Class: DSPy::Evaluate

Inherits:
Object
  • Object
show all
Extended by:
T::Sig
Defined in:
lib/dspy/evaluate.rb

Overview

Core evaluation framework for DSPy programs Supports single evaluations, batch evaluations, and optimization workflows

Defined Under Namespace

Classes: BatchEvaluationResult, EvaluationResult

Instance Attribute Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(program, metric: nil, num_threads: 1, max_errors: 5, provide_traceback: true) ⇒ Evaluate

Returns a new instance of Evaluate.



129
130
131
132
133
134
135
 
# File 'lib/dspy/evaluate.rb', line 129

def initialize(program, metric: nil, num_threads: 1, max_errors: 5, provide_traceback: true)
  @program = program
  @metric = metric
  @num_threads = num_threads || 1
  @max_errors = max_errors || 5
  @provide_traceback = provide_traceback
end

Instance Attribute Details

#max_errors ⇒ Object (readonly)

Returns the value of attribute max_errors.



115
116
117
 
# File 'lib/dspy/evaluate.rb', line 115

def max_errors
  @max_errors
end

#metric ⇒ Object (readonly)

Returns the value of attribute metric.



109
110
111
 
# File 'lib/dspy/evaluate.rb', line 109

def metric
  @metric
end

#num_threads ⇒ Object (readonly)

Returns the value of attribute num_threads.



112
113
114
 
# File 'lib/dspy/evaluate.rb', line 112

def num_threads
  @num_threads
end

#program ⇒ Object (readonly)

Returns the value of attribute program.



106
107
108
 
# File 'lib/dspy/evaluate.rb', line 106

def program
  @program
end

#provide_traceback ⇒ Object (readonly)

Returns the value of attribute provide_traceback.



118
119
120
 
# File 'lib/dspy/evaluate.rb', line 118

def provide_traceback
  @provide_traceback
end

Instance Method Details

#call(example, trace: nil) ⇒ Object 



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
 
# File 'lib/dspy/evaluate.rb', line 139

def call(example, trace: nil)
  DSPy::Context.with_span(
    operation: 'evaluation.example',
    'dspy.module' => 'Evaluator',
    'evaluation.program' => @program.class.name,
    'evaluation.has_metric' => !@metric.nil?
  ) do
    begin
      # Extract input from example - support both hash and object formats
      input_values = extract_input_values(example)
      
      # Run prediction
      prediction = @program.call(**input_values)
      
      # Calculate metrics if provided
      metrics = {}
      passed = true
      
      if @metric
        begin
          metric_result = @metric.call(example, prediction)
          if metric_result.is_a?(Hash)
            metrics = metric_result
            passed = metrics[:passed] || metrics['passed'] || true
          else
            passed = !!metric_result
            metrics[:passed] = passed
          end
        rescue => e
          passed = false
          metrics[:error] = e.message
          metrics[:passed] = false
        end
      end
      
      EvaluationResult.new(
        example: example,
        prediction: prediction,
        trace: trace,
        metrics: metrics,
        passed: passed
      )
    rescue => e
      # Return failed evaluation result
      error_metrics = {
        error: e.message,
        passed: false
      }
      
      if @provide_traceback
        error_metrics[:traceback] = e.backtrace&.first(10) || []
      end
      
      EvaluationResult.new(
        example: example,
        prediction: nil,
        trace: trace,
        metrics: error_metrics,
        passed: false
      )
    end
  end
end

#evaluate(devset, display_progress: true, display_table: false, return_outputs: true) ⇒ Object 



212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
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
 
# File 'lib/dspy/evaluate.rb', line 212

def evaluate(devset, display_progress: true, display_table: false, return_outputs: true)
  DSPy::Context.with_span(
    operation: 'evaluation.batch',
    'dspy.module' => 'Evaluator',
    'evaluation.program' => @program.class.name,
    'evaluation.num_examples' => devset.length,
    'evaluation.has_metric' => !@metric.nil?,
    'evaluation.num_threads' => @num_threads
  ) do
    results = []
    errors = 0
    
    if display_progress
      puts "Evaluating #{devset.length} examples..."
    end
    
    devset.each_with_index do |example, index|
      break if errors >= @max_errors
      
      begin
        result = call(example)
        results << result
        
        unless result.passed
          errors += 1
        end
        
        if display_progress && (index + 1) % 10 == 0
          puts "Processed #{index + 1}/#{devset.length} examples (#{results.count(&:passed)} passed)"
        end
        
      rescue => e
        errors += 1
        puts "Error processing example #{index}: #{e.message}" if display_progress
        
        # Create error result
        error_result = EvaluationResult.new(
          example: example,
          prediction: nil,
          trace: nil,
          metrics: { error: e.message, passed: false },
          passed: false
        )
        results << error_result
      end
    end
    
    # Aggregate metrics
    aggregated_metrics = aggregate_metrics(results)
    
    batch_result = BatchEvaluationResult.new(
      results: results,
      aggregated_metrics: aggregated_metrics
    )
    
    if display_table
      display_results_table(batch_result)
    end
    
    # Emit batch completion event
    DSPy.log('evaluation.batch_complete', **{
      'evaluation.program_class' => @program.class.name,
      'evaluation.total_examples' => batch_result.total_examples,
      'evaluation.passed_examples' => batch_result.passed_examples,
      'evaluation.pass_rate' => batch_result.pass_rate,
      'evaluation.aggregated_metrics' => aggregated_metrics
    })
    
    if display_progress
      puts "Evaluation complete: #{batch_result.passed_examples}/#{batch_result.total_examples} passed (#{(batch_result.pass_rate * 100).round(1)}%)"
    end
    
    batch_result
  end
end