Show Menu
TOPICS×

Experiments

Model development and training occurs at the Experiment level, where an Experiment consists of an MLInstance, training runs, and scoring runs.

Create an Experiment

You can create an Experiment by performing a POST request while providing a name and a valid MLInstance ID in the request payload.
Unlike model training in the UI, creating an Experiment through an explicit API call does not automatically create and execute a training run.
API Format
POST /experiments

Request
curl -X POST \
    https://platform.adobe.io/data/sensei/experiments \
    -H 'Authorization: Bearer {ACCESS_TOKEN}' \
    -H 'x-api-key: {API_KEY}' \
    -H 'x-gw-ims-org-id: {IMS_ORG}' \
    -H 'x-sandbox-name: {SANDBOX_NAME}' \
    -H 'content-type: application/vnd.adobe.platform.sensei+json;profile=experiment.v1.json' \
    -d '{
        "name": "a name for this Experiment",
        "mlInstanceId": "46986c8f-7739-4376-8509-0178bdf32cda"
    }'

Property
Description
name
The desired name for the Experiment. The training run corresponding to this Experiment will inherit this value to be displayed in the UI as the training run name.
mlInstanceId
A valid MLInstance ID.
Response
A successful response returns a payload containing the details of the newly created Experiment including its unique identifier ( id ).
{
    "id": "5cb25a2d-2cbd-4c99-a619-8ddae5250a7b",
    "name": "A name for this Experiment",
    "mlInstanceId": "46986c8f-7739-4376-8509-0178bdf32cda",
    "created": "2019-01-01T00:00:00.000Z",
    "createdBy": {
        "userId": "Jane_Doe@AdobeID"
    },
    "updated": "2019-01-01T00:00:00.000Z",
    "createdByService": false
}

Create and execute a training or scoring run

You can create training or scoring runs by performing a POST request and providing a valid Experiment ID and specifying the run task. Scoring runs can be created only if the Experiment has an existing and successful training run. Successfully creating a training run will initialize the model training procedure and its successful completion will generate a trained model. Generating trained models will replace any previously existing ones such that an Experiment can only utilize a single trained model at any given time.
API Format
POST /experiments/{EXPERIMENT_ID}/runs

Parameter
Description
{EXPERIMENT_ID}
A valid Experiment ID.
Request
curl -X POST \
    https://platform.adobe.io/data/sensei/experiments/5cb25a2d-2cbd-4c99-a619-8ddae5250a7b/runs \
    -H 'Authorization: Bearer {ACCESS_TOKEN}' \
    -H 'x-api-key: {API_KEY}' \
    -H 'x-gw-ims-org-id: {IMS_ORG}' \
    -H 'x-sandbox-name: {SANDBOX_NAME}' \
    -H 'content-type: application/vnd.adobe.platform.sensei+json;profile=experimentRun.v1.json' \
    -d '{
        "mode": "{TASK}"
    }'

Property
Description
{TASK}
Specifies the run's task. Set this value as either train for training, score for scoring, or featurePipeline for feature pipeline.
Response
A successful response returns a payload containing the details of the newly created run including the inherited default training or scoring parameters, and the run's unique ID ( {RUN_ID} ).
{
    "id": "33408593-2871-4198-a812-6d1b7d939cda",
    "mode": "{TASK}",
    "experimentId": "5cb25a2d-2cbd-4c99-a619-8ddae5250a7b",
    "created": "2019-01-01T00:00:00.000Z",
    "createdBy": {
        "userId": "Jane_Doe@AdobeID"
    },
    "updated": "2019-01-01T00:00:00.000Z",
    "createdBySchedule": false,
    "tasks": [
        {
            "name": "{TASK}",
            "parameters": [
                {
                    "key": "parameter",
                    "value": "parameter value"
                }
            ]
        }
    ]
}

Retrieve a list of Experiments

You can retrieve a list of Experiments belonging to a particular MLInstance by performing a single GET request and providing a valid MLInstance ID as a query parameter. For a list of available queries, refer to the appendix section on query parameters for asset retrieval .
API Format
GET /experiments
GET /experiments?property=mlInstanceId=={MLINSTANCE_ID}

Parameter
Description
{MLINSTANCE_ID}
Provide a valid MLInstance ID to retrieve a list of Experiments belonging to that particular MLInstance.
Request
curl -X GET \
    https://platform.adobe.io/data/sensei/experiments?property=mlInstanceId==46986c8f-7739-4376-8509-0178bdf32cda \
    -H 'Authorization: Bearer {ACCESS_TOKEN}' \
    -H 'x-api-key: {API_KEY}' \
    -H 'x-gw-ims-org-id: {IMS_ORG}' \
    -H 'x-sandbox-name: {SANDBOX_NAME}'

Response
A successful response returns a list of Experiments sharing the same MLInstance ID ( {MLINSTANCE_ID} ).
{
    "children": [
        {
            "id": "5cb25a2d-2cbd-4c99-a619-8ddae5250a7b",
            "name": "A name for this Experiment",
            "mlInstanceId": "46986c8f-7739-4376-8509-0178bdf32cda",
            "created": "2019-01-01T00:00:00.000Z",
            "updated": "2019-01-01T00:00:00.000Z",
            "createdByService": false
        },
        {
            "id": "6cb25a2d-2cbd-4c99-a619-8ddae5250a7b",
            "name": "Training Run 1",
            "mlInstanceId": "46986c8f-7839-4376-8509-0178bdf32cda",
            "created": "2019-01-01T00:00:00.000Z",
            "updated": "2019-01-01T00:00:00.000Z",
            "createdByService": false
        },
        {
            "id": "7cb25a2d-2cbd-4c99-a619-8ddae5250a7b",
            "name": "Training Run 2",
            "mlInstanceId": "46986c8f-7939-4376-8509-0178bdf32cda",
            "created": "2019-01-01T00:00:00.000Z",
            "updated": "2019-01-01T00:00:00.000Z",
            "createdByService": false
        }
    ],
    "_page": {
        "property": "deleted==false",
        "count": 3
    }
}

Retrieve a specific Experiment

You can retrieve the details of a specific Experiment by performing a GET request that includes the desired Experiment's ID in the request path.
API Format
GET /experiments/{EXPERIMENT_ID}

Parameter
Description
{EXPERIMENT_ID}
A valid Experiment ID.
Request
curl -X GET \
    https://platform.adobe.io/data/sensei/experiments/5cb25a2d-2cbd-4c99-a619-8ddae5250a7b \
    -H 'Authorization: Bearer {ACCESS_TOKEN}' \
    -H 'x-api-key: {API_KEY}' \
    -H 'x-gw-ims-org-id: {IMS_ORG}' \
    -H 'x-sandbox-name: {SANDBOX_NAME}'

Response
A successful response returns a payload containing the details of the requested Experiment.
{
    "id": "5cb25a2d-2cbd-4c99-a619-8ddae5250a7b",
    "name": "A name for this Experiment",
    "mlInstanceId": "46986c8f-7739-4376-8509-0178bdf32cda",
    "created": "2019-01-01T00:00:00.000Z",
    "createdBy": {
        "userId": "Jane_Doe@AdobeID"
    },
    "updated": "2019-01-01T00:00:00.000Z",
    "createdByService": false
}

Retrieve a list of Experiment runs

You can retrieve a list of training or scoring runs belonging to a particular Experiment by performing a single GET request and providing a valid Experiment ID. To help filter results, you can specify query parameters in the request path. For a complete list of available query parameters, see the appendix section on query parameters for asset retrieval .
When combining multiple query parameters, they must be separated by ampersands (&).
API Format
GET /experiments/{EXPERIMENT_ID}/runs
GET /experiments/{EXPERIMENT_ID}/runs?{QUERY_PARAMETER}={VALUE}
GET /experiments/{EXPERIMENT_ID}/runs?{QUERY_PARAMETER_1}={VALUE_1}&{QUERY_PARAMETER_2}={VALUE_2}

Parameter
Description
{EXPERIMENT_ID}
A valid Experiment ID.
{QUERY_PARAMETER}
One of the available query parameters used to filter results.
{VALUE}
The value for the preceding query parameter.
Request
The following request contains a query and retrieves a list of training runs belonging to some Experiment.
curl -X GET \
    https://platform.adobe.io/data/sensei/experiments/5cb25a2d-2cbd-4c99-a619-8ddae5250a7b/runs?property=mode==train \
    -H 'Authorization: Bearer {ACCESS_TOKEN}' \
    -H 'x-api-key: {API_KEY}' \
    -H 'x-gw-ims-org-id: {IMS_ORG}' \
    -H 'x-sandbox-name: {SANDBOX_NAME}'

Response
A successful response returns a payload containing a list of runs and each of their details including their Experiment run ID ( {RUN_ID} ).
{
    "children": [
        {
            "id": "33408593-2871-4198-a812-6d1b7d939cda",
            "mode": "train",
            "experimentId": "5cb25a2d-2cbd-4c99-a619-8ddae5250a7b",
            "created": "2019-01-01T00:00:00.000Z",
            "createdBy": {
                "userId": "Jane_Doe@AdobeID"
            },
            "createdBySchedule": false
        }
    ],
    "_page": {
        "property": "mode==train,experimentId==5cb25a2d-2cbd-4c99-a619-8ddae5250a7b,deleted==false",
        "totalCount": 1,
        "count": 1
    }
}

Update an Experiment

You can update an existing Experiment by overwriting its properties through a PUT request that includes the target Experiment's ID in the request path and providing a JSON payload containing updated properties.
In order to ensure the success of this PUT request, it is suggested that first you perform a GET request to retrieve the Experiment by ID . Then, modify and update the returned JSON object and apply the entirety of the modified JSON object as the payload for the PUT request.
The following sample API call updates an Experiments's name while having these properties initially:
{
    "name": "A name for this Experiment",
    "mlInstanceId": "46986c8f-7739-4376-8509-0178bdf32cda",
    "created": "2019-01-01T00:00:00.000Z",
    "createdBy": {
        "userId": "Jane_Doe@AdobeID"
    },
    "createdByService": false
}

API Format
PUT /experiments/{EXPERIMENT_ID}

Parameter
Description
{EXPERIMENT_ID}
A valid Experiment ID.
Request
curl -X PUT \
    https://platform.adobe.io/data/sensei/experiments/5cb25a2d-2cbd-4c99-a619-8ddae5250a7b \
    -H 'Authorization: Bearer {ACCESS_TOKEN}' \
    -H 'x-api-key: {API_KEY}' \
    -H 'x-gw-ims-org-id: {IMS_ORG}' \
    -H 'x-sandbox-name: {SANDBOX_NAME}' \
    -H 'content-type: application/vnd.adobe.platform.sensei+json;profile=experiments.v1.json' \
    -d '{
        "name": "An upated name",
        "mlInstanceId": "46986c8f-7739-4376-8509-0178bdf32cda",
        "created": "2019-01-01T00:00:00.000Z",
        "createdBy": {
            "userId": "Jane_Doe@AdobeID"
        },
        "createdByService": false
    }'

Response
A successful response returns a payload containing the Experiment's updated details.
{
    "id": "5cb25a2d-2cbd-4c99-a619-8ddae5250a7b",
    "name": "An updated name",
    "mlInstanceId": "46986c8f-7739-4376-8509-0178bdf32cda",
    "created": "2019-01-01T00:00:00.000Z",
    "createdBy": {
        "userId": "Jane_Doe@AdobeID"
    },
    "updated": "2019-01-02T00:00:00.000Z",
    "createdByService": false
}

Delete an Experiment

You can delete a single Experiment by performing a DELETE request that includes the target Experiment's ID in the request path.
API Format
DELETE /experiments/{EXPERIMENT_ID}

Parameter
Description
{EXPERIMENT_ID}
A valid Experiment ID.
Request
curl -X DELETE \
    https://platform.adobe.io/data/sensei/experiments/5cb25a2d-2cbd-4c99-a619-8ddae5250a7b \
    -H 'Authorization: Bearer {ACCESS_TOKEN}' \
    -H 'x-api-key: {API_KEY}' \
    -H 'x-gw-ims-org-id: {IMS_ORG}' \
    -H 'x-sandbox-name: {SANDBOX_NAME}'

Response
{
    "title": "Success",
    "status": 200,
    "detail": "Experiment successfully deleted"
}

Delete Experiments by MLInstance ID

You can delete all Experiments belonging to a particular MLInstance by performing a DELETE request that includes the MLInstance ID as a query parameter.
API Format
DELETE /experiments?mlInstanceId={MLINSTANCE_ID}

Parameter
Description
{MLINSTANCE_ID}
A valid MLInstance ID.
Request
curl -X DELETE \
    https://platform.adobe.io/data/sensei/experiments?mlInstanceId=46986c8f-7739-4376-8509-0178bdf32cda \
    -H 'Authorization: Bearer {ACCESS_TOKEN}' \
    -H 'x-api-key: {API_KEY}' \
    -H 'x-gw-ims-org-id: {IMS_ORG}' \
    -H 'x-sandbox-name: {SANDBOX_NAME}'

Response
{
    "title": "Success",
    "status": 200,
    "detail": "Experiments successfully deleted"
}