Python SDKIntroductory pagesNew features

Relation Confidence

The relation confidence feature includes a confidence value for each relation predicted.

Example NLP API results:

Specifically the following is a snippet showing an example of a measurement relation’s confidence:

"measurements": [
  {
    "label": "RMS0",
    "attributes": {
      "confidence": 0.9999994039535522
    },
    "args": {
      "value": {
        "ref": "E0",
        "text": [
          "97/52"
        ]
      },
      "subject": {
        "ref": "E3",
        "text": [
          "Blood pressure"
        ]
      }
    }
  },

Changed

Python API

All the relations now have a .confidence attribute which is a float (e.g. confidence).

Not all relation types currently support confidence scores, so the .confidence attribute may be None. See NLP API documentation on which relations will include confidence scores.

Example usage:

import json
import pathlib
from emtellipro.data import AnnotatedDocument
import emtellipro.load
json_data = pathlib.Path('data/relation_confidence.json').read_text()
doc_dict = json.loads(json_data)['documents'][0]
ann_doc = AnnotatedDocument(doc_dict)
for measurement in ann_doc.relations['measurement']:
    print(
        measurement.value.text,
        measurement.subject.text,
        measurement.confidence,
    )
Output of above code.
['97/52'] ['Blood pressure'] 0.9999994039535522
['79'] ['pulse'] 0.9999290704727173

Schema changes

Each relation table now contains a confidence column:

  • experiencerrelation.confidence
  • followuprelation.confidence
  • measurementrelation.confidence
  • imagelinkrelation.confidence
  • qualifierrelation.confidence
  • medicationrelation.confidence
  • temporalityrelation.confidence
  • reportedeventrelation.confidence
  • anatomicsiterelation.confidence

For example, here is the new schema for measurementrelation (in PostgreSQL):

                                   Table "public.measurementrelation"
   Column   |       Type       | Collation | Nullable |                     Default
------------+------------------+-----------+----------+-------------------------------------------------
 id         | integer          |           | not null | nextval('measurementrelation_id_seq'::regclass)
 subject_id | integer          |           |          |
 value_id   | integer          |           |          |
 confidence | double precision |           |          |
Indexes:
    "pk_measurementrelation" PRIMARY KEY, btree (id)
Foreign-key constraints:
    "fk_measurementrelation_subject_id_entity" FOREIGN KEY (subject_id) REFERENCES entity(id)
    "fk_measurementrelation_value_id_entity" FOREIGN KEY (value_id) REFERENCES entity(id)

The following is an example of what you can expect to see in that table:

testdb=# select * from measurementrelation;
 id | subject_id | value_id |     confidence
----+------------+----------+--------------------
  1 |          3 |        1 | 0.9999994039535522
  2 |          4 |        2 | 0.9999290704727173
(2 rows)