Class: RubricLLM::Report

Inherits:
Object
  • Object
show all
Defined in:
lib/rubric_llm/report.rb

Instance Attribute Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(results:, duration: nil) ⇒ Report

Returns a new instance of Report.



9
10
11
12
 
# File 'lib/rubric_llm/report.rb', line 9

def initialize(results:, duration: nil)
  @results = results
  @duration = duration
end

Instance Attribute Details

#duration ⇒ Object (readonly)

Returns the value of attribute duration.



7
8
9
 
# File 'lib/rubric_llm/report.rb', line 7

def duration
  @duration
end

#results ⇒ Object (readonly)

Returns the value of attribute results.



7
8
9
 
# File 'lib/rubric_llm/report.rb', line 7

def results
  @results
end

Instance Method Details

#export_csv(path) ⇒ Object 



40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
 
# File 'lib/rubric_llm/report.rb', line 40

def export_csv(path)
  require "csv" # optional dependency — add `gem "csv"` to your Gemfile if missing
  metrics = all_metric_names
  CSV.open(path, "w") do |csv|
    csv << ["question", "answer", "overall", *metrics]
    results.each do |result|
      csv << [
        result.sample[:question],
        result.sample[:answer],
        result.overall,
        *metrics.map { |m| result.scores[m] }
      ]
    end
  end
end

#export_json(path) ⇒ Object 



56
57
58
 
# File 'lib/rubric_llm/report.rb', line 56

def export_json(path)
  File.write(path, JSON.pretty_generate(serializable_hash))
end

#failures(threshold: 0.8) ⇒ Object 



22
23
24
 
# File 'lib/rubric_llm/report.rb', line 22

def failures(threshold: 0.8)
  results.reject { |r| r.pass?(threshold:) }
end

#metric_stats ⇒ Object 



14
15
16
 
# File 'lib/rubric_llm/report.rb', line 14

def metric_stats
  @metric_stats ||= compute_stats
end

#scores_for(metric) ⇒ Object

Scores for a single metric across all results (for Comparison).



65
66
67
 
# File 'lib/rubric_llm/report.rb', line 65

def scores_for(metric)
  results.map { |r| r.scores[metric] }
end

#summary ⇒ Object 



26
27
28
29
30
31
32
33
34
35
36
37
38
 
# File 'lib/rubric_llm/report.rb', line 26

def summary
  lines = ["RubricLLM Evaluation Report"]
  lines << ("=" * 40)
  lines << "Samples: #{results.size}"
  lines << "Duration: #{"%.1f" % duration}s" if duration

  metric_stats.each do |metric, stats|
    lines << format("  %-20s  mean=%.3f  std=%.3f  min=%.3f  max=%.3f  n=%d",
                    metric, stats[:mean], stats[:std], stats[:min], stats[:max], stats[:count])
  end

  lines.join("\n")
end

#to_json ⇒ Object 



60
61
62
 
# File 'lib/rubric_llm/report.rb', line 60

def to_json(*)
  JSON.generate(serializable_hash, *)
end

#worst(n) ⇒ Object 



18
19
20
 
# File 'lib/rubric_llm/report.rb', line 18

def worst(n)
  results.sort_by { |r| r.overall || Float::INFINITY }.first(n)
end