{
  "alt": "Dot chart with two dots per label count. Recalibrated Jev sits near the right at every count, with a wide range only at the smallest. The classifier starts far to the left and approaches raw Jev only at the full sample.",
  "caption": "Read down the rows: both arms use the same labeled reviews, and the line marks raw Jev. Each count is every local label used, including the classifier's calibration labels; the reviews being scored are not among them. Whiskers are ranges across draws and not confidence intervals. Draws differ from the previous chapter's curve, so its numbers differ a little.",
  "data": [
    {
      "conf_high": 2475,
      "conf_low": 0,
      "labels": "50 labels",
      "model": "Classifier, fit and calibrated on these labels",
      "n_labels": 50,
      "net": 1502
    },
    {
      "conf_high": 5337,
      "conf_low": 3387,
      "labels": "50 labels",
      "model": "Jev, recalibrated on these labels",
      "n_labels": 50,
      "net": 5058
    },
    {
      "conf_high": 3709,
      "conf_low": 1570,
      "labels": "100 labels",
      "model": "Classifier, fit and calibrated on these labels",
      "n_labels": 100,
      "net": 2981
    },
    {
      "conf_high": 5329,
      "conf_low": 5029,
      "labels": "100 labels",
      "model": "Jev, recalibrated on these labels",
      "n_labels": 100,
      "net": 5188
    },
    {
      "conf_high": 3885,
      "conf_low": 1858,
      "labels": "200 labels",
      "model": "Classifier, fit and calibrated on these labels",
      "n_labels": 200,
      "net": 3257
    },
    {
      "conf_high": 5341,
      "conf_low": 5072,
      "labels": "200 labels",
      "model": "Jev, recalibrated on these labels",
      "n_labels": 200,
      "net": 5257
    },
    {
      "conf_high": 4387,
      "conf_low": 3338,
      "labels": "500 labels",
      "model": "Classifier, fit and calibrated on these labels",
      "n_labels": 500,
      "net": 4073
    },
    {
      "conf_high": 5336,
      "conf_low": 5185,
      "labels": "500 labels",
      "model": "Jev, recalibrated on these labels",
      "n_labels": 500,
      "net": 5285
    },
    {
      "conf_high": 4628,
      "conf_low": 4386,
      "labels": "1,000 labels",
      "model": "Classifier, fit and calibrated on these labels",
      "n_labels": 1000,
      "net": 4503
    },
    {
      "conf_high": 5335,
      "conf_low": 5248,
      "labels": "1,000 labels",
      "model": "Jev, recalibrated on these labels",
      "n_labels": 1000,
      "net": 5300
    },
    {
      "conf_high": 4683,
      "conf_low": 4490,
      "labels": "2,000 labels",
      "model": "Classifier, fit and calibrated on these labels",
      "n_labels": 2000,
      "net": 4581
    },
    {
      "conf_high": 5337,
      "conf_low": 5225,
      "labels": "2,000 labels",
      "model": "Jev, recalibrated on these labels",
      "n_labels": 2000,
      "net": 5307
    },
    {
      "conf_high": 4872,
      "conf_low": 4749,
      "labels": "5,000 labels",
      "model": "Classifier, fit and calibrated on these labels",
      "n_labels": 5000,
      "net": 4816
    },
    {
      "conf_high": 5344,
      "conf_low": 5270,
      "labels": "5,000 labels",
      "model": "Jev, recalibrated on these labels",
      "n_labels": 5000,
      "net": 5320
    },
    {
      "conf_high": 4974,
      "conf_low": 4897,
      "labels": "10,000 labels",
      "model": "Classifier, fit and calibrated on these labels",
      "n_labels": 10000,
      "net": 4939
    },
    {
      "conf_high": 5342,
      "conf_low": 5304,
      "labels": "10,000 labels",
      "model": "Jev, recalibrated on these labels",
      "n_labels": 10000,
      "net": 5333
    },
    {
      "conf_high": 5045,
      "conf_low": 5008,
      "labels": "20,000 labels",
      "model": "Classifier, fit and calibrated on these labels",
      "n_labels": 20000,
      "net": 5032
    },
    {
      "conf_high": null,
      "conf_low": null,
      "labels": "20,000 labels",
      "model": "Jev, recalibrated on these labels",
      "n_labels": 20000,
      "net": 5342
    }
  ],
  "id": "value-by-labels",
  "notes": [
    "Platt scaling in place of isotonic calibration gives the same reading."
  ],
  "source": "Amazon Reviews 2023, coffee products; labels from reviews of other products; policy scored on the sampled evaluation reviews, weighted; engine calls October 2026; assumed costs",
  "spec": {
    "colors": {
      "Classifier, fit and calibrated on these labels": "cat5",
      "Jev, recalibrated on these labels": "cat1"
    },
    "group": "model",
    "hi": "conf_high",
    "lo": "conf_low",
    "reference": 5076,
    "referenceLabel": "Raw Jev",
    "sort": "data",
    "x": "net",
    "xDomain": [
      0,
      6000
    ],
    "xFormat": {
      "digits": 0,
      "prefix": "$",
      "thousands": true
    },
    "xLabel": "Net value (US dollars per thousand reviews)",
    "y": "labels"
  },
  "subtitle": "Net value of the contact policy per thousand reviews against local labels used, weighted; whiskers run from the lowest to the highest draw",
  "title": "For the same labels, recalibrating Jev earns more than fitting a classifier at every size tried",
  "type": "dot"
}