possibly semantically
+#                annotated by Salsa
+#    FrameNet: data from the FrameNet project
+#
+#    Don't set 'origin' if none of these origins apply
+#
+# - encoding: utf8 / iso / hex / <not specified>
+#                 Default: iso
+
+language = de
+#origin = 
+format = Plain
+encoding = iso
+
+#############################
+# Which preprocessing steps to take?
+#
+# Data can be parsed, lemmatized and POS-tagged,
+# but this happens only if it is specified in the
+# experiment file.
+#
+# Set these booleans to true to trigger the respective
+# type of preprocessing. The default value is false.
+
+do_lemmatize = true
+do_postag = false
+do_parse = true
+
+#############################
+# directory where frprep puts its internal data
+#
+
+frprep_directory = <%= File.expand_path('test/functional/output/') %>
+
+#############################
+# Syntax/semantics interface repair:
+# FrameNet annotated data has some annotation choices
+# that may make it harder to learn the mapping from
+# syntactic structure to semantic roles.
+#
+# If you are using FrameNet data for training a 
+# semantic role labeler, set the following two settings
+# to true (default is false) to 'repair' semantic role labels
+# to closer match the syntactic structure
+
+fe_syn_repair = true
+fe_rel_repair = false
+
+
+#################
+# Location of tools and resources used by Fred
+
+# currently known to the system:
+# (Saarbruecken paths given)
+#
+# - POS tagging:
+#   - pos_tagger = treetagger
+#     pos_tagger_path = /proj/llx/Software/treetagger/cmd/tree-tagger-english-notokenisation
+#
+# - Lemmatization:
+#   - lemmatizer = treetagger
+#     lemmatizer_path = /proj/llx/Software/treetagger/cmd/tree-tagger-english-notokenisation
+#     lemmatizer_path = /proj/llx/Software/treetagger/cmd/tree-tagger-german-notokenisation
+#
+# - Parser:
+#   - parser = collins  (English)
+#     parser_path = /proj/llx/Software/Parsers/COLLINS-PARSER/
+#   - parser = sleepy   (German)
+#     parser_path = /proj/corpora/sleepy3/
+#   - parser = minipar (English)
+#     parser_path = /proj/llx/Software/Parsers/minipar-linux/
+#
+pos_tagger = treetagger
+pos_tagger_path = <%= File.expand_path('tools/treetagger/shal-ger') %>
+
+lemmatizer = treetagger
+lemmatizer_path = <%= File.expand_path('tools/treetagger/shal-ger') %>
+
+parser = berkeley
+parser_path = <%= File.expand_path('tools/berkeleyParser') %>
+
+# parser: 
+# maximum no. of sentences in a parse file,
+# maximum sentence length to be parsed
+
+parser_max_sent_num = 2000
+parser_max_sent_len = 80