This is a migration guide for developers to migrate their existing assets from WML v3/v4 beta to WML v4.
- Specifying credentials
- Creating a client object
- Creating deployment space and setting up default deployement space for your project
- Specifying metadata for storing the model
- Storing a model (remains same)
- Specifying metadata for deployment (remains same)
- Specifying input payload for prediction (remains same)
- Getting predictions (remains same)
wml_credentials = {
"apikey":"",
"url": "",
"instance_id": ""
}NOTE:
instance_idis no longer required.urlwill change based on the region where the WML Instance is deployed in your IBM Cloud, Example:- London: https://eu-gb.ml.cloud.ibm.com
- Dallas: https://us-south.ml.cloud.ibm.com
apikey = "<APIKEY_FROM_IAM>"wml_credentials = {
"apikey": apikey,
"url": "https://eu-gb.ml.cloud.ibm.com"
}from watson_machine_learning_client import WatsonMachineLearningAPIClient
client = WatsonMachineLearningAPIClient(wml_credentials)!pip install -U ibm-watson-machine-learningfrom ibm_watson_machine_learning import APIClient
client = APIClient(wml_credentials)In IBM Cloud, goto resources and select Watson Studio instance, click on Get Started to launch Watson Studio / Cloud Pak for Data.
In Watson Studio / Cloud Pak for Data, click on the hamburger menu on the top left corner and select Deployment spaces > View all spaces.
In deployment spaces, click on New deployment space +.
Select Create an empty space when prompted.
Make sure you select the appropriate Cloud object storage service as well as Machine learning service.
NOTE: In v4 Machine learning assets are stored in Cloud Object Storage rather than in the Watson Machine Learning repository.
Once the deployment space is created, click on View Space to view the details.
Click on Settings and copy the
space ID.
Learn more about deployment space here.
space_id = "<DEPLOYMENT_SPACE_ID_FROM_WATSON_STUDIO>"client.spaces.list(limit=10)client.set.default_space(space_id)metadata = {
client.repository.ModelMetaNames.NAME: 'Personal Loan Prediction Model',
client.repository.ModelMetaNames.TYPE: "scikit-learn_0.20",
client.repository.ModelMetaNames.RUNTIME_UID: "scikit-learn_0.20-py3.6"
}sofware_spec_uid = client.software_specifications.get_id_by_name("default_py3.7")metadata = {
client.repository.ModelMetaNames.NAME: 'Personal Loan Prediction model',
client.repository.ModelMetaNames.TYPE: 'scikit-learn_0.23',
client.repository.ModelMetaNames.SOFTWARE_SPEC_UID: sofware_spec_uid
}Optionally, to get a list of software specifications run the following:
client.software_specifications.list()published_model = client.repository.store_model(model=naive_model, meta_props=metadata, \
training_data=train_set, training_target=train_labels)import json
published_model_uid = client.repository.get_model_uid(published_model)
model_details = client.repository.get_details(published_model_uid)
print(json.dumps(model_details, indent=2))deploy_meta = {
client.deployments.ConfigurationMetaNames.NAME: 'Deployment of Personal Loan Prediction',
client.deployments.ConfigurationMetaNames.ONLINE: {}
}created_deployment = client.deployments.create(published_model_uid, meta_props=deploy_meta)deployments = client.deployments.get_details()job_payload = {
client.deployments.ScoringMetaNames.INPUT_DATA: [{
'values': [[39,139,95616,3,3.4,1,483,0,0,1,0]]
}]
}did = client.deployments.get_uid(created_deployment)predictions = client.deployments.score(did, job_payload)print(json.dumps(predictions, indent=2))print(predictions["predictions"][0]['values'][0][0])