Partager via

Data Science Toolkit - Custom model parser service

  • Article

The Custom Model Parser service lets you check the validity of decision trees written in our Bonsai Language. You should use this service to identify and resolve any Bonsai syntax or feature errors before using the Custom Model Service to upload trees for use in campaigns.

This page walks you through the validation process. For background information on the purpose of custom models, see Custom Models.

Step 1. Base64-encode your Bonsai decision tree

Once you have written your decision tree in our Bonsai Language, base64-encode it.

$ cat decision_tree.txt 
 
# This tree determines a bid price as follows:
# 1. If the user is in California, and the hour is between 12pm and 14pm there, bid $1.50.
# 2. If the user is in New York, and the hour is between 1am and 3am there, bid $0.10.
# 3. Otherwise, bid $0.50.
 
if every region = "US:CA", user_hour range (12,14):
        1.5
elif every region = "US:NY", user_hour range (1,3):
        0.1
else:
        0.5
$ cat decision_tree.txt |base64
 
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

Step 2. Create a JSON file that contains your encoded tree

Create a JSON file as shown below. The main object must contain a custom-model-parser object with the encoded tree as a string in the model_text field.

$ cat check_tree.json 
 
{
    "custom-model-parser": {
                "model_text": "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"
    }
}

Step 3. POST file to custom model parser service

Make a POST call to the Custom Model Parser Service as shown below.

$ curl -b cookies -c cookies -X POST -d @check_tree.json 'https://api.appnexus.com/custom-model-parser'

Step 4. Check response

Tree is valid

If your Bonsai decision tree is valid, the custom-model-parser object in the response will contain the following fields:

Field Description
model_text Your decision tree.
size The size of your decision tree, in bytes.
Size Limit: Your tree must be smaller than 3 MB, or 3,145,728 bytes. If it is larger than that, you will not be able to add your tree using the Custom Model Service.

Note

The custom-model-parser endpoint used to return the Lisp storage format of a decision tree under the "model_text" field. We are deprecating this internal representation, as it was only an intermediate step and is not used anywhere on the platform. The field will now always contain the value "<removed>". The "size" field on the response will continue to be filled accurately and you should keep relying on it to know if your model is too large for upload.

{
    "response": {
        "service": "custom-model-parser",
        "method": "post",
        "custom-model-parser": {
            "model_text": "(if (and (region = 3922) (user_hour range 12 14)) 1.5 (if (and (region = 3950) (user_hour range 1 3)) 0.1 0.5))",
            "size": 111
        }
    }
    }

Tree is not valid

If your Bonsai decision tree is not valid, the error field in the response will identify the issue. An error can result from either invalid Bonsai syntax or invalid usage of Bonsai features. See Error Messages below for more details.

{
    "response": {
        "error_id": "SYNTAX",
        "error": "ERROR: User hour expected on line 6 at column 49; user_hour must be between 0 and 23; found 24",
        "error_description": null,
        "service": "custom-model-parser",
        "method": "post",
        "error_code": "INVALID_SYNTAX"
    }
}

Error messages

Syntax errors

  • Illegal character found:

    ERROR: Illegal character <character> found on line <line> at column <column>

  • Second root node found:

    ERROR: End of file expected; <token> found.

  • Numeric operator used with non-numeric type:

    ERROR: Numeric operator <operator> invalid with non-numeric type <type> at line <line> at column <column>.

  • Incorrect number of elements in a list (like having 3 elements in a range expression):

    ERROR: Invalid number of elements in list at line <line> at column <column>. Required number of elements is <required number>, <number> elements found

Feature errors

There are two types of features: Features we validate against the DB, and features whose values must be certain numeric values. Validating against the DB, the error message seen for an invalid value is:

ERROR: <description> expected on line <line> at column <column>; found <value found>.

Validating numeric values, the error message seen for an invalid value is:

ERROR: <description> expected on line <line> at column <column>; <numeric restriction>

Features, descriptions, numeric restrictions are as follows:

  • country: Country string
  • region: Region string
  • city: City string
  • supply_type: Supply type
  • domain: URL string
  • browser: Browser string
  • carrier: Carrier string
  • os_family: OS family string
  • placement: Valid placement id
  • size: Size string
  • placement_group: Valid placement group id
  • publisher: Valid publisher id
  • mobile_app: Valid mobile app id
  • cookie_age: Cookie age in minutes
  • user_hour: User hour must be between 0 and 23
  • user_day: User day must be between 0 and 6
  • OBJECT[ID].lifetime_frequency: Here the object is advertiser, line_item, or campaign (representing split), and ID is the object ID. The life frequency must be a positive integer. In this scenario, object is advertiser.
  • OBJECT[ID].day_freq: Here the object is advertiser, line_item, or campaign (representing split), and ID is the object ID. The day frequency must be a positive integer. In this scenario, object is advertiser.
  • advertiser_recency: Advertiser recency must be positive integer or -1
  • device_type: Device type string
  • estimated_iab_viewthrough_rate: IAB viewthrough rate must be a number between 0 and 1