Class: Evoc::Experiment

Inherits:
Object
  • Object
show all
Includes:
Logging
Defined in:
lib/evoc/experiment.rb

Constant Summary collapse

@@logger_level = 
:error

Instance Attribute Summary collapse

Instance Method Summary collapse

Methods included from Logging

configure_logger_for, #logger, logger_for, set_level

Constructor Details

#initialize(opts = Hash.new) ⇒ Experiment 



8
9
10
11
12
13
14
15
16
17
 
# File 'lib/evoc/experiment.rb', line 8

def initialize(opts = Hash.new)
  self.opts = opts
  # Set logger level
  Logging.set_level(self.opts[:logger_level])
  logger.debug "Initialized experiment with options: #{opts}"
  # setup history
  if !opts[:transactions].nil?
    Evoc::HistoryStore.initialize(path: self.opts[:transactions],case_id: self.opts[:case_id], granularity: self.opts[:granularity])
  end
end

Instance Attribute Details

#opts ⇒ Object

Returns the value of attribute opts.



4
5
6
 
# File 'lib/evoc/experiment.rb', line 4

def opts
  @opts
end

Instance Method Details

#execute_scenarios ⇒ Object

Execute a set of scenarios

@return: json lines stream to stdout (jsonlines.org/)



229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
 
# File 'lib/evoc/experiment.rb', line 229

def execute_scenarios

  ######
  # Setup factors
  #####

  # Factor: model size aka model/learning/history size
  factor_model_size = self.opts[:model_size].nil? ? nil : self.opts[:model_size].map {|s| [ 'model_size',s ]}
  # Factor: Max size aka filtering constraint on history
  factor_max_size = self.opts[:max_size].nil? ? nil : self.opts[:max_size].map {|s| [ 'max_size',s ]}
  # Factor: Model age aka number of commits between query and last tx in history
  factor_model_age = self.opts[:model_age].nil? ? nil : self.opts[:model_age].map {|s| [ 'model_age',s ]}
  # Factor: Measures
  factor_measures = self.opts[:measures].map {|c| ['measures',c]}
  # Factor: Aggregator
  # if aggregation is requested, we also assume that no aggregation wants to be tested
  # the non-aggregation rows are specified with 'aggregator' == nil
  factor_aggregators = self.opts[:aggregators].nil? ? nil : (self.opts[:aggregators]+[nil]).map {|a| ['aggregator',a]}
  # deprecated factor, allways set to 1 for backwards compatibility
  # factor_permutation = self.opts[:permutation].nil? ? nil : (1..self.opts[:permutation]).to_a.map {|p| [ 'permutation',p ]}
  factor_permutation =  [[ 'permutation',1 ]]

  ####
  # Iterate over the queries provided and execute each query in each scenario
  #
  # queries CSV header:
  # tx_id,tx_index,tx,query,query_size,expected_outcome,expected_outcome_size

  # Count the number of lines so we can properly format the output json
  num_lines = File.read(self.opts[:queries]).each_line.count-1
  current_line = 1

  # compact removes nil values (not used factors)
  factors = [factor_model_size,factor_max_size,factor_model_age,factor_measures,factor_permutation,factor_aggregators].compact
  num_of_scenarios = factors.inject(1) {|product,f| product * f.size}
  invalid_configuration = 0 
  last_error = 'no errors'
  CSV.foreach(self.opts[:queries], headers: true) do |query|
    # abort if the failsafe file is present 
    if !self.opts[:fail_safe].nil?
      if File.exists?(self.opts[:fail_safe])
        $stderr.puts "\nFail safe detected, exiting.."
        break
      end
    end
    # get query
    query_hash = query.to_h
    # convert query string to array of items
    query_hash['query'] = query_hash['query'].split(',').map(&:to_i)
    # verify query before executing
    tx = nil
    if tx = Evoc::HistoryStore.base_history.get_tx(id: query_hash['tx_id'],id_type: :id)
      if !(query_hash['query'] - tx.items).empty?
        raise Evoc::Exceptions::ConfigurationError.new "The query generated from #{query_hash['tx_id']} was not a subset of the same tx in the loaded history. The query was: '#{query_hash['query']}', the tx was '#{tx.items}'"
      end
    else
      raise Evoc::Exceptions::ConfigurationError.new "Could not find the tx: '#{query_hash['tx_id']}' from #{self.opts[:queries]} in the history #{self.opts[:transactions]}"
    end

    current_scenario = 1
    # - the splat operator '*' turns the array into parameters for #product
    # - the block form of #product makes it lazy (i.e., the whole cartesian product isn't generated at once)
    factors.first.product(*factors[1..-1]).each do |scenario|
      params = query_hash.merge(scenario.to_h)
      params[:case_id] = self.opts[:case_id]
      params[:granularity] = self.opts[:granularity]
      # initialize scenario
      s =  Evoc::Scenario.new(params)
      scenario_stats = {}
      if self.opts[:stats]
        scenario_stats = s.stats
      end
      # Factor: Algorithm
      self.opts[:algorithms].each do |algorithm|
        s.algorithm = algorithm
        # Print progress to stderr
        STDERR.print "(#{self.opts[:case_id]}) Executing #{algorithm} on scenario #{current_scenario} of #{num_of_scenarios} on query #{current_line} of #{num_lines}"
        if invalid_configuration > 0
          STDERR.print " (scenarios skipped: #{invalid_configuration},last reason: #{last_error[0..20]}...)                             \r"
        else
          STDERR.print "                                \r"
        end

        begin
          Evoc::RecommendationCache.get_recommendation(algorithm: algorithm,
                                                       query: s.query,
                                                       model_start: s.model_start,
                                                       model_end: s.model_end,
                                                       max_size: s.max_size,
                                                       aggregator: s.aggregator,
                                                       measures: s.measures)
          Evoc::RecommendationCache.evaluate_last(evaluators: self.opts[:evaluators],
                                                  topk: self.opts[:topk],
                                                  unique_consequents: self.opts[:unique_consequents],
                                                  expected_outcome: s.expected_outcome,
                                                  measure_combination: s.measures)

          # build json line by merging hashes
          $stdout.puts s.to_h
                         .merge(scenario_stats)
                         .merge({topk: self.opts[:topk], date: tx.date})
                         .merge(Evoc::RecommendationCache.to_h(measures: s.measures))
                         .to_json
        rescue ArgumentError => e
          invalid_configuration += 1
          last_error = e.message
        end
      end
      current_scenario += 1
    end
    current_line += 1
  end
  STDERR.puts "\n(#{self.opts[:case_id]}) DONE"
end

#generate_queries ⇒ Object

Generates a CSV of queries according to the given options

CSV HEADER:

tx_id, query



113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
 
# File 'lib/evoc/experiment.rb', line 113

def generate_queries
  # initialze a random number generator with fixed seed
  rand = Random.new(self.opts[:seed])
  ##
  # write dict
  ##
  if path = self.opts[:write_dict]
    tmp = Tempfile.new('dict')
    begin
      tmp.puts("id,name")
      Evoc::HistoryStore.base_history.int_2_name.each do |id,name|
        tmp.puts("#{id},#{name}")
      end
      tmp.close
      FileUtils.mv(tmp.path,path)
    ensure
      tmp.close
      tmp.unlink
    end
  end
  ##
  # WRITE CSV HEADER
  CSV {|row| row << %W(tx_id query)}

  ###
  # Iterate over sampled tx ids
  CSV.foreach(self.opts[:transaction_ids_path], headers: true) do |row|
    tx_id = row['tx_id']
    ##
    # GET THE TRANSACTION
    if tx = Evoc::HistoryStore.base_history.get_tx(id: tx_id, id_type: :id)
      items = tx.items
      tx_size = items.size
      ##
      # SAMPLE QUERIES
      #
      # We have 3 different strategies, which may produce the same sizes,
      # but the same size does not need to be executed several times, 
      # so duplicates are removed
      specified_sizes = []
      if !self.opts[:select].nil? then specified_sizes << self.opts[:select].map(&:to_i) end
      if !self.opts[:reverse_select].nil? then specified_sizes << self.opts[:reverse_select].map {|i| tx_size-i.to_i} end
      if !self.opts[:percentage].nil? then specified_sizes << self.opts[:percentage].map {|p| (p.to_f/100*tx_size).ceil} end
      # filter out sizes <= 1
      specified_sizes.flatten!.select! {|s| s > 0}
      specified_sizes.uniq!

      random_sizes = []
      if self.opts[:random_select] then random_sizes << Random.new.rand(self.opts[:minimum_query_size]..(tx_size-1)) end

      sampled_queries = []
      # only specified sizes
      if random_sizes.empty? & !specified_sizes.empty?
        sampled_queries = specified_sizes.map {|s| items.sample(s, random: rand)}
      # only random sizes
      elsif !random_sizes.empty? & specified_sizes.empty?
        sampled_queries = random_sizes.map {|s| items.sample(s, random: rand)}
      # random + specified = randomly sample in range defined by specified
      # ex:
      # specified = [1,3,10,20]
      # tx size = 4
      #
      # 1. remove X in specified that are larger than or equal to 4
      # 2. randomly select X in specified = Y
      # 3. randomly select Y in tx
      elsif !random_sizes.empty? & !specified_sizes.empty?
        specified_sizes.select! {|s| (s < tx_size) & (s >= self.opts[:minimum_query_size])} #1.
        if randomly_sampled_size = specified_sizes.sample(random: rand) #2.
          sampled_queries = [items.sample(randomly_sampled_size, random: rand)] #3.
        end
      end

      if sampled_queries.empty?
        logger.warn "Unable to generate query from tx: #{items}, with params #{self.opts}"
      end

      ##
      # WRITE CSV
      sampled_queries.each do |query| 
        if query.size == tx_size
          logger.debug "The size of the sampled query was equal to the size of the transaction, skipping.. Tx ID: #{tx_id}. Query size: #{query.size}"
          next
        end
        if query.size < self.opts[:minimum_query_size]
          next
        end
        CSV {|row| row << [tx_id,query.join(',')]}
      end
    else
      raise ArgumentError, "The tx with id '#{tx_id}' was not found in the history: #{self.opts[:transactions]}, wrong file?"
    end
  end

  #TODO  possibly move this to execution
  #############################################
  # Filter the generated queries if requested #
  #############################################
  #
  #
  #      logger.debug "Number of queries before filtering: #{query_store.size}"
  #
  #      if self.opts[:filter_expected_outcome]
  #        STDERR.puts "Filtering expected outcome.."
  #        init_size = query_store.size
  #        query_store.filter_expected_outcome!
  #        STDERR.puts "Had to remove #{init_size - query_store.size} queries as a result of the expected outcome now being empty"
  #      end
  #      if query_store.empty?
  #        STDERR.puts "WARNING: Returning 0 queries (maybe increase sample size?)"
  #      end
end

#sample_transactions ⇒ Object 



19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
 
# File 'lib/evoc/experiment.rb', line 19

def sample_transactions
  # initialze a random number generator with fixed seed
  rand = Random.new(self.opts[:seed])
  # by default we can sample from the whole history
  sampling_history = Evoc::HistoryStore.base_history
  STDERR.puts "Sampling transactions from a pool of #{sampling_history.size}.."
  sample = []

  #################################################################################
  # performing filtering steps on min/max commits size and minimum previous history
  #################################################################################

  if !self.opts[:recent].nil?
    size = sampling_history.size
    sampling_history = sampling_history[[0,size-self.opts[:recent]].max..-1]
    STDERR.puts "    Filtering to the #{self.opts[:recent]} most recent transactions (new pool size: #{sampling_history.size})"
  end
  # filter out transactions larger than X
  if !self.opts[:minimum_commit_size].nil?
    sampling_history = sampling_history.select {|tx| tx.size >= self.opts[:minimum_commit_size]}
    STDERR.puts "    Filtering to txes larger than or equal to #{self.opts[:minimum_commit_size]} (new pool size: #{sampling_history.size})"
  end
  if !self.opts[:maximum_commit_size].nil?
    sampling_history = sampling_history.select {|tx| tx.size <= self.opts[:maximum_commit_size]}
    STDERR.puts "    Filtering to txes smaller than or equal to #{self.opts[:maximum_commit_size]} (new pool size: #{sampling_history.size})"
  end
  # only sample transactions that have at least 'minimum_history' previous history
  if !self.opts[:minimum_history].nil?
    sampling_history = sampling_history.select {|tx| tx.index >= self.opts[:minimum_history]}
    STDERR.puts "    Filtering to txes with at least #{self.opts[:minimum_history]} previous txes (new pool size: #{sampling_history.size})"
  end
  if !self.opts[:recent_viable].nil?
    size = sampling_history.size
    sampling_history = sampling_history[[0,size-self.opts[:recent_viable]].max..-1]
    STDERR.puts "    Filtering to the #{self.opts[:recent_viable]} most recent viable transactions (new pool size: #{sampling_history.size})"
  end

  filtering_switches = [:recent,:recent_viable,:minimum_commit_size,:maximum_commit_size,:minimum_history]
  if filtering_switches.any? {|s| !self.opts[s].nil?}
    if sampling_history.size == 0
      STDERR.puts "WARNING: All transactions were filtered out, unable to sample"
      return []
    end
  end

  if self.opts[:sample_size] > sampling_history.size 
    STDERR.puts "WARNING: The sample size is larger than the available transactions"
  end

  ######################
  # performing sampling
  ######################

  # group the txes by size
  groups = sampling_history.group_by {|tx| tx.size}
  # sort the sample_groups option to reduce the need for maintaining control over which txes that have been sampled
  # i.e., random sampling is done first, then the sampled txes are removed from the sampling
  tx_sizes_to_sample_from = self.opts[:sample_groups].sort_by(&:to_s)
  tx_sizes_to_sample_from.each do |group_size|
    if group_size == '*'
      sampled_ids = sampling_history.map(&:id).sample(self.opts[:sample_size], random: rand)
      sample << sampled_ids
      STDERR.puts "Sampled #{sampled_ids.size} txes"
      # remove sampled txes from sampling_history
      filtered_hist = sampling_history.reject {|tx| sampled_ids.include? tx.id}
      sampling_history.clear
      filtered_hist.each {|tx| sampling_history << tx}
    elsif group_size.to_i
      # check if there were any txes of this size
      if group = groups[group_size.to_i]
        if group.size < self.opts[:sample_size]
          logger.warn "Only #{group.size} transactions found of size #{group_size}, asked for #{self.opts[:sample_size]}"
        end
        sampled_ids = group.sample(self.opts[:sample_size], random: rand).map(&:id)
        sample << sampled_ids
        STDERR.puts "Sampled #{sampled_ids.size} txes of size #{group_size}"
      else
        logger.warn "No transactions found of size #{group_size}, asked for #{self.opts[:sample_size]} (minimum history: #{self.opts[:minimum_history]})"
      end
    else
      raise ArgumentError.new, "Tx size for sampling must either be specified by an Integer or '*' (was #{group_size}:#{group_size.class})"
    end
  end
  sample.flatten.uniq
end