predictions.md

Predictions

• A Prediction is a prediction made by a Model on a Document

After a Model have been trained and a Document have been created a Prediction can be made.

>> las predictions create las:document:fe1cfdb5de254bb68d7c2763d3861baf las:model:03bce190cf4343efb3eb5065bd999844  
{
  "predictionId": "las:prediction:b68233f457f54342bd65acbdb7053637",
  "modelId": "las:model:03bce190cf4343efb3eb5065bd999844",
  "documentId": "las:document:fe1cfdb5de254bb68d7c2763d3861baf",
  "predictions": [
    {
      "label": "bank_account",
      "value": "05395598149",
      "confidence": 0.9996034754459902
    },
    {
      "label": "customer_id",
      "value": "218395942l7317305",
      "confidence": 0.9697508355421477
    },
    {
      "label": "total_amount",
      "value": "439.00",
      "confidence": 0.9529096553664052
    },
    {
      "label": "due_date",
      "value": "2020-01-25",
      "confidence": 0.035909883433622676
    }
  ],
  "timestamp": 1612950528,
  "inferenceTime": 3.6797611713409424
}

Common for all Predictions is that they have a list of dictionaries on the form

    {
      "label": "<field-name>",
      "value": "<predicted-value>",
      "confidence": <confidence-of-prediction>
    }

By delivering confidence with every prediction the user can make decisions on whether to use or discard the result depending on the accuracy required for a specific case. In the example above the confidence reveals that the due_date is likely to be wrong, while the bank_account is very likely to be correct.

Predictions can also be listed, which can be useful for making statistics or inspecting the performance of the model.

>>las predictions list
{
  "predictions": [
    {
        ...
    }
  ],
  "nextToken": null, 
}