Class: CompletionKit::MetricCalibrationStats

Inherits:
Object
  • Object
show all
Defined in:
app/services/completion_kit/metric_calibration_stats.rb

Defined Under Namespace

Classes: Result

Constant Summary collapse

PROVISIONAL_MIN = 
10
FIRM_MIN = 
30

Class Method Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(metric:, judge_version: nil) ⇒ MetricCalibrationStats

Returns a new instance of MetricCalibrationStats.



38
39
40
41
 
# File 'app/services/completion_kit/metric_calibration_stats.rb', line 38

def initialize(metric:, judge_version: nil)
  @metric = metric
  @judge_version = judge_version
end

Class Method Details

.for(metric, judge_version: nil) ⇒ Object 



34
35
36
 
# File 'app/services/completion_kit/metric_calibration_stats.rb', line 34

def self.for(metric, judge_version: nil)
  new(metric: metric, judge_version: judge_version).call
end

Instance Method Details

#call ⇒ Object 



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
 
# File 'app/services/completion_kit/metric_calibration_stats.rb', line 43

def call
  scope = Calibration.where(metric_id: @metric.id)
  scope = scope.where(judge_version_id: @judge_version.id) if @judge_version

  verdicts = scope.pluck(:verdict, :corrected_score, :response_id)
  n = verdicts.length
  agrees = verdicts.count { |v, _, _| v == "agree" }
  disagrees = verdicts.count { |v, _, _| v == "disagree" }
  borderlines = verdicts.count { |v, _, _| v == "borderline" }

  ci = CalibrationMath.wilson_interval(successes: agrees, n: n)

  pairs = score_pairs(verdicts)
  mae_value = CalibrationMath.mae(pairs)
  pearson_value = CalibrationMath.pearson(pairs)
  kappa_value = CalibrationMath.quadratic_weighted_kappa(pairs, categories: 1..5)

  Result.new(
    sample_size: n,
    agree_count: agrees,
    disagree_count: disagrees,
    borderline_count: borderlines,
    agreement_point: ci[:point],
    agreement_low: ci[:low],
    agreement_high: ci[:high],
    borderline_rate: n.zero? ? nil : borderlines.to_f / n,
    mae: mae_value,
    pearson: pearson_value,
    kappa: kappa_value,
    gate: gate_for(n)
  )
end