{
  "alt": "Dot chart with one row per training size. The classifier's net value rises with each larger size and reaches the vertical line marking Jev only at the largest size.",
  "caption": "Read down the rows: each dot is the mean of refits on that many labeled reviews, and its whisker runs from the lowest to the highest refit, which is not a confidence interval. Each count includes the fifth held out for calibration. The vertical line is Jev, which trained on none of them. Costs are the next chapter's assumptions, and checking Jev on any review set still needs some labels.",
  "data": [
    {
      "conf_high": 3009,
      "conf_low": 0,
      "labels": "50 labels",
      "net": 1772
    },
    {
      "conf_high": 3204,
      "conf_low": 0,
      "labels": "100 labels",
      "net": 2224
    },
    {
      "conf_high": 3758,
      "conf_low": 2882,
      "labels": "200 labels",
      "net": 3307
    },
    {
      "conf_high": 4191,
      "conf_low": 3684,
      "labels": "500 labels",
      "net": 4022
    },
    {
      "conf_high": 4559,
      "conf_low": 4208,
      "labels": "1,000 labels",
      "net": 4466
    },
    {
      "conf_high": 4673,
      "conf_low": 4331,
      "labels": "2,000 labels",
      "net": 4551
    },
    {
      "conf_high": 4877,
      "conf_low": 4731,
      "labels": "5,000 labels",
      "net": 4816
    },
    {
      "conf_high": 4986,
      "conf_low": 4906,
      "labels": "10,000 labels",
      "net": 4944
    },
    {
      "conf_high": 5058,
      "conf_low": 5036,
      "labels": "20,000 labels",
      "net": 5047
    }
  ],
  "id": "labels-to-catch-up",
  "source": "Amazon Reviews 2023, coffee products; 5,000 sampled reviews, weighted; ten refits per size, three at the largest; engine calls October 2026; assumed costs",
  "spec": {
    "hi": "conf_high",
    "lo": "conf_low",
    "reference": 5076,
    "referenceLabel": "Jev, no labels",
    "sort": "data",
    "valueLabels": true,
    "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 labeled reviews used to train the classifier, weighted",
  "title": "The review classifier needs thousands of labels to approach what Jev earns with none",
  "type": "dot"
}