Observation Create

https://api.sigopt.com/v1/experiments/EXPERIMENT_ID/observations

Creates a new Observation. Every observation reports on a set of parameters which can come from a Suggestion or manually specified Assignments.

Request Method: POST

Parameters

Name	Type	Required?	Description
assignments	Assignments	N	Manually specify values that the parameters held during this trial. You must specify exactly one value for each parameter in the experiment. Prefer using suggestion.
failed	boolean	N	A boolean indicating whether this observation failed for some reason. SigOpt takes this into consideration when optimizing. It is invalid for both failed to be true, and any of value, value_stddev, or values to be present. The default value is false.
metadata	Metadata	N	Optional user-provided object. See Using Metadata for more information.
suggestion	string	N	The id of the Suggestion you are reporting on. Preferred over assignments.
values	array<Metric Evaluation>	N	An array of Metric Evaluation objects. For experiments with more than one Metric, you must use this instead of value and value_stddev.

Deprecated Parameters

These parameters should no longer be used because there are better alternatives.

Name	Type	Required?	Description
value	float	N	The observed metric value from this trial.
value_stddev	float	N	The standard deviation in value that was observed in this trial.

Response

Observation object.

Examples

Create an Observation

Python

observation = conn.experiments(EXPERIMENT_ID).observations().create(
  suggestion="1",
  values=[
    dict(
      name="accuracy",
      value=0.95
      ),
    dict(
      name="loss",
      value=123.45
      )
    ]
  )

Bash

accuracy=0.95
loss=123.45
OBSERVATION=`curl -s -X POST https://api.sigopt.com/v1/experiments/EXPERIMENT_ID/observations -u "$SIGOPT_API_TOKEN": \
  -H 'Content-Type: application/json' \
  -d "{\"suggestion\":\"1\",\"values\":[{\"name\":\"accuracy\",\"value\":$accuracy},{\"name\":\"loss\",\"value\":$loss}]}"`

Java

Observation observation = new Experiment(EXPERIMENT_ID).observations().create()
  .data(
    new Observation.Builder()
      .suggestion("1")
      .values(java.util.Arrays.asList(
        new Value.Builder()
          .name("accuracy")
          .value(0.95)
          .build(),
        new Value.Builder()
          .name("loss")
          .value(123.45)
          .build()
        ))
      .build()
    )
  .call();

Response

{
  "assignments": {
    "degree": 2,
    "gamma": 3.6,
    "kernel": "rbf"
  },
  "created": 1414800000,
  "experiment": "1",
  "failed": false,
  "id": "1",
  "metadata": null,
  "object": "observation",
  "suggestion": "1",
  "value": 1,
  "value_stddev": null,
  "values": [
    {
      "name": "Accuracy",
      "object": "metric_evaluation",
      "value": 1,
      "value_stddev": null
    }
  ]
}

Create an Observation with custom Assignments

Python

observation = conn.experiments(EXPERIMENT_ID).observations().create(
  assignments=dict(
    degree=2,
    gamma=3.6,
    kernel="rbf"
    ),
  values=[
    dict(
      name="Accuracy",
      value=1
      )
    ]
  )

Bash

OBSERVATION=`curl -s -X POST https://api.sigopt.com/v1/experiments/EXPERIMENT_ID/observations -u "$SIGOPT_API_TOKEN": \
  -H 'Content-Type: application/json' \
  -d "{\"assignments\":{\"degree\":2,\"gamma"\:3.6,\"kernel\":\"rbf\"},\"values\":[{\"name\":\"Accuracy\",\"value\":1}]}"`

Java

Observation observation = new Experiment(EXPERIMENT_ID).observations().create()
  .data(
    new Observation.Builder()
      .assignments(new Assignments.Builder()
        .set("degree", 2)
        .set("gamma", 3.6)
        .set("kernel", "rbf")
        .build())
      .values(java.util.Arrays.asList(
        new MetricEvaluation.Builder()
          .name("Accuracy")
          .value(1)
          .build()
        ))
      .build()
    )
  .call();

Response

{
  "assignments": {
    "degree": 2,
    "gamma": 3.6,
    "kernel": "rbf"
  },
  "created": 1414800000,
  "experiment": "1",
  "failed": false,
  "id": "1",
  "metadata": null,
  "object": "observation",
  "suggestion": "1",
  "value": 1,
  "value_stddev": null,
  "values": [
    {
      "name": "Accuracy",
      "object": "metric_evaluation",
      "value": 1,
      "value_stddev": null
    }
  ]
}

Create a Failed Observation

Python

observation = conn.experiments(EXPERIMENT_ID).observations().create(
  failed=True,
  suggestion="2"
  )

Bash

OBSERVATION=`curl -s -X POST https://api.sigopt.com/v1/experiments/EXPERIMENT_ID/observations -u "$SIGOPT_API_TOKEN": \
  -H 'Content-Type: application/json' \
  -d "{\"suggestion\":\"2\",\"failed\":true}"`

Java

Observation observation = new Experiment(EXPERIMENT_ID).observations().create()
  .data(
    new Observation.Builder()
      .failed(true)
      .suggestion("2")
      .build()
    )
  .call();

Response

{
  "assignments": {
    "degree": 2,
    "gamma": 3.6,
    "kernel": "rbf"
  },
  "created": 1414800000,
  "experiment": "1",
  "failed": true,
  "id": "3",
  "metadata": null,
  "object": "observation",
  "suggestion": "2",
  "value": null,
  "value_stddev": null,
  "values": []
}

Create an Observation with Metadata

Python

observation = conn.experiments(EXPERIMENT_ID).observations().create(
  metadata=dict(
    estimated_training_time_s=302
    ),
  suggestion="1",
  values=[
    dict(
      name="Accuracy",
      value=1,
      value_stddev=0.1
      )
    ]
  )

Bash

OBSERVATION=`curl -s -X POST https://api.sigopt.com/v1/experiments/EXPERIMENT_ID/observations -u "$SIGOPT_API_TOKEN": \
  -H 'Content-Type: application/json' \
  -d "{\"suggestion\":\"1\",\"values\":[{\"name\":\"Accuracy\",\"value\":1,\"value_stddev\":0.1}],\"metadata\":{\"estimated_training_time_s\":302}}"`

Java

Observation observation = new Experiment(EXPERIMENT_ID).observations().create()
  .data(
    new Observation.Builder()
      .metadata(new Metadata.Builder()
        .set("estimated_training_time_s", 302)
        .build())
      .suggestion("1")
      .values(java.util.Arrays.asList(
        new MetricEvaluation.Builder()
          .name("Accuracy")
          .value(1)
          .valueStddev(0.1)
          .build()
        ))
      .build()
    )
  .call();

Response

{
  "assignments": {
    "degree": 2,
    "gamma": 3.6,
    "kernel": "rbf"
  },
  "created": 1414800000,
  "experiment": "1",
  "failed": false,
  "id": "2",
  "metadata": {
    "estimated_training_time_s": 302
  },
  "object": "observation",
  "suggestion": "1",
  "value": 1,
  "value_stddev": 0.1,
  "values": [
    {
      "name": "Accuracy",
      "object": "metric_evaluation",
      "value": 1,
      "value_stddev": 0.1
    }
  ]
}