Deploy Scikit-learn Models on EAS via PMML Processor - Platform for AI

Deploy PMML models as online services for inference using the built-in processor.

Prerequisites

Models exported from Machine Learning Designer deploy directly to EAS. Models trained with open-source frameworks (scikit-learn, XGBoost) require conversion to PMML format before deployment. For conversion instructions, see Export a general-purpose model.

For PMML specification details, see PMML 4.4 - General Structure.

Missing value handling

If the PMML model file does not specify a missing value (isMissing) policy for features, the processor applies default values by data type.

Data type	Default value
BOOLEAN	false
DOUBLE	0.0
FLOAT	0.0
INT	0
STRING	""

Deploy the service

Set processor type to pmml in your configuration file when deploying with EASCMD.

{
  "name": "eas_lr_example",
  "processor": "pmml",
  "model_path": "http://examplebucket.oss-cn-shanghai.aliyuncs.com/models/lr.pmml",
  "metadata": {
    "instance": 1,
    "cpu": 1
  }
}

For EASCMD deployment instructions, see Deploy a service using EASCMD or DSW.

To deploy from the console, see Deploy a service from the console.

Call the service

Complete these steps to call the deployed service.

Go to Elastic Algorithm Service (EAS) page. Find your service and click View Invocation Method in Service Method column to obtain the endpoint and authentication token.

Construct a service request.

PMML service input and output must use JSON format. A single request can contain multiple samples with feature data in key-value format.

[
    {
        "address": 12,
        "age": 22,
        "ed": 4,
        "marital": 3.0,
        "region": 2.0,
        "tenure": 4.0
    },
    {
        "address": 2,
        "age": 34,
        "ed": 6,
        "marital": 1.3,
        "region": 2.1,
        "tenure": 4.2
    }
]

Note

Remove line breaks and spaces from the JSON request to reduce network traffic and improve performance.

Send the service request.

Use one of these methods to send the request:

Important

Passing the token in HTTP header transmits it in plaintext. Using EAS SDK signs the request with the token before transmission, providing higher security.

Use curl to test

Pass the authentication token in HTTP header when sending requests to the service endpoint.

// Send the request.
curl -v 18284888792***.cn-shanghai.pai-eas.aliyuncs.com/api/predict/pmml_test \
      -H 'Authorization: YmE3NDkyMzdiMzNmMGM3ZmE4ZmNjZDk0M2NiMDA***' \
      -d '[{"region": 2.0, "marital":3.0,"tenure":4.0, "age":22,"address":12,"ed":4},{"region": 2.1, "marital":1.3,"tenure":4.2, "age":34,"address":2,"ed":6}]'

Parameters:

18284888792***.cn-shanghai.pai-eas.aliyuncs.com/api/predict/pmml_test: Replace with endpoint from Step 1.
Authorization: Replace with token from Step 1.

Sample response:

[{"p_0":0.2856724025030001,"p_1":0.7143275974969999},{"p_0":0.18324957921381624,"p_1":0.8167504207861838}]

Use Python SDK. For SDK documentation, see Python SDK usage guide.

Example:

#!/usr/bin/env python

from eas_prediction import PredictClient
from eas_prediction import StringRequest

if __name__ == '__main__':
    # Replace with endpoint and service name from Step 1.
    client = PredictClient('1828488879222***.cn-shanghai.pai-eas.aliyuncs.com', 'pmml_test')
    # Replace with token from Step 1.
    client.set_token('YmE3NDkyMzdiMzNmMGM3ZmE4ZmNjZDk0M2NiMDA***')
    client.init()


    req = StringRequest('[{"region": 2.0, "marital":3.0,"tenure":4.0, "age":22,"address":12,"ed":4},{"region": 2.1, "marital":1.3,"tenure":4.2, "age":34,"address":2,"ed":6}]')
    for x in range(100):
        resp = client.predict(req)
        print(resp)

For SDKs in other languages, see Service invocation SDKs.

Related information

For other processors, see Deploy a service using a pre-built processor.
For custom processor development, see Deploy a service using a custom processor.