Deploying machine learning custom model using IBM Watson

Overview:

In this project, we will understand step by step how to take a trained Scikit-Learn model and deploy it to the Cloud using Watson Machine Learning. This model can then be used for a whole bunch of applications and can even be used in different languages like Javascript, Scala, Java and Go.

In this project we’ll learn how to:

• Save the model to Watson Machine Learning
• Creating online deployments with Python
• Scoring your model using the Python API

In this project, we will mainly focus on model deployment.

For more details on Machine Learning Model that we used for this project are here

Watson Machine Learning:

IBM Watson Machine Learning is a full-service IBM Cloud offering that makes it easy for developers and data scientists to work together to integrate predictive capabilities with their applications.

1) Import and install dependencies:

install IBM Watson Machine Learning

!pip install ibm_watson_machine_learning

Import dependencies

from ibm_watson_machine_learning import APIClient
import json 
import numpy as np

2) Create deployment space:

• go to cloud.ibm.com and login
• go to catelog

• search for machine learning

• select location and lite plan
• Access in Watson studio

• go to deployments and select new deployment space

• name the deployment space and select the machine learning service that you've created in previous steps.

You will be able to see your space in deployments

3) Authentication:

from ibm_watson_machine_learning import APIClient
wml_credentials = {
                    "url":"https://us-south.ml.cloud.ibm.com",
                    "apikey":"<your api key>"
}

client = APIClient(wml_credentials)

For api key follow the following steps:

• go to ibm cloud
• select manage -> Access(IAM) -> API keys

Where to get you space ID?

# create a new deployment space in services and software
space_uid = guid_from_space_name(client, "DeploymentSpace")
print("space_uid : ", space_uid)

Other way to get the space ID:

• go to deployments -> manage

client.set.default_space(space_uid)

4) Check the software specifications:

client.software_specifications.list()

-----------------------------  ------------------------------------  ----
NAME                           ASSET_ID                              TYPE
default_py3.6                  0062b8c9-8b7d-44a0-a9b9-46c416adcbd9  base
pytorch-onnx_1.3-py3.7-edt     069ea134-3346-5748-b513-49120e15d288  base
scikit-learn_0.20-py3.6        09c5a1d0-9c1e-4473-a344-eb7b665ff687  base
spark-mllib_3.0-scala_2.12     09f4cff0-90a7-5899-b9ed-1ef348aebdee  base
ai-function_0.1-py3.6          0cdb0f1e-5376-4f4d-92dd-da3b69aa9bda  base
shiny-r3.6                     0e6e79df-875e-4f24-8ae9-62dcc2148306  base
tensorflow_2.4-py3.7-horovod   1092590a-307d-563d-9b62-4eb7d64b3f22  base
pytorch_1.1-py3.6              10ac12d6-6b30-4ccd-8392-3e922c096a92  base
tensorflow_1.15-py3.6-ddl      111e41b3-de2d-5422-a4d6-bf776828c4b7  base
scikit-learn_0.22-py3.6        154010fa-5b3b-4ac1-82af-4d5ee5abbc85  base
default_r3.6                   1b70aec3-ab34-4b87-8aa0-a4a3c8296a36  base
pytorch-onnx_1.3-py3.6         1bc6029a-cc97-56da-b8e0-39c3880dbbe7  base
tensorflow_2.1-py3.6           1eb25b84-d6ed-5dde-b6a5-3fbdf1665666  base
tensorflow_2.4-py3.8-horovod   217c16f6-178f-56bf-824a-b19f20564c49  base
do_py3.8                       295addb5-9ef9-547e-9bf4-92ae3563e720  base
autoai-ts_3.8-py3.8            2aa0c932-798f-5ae9-abd6-15e0c2402fb5  base
tensorflow_1.15-py3.6          2b73a275-7cbf-420b-a912-eae7f436e0bc  base
pytorch_1.2-py3.6              2c8ef57d-2687-4b7d-acce-01f94976dac1  base
spark-mllib_2.3                2e51f700-bca0-4b0d-88dc-5c6791338875  base
pytorch-onnx_1.1-py3.6-edt     32983cea-3f32-4400-8965-dde874a8d67e  base
spark-mllib_3.0-py37           36507ebe-8770-55ba-ab2a-eafe787600e9  base
spark-mllib_2.4                390d21f8-e58b-4fac-9c55-d7ceda621326  base
xgboost_0.82-py3.6             39e31acd-5f30-41dc-ae44-60233c80306e  base
pytorch-onnx_1.2-py3.6-edt     40589d0e-7019-4e28-8daa-fb03b6f4fe12  base
autoai-obm_3.0                 42b92e18-d9ab-567f-988a-4240ba1ed5f7  base
spark-mllib_2.4-r_3.6          49403dff-92e9-4c87-a3d7-a42d0021c095  base
xgboost_0.90-py3.6             4ff8d6c2-1343-4c18-85e1-689c965304d3  base
pytorch-onnx_1.1-py3.6         50f95b2a-bc16-43bb-bc94-b0bed208c60b  base
autoai-ts_3.9-py3.8            52c57136-80fa-572e-8728-a5e7cbb42cde  base
spark-mllib_2.4-scala_2.11     55a70f99-7320-4be5-9fb9-9edb5a443af5  base
spark-mllib_3.0                5c1b0ca2-4977-5c2e-9439-ffd44ea8ffe9  base
autoai-obm_2.0                 5c2e37fa-80b8-5e77-840f-d912469614ee  base
spss-modeler_18.1              5c3cad7e-507f-4b2a-a9a3-ab53a21dee8b  base
cuda-py3.8                     5d3232bf-c86b-5df4-a2cd-7bb870a1cd4e  base
autoai-kb_3.1-py3.7            632d4b22-10aa-5180-88f0-f52dfb6444d7  base
pytorch-onnx_1.7-py3.8         634d3cdc-b562-5bf9-a2d4-ea90a478456b  base
spark-mllib_2.3-r_3.6          6586b9e3-ccd6-4f92-900f-0f8cb2bd6f0c  base
tensorflow_2.4-py3.7           65e171d7-72d1-55d9-8ebb-f813d620c9bb  base
spss-modeler_18.2              687eddc9-028a-4117-b9dd-e57b36f1efa5  base
pytorch-onnx_1.2-py3.6         692a6a4d-2c4d-45ff-a1ed-b167ee55469a  base
do_12.9                        75a3a4b0-6aa0-41b3-a618-48b1f56332a6  base
spark-mllib_2.3-scala_2.11     7963efe5-bbec-417e-92cf-0574e21b4e8d  base
spark-mllib_2.4-py37           7abc992b-b685-532b-a122-a396a3cdbaab  base
caffe_1.0-py3.6                7bb3dbe2-da6e-4145-918d-b6d84aa93b6b  base
pytorch-onnx_1.7-py3.7         812c6631-42b7-5613-982b-02098e6c909c  base
cuda-py3.6                     82c79ece-4d12-40e6-8787-a7b9e0f62770  base
tensorflow_1.15-py3.6-horovod  8964680e-d5e4-5bb8-919b-8342c6c0dfd8  base
hybrid_0.1                     8c1a58c6-62b5-4dc4-987a-df751c2756b6  base
pytorch-onnx_1.3-py3.7         8d5d8a87-a912-54cf-81ec-3914adaa988d  base
caffe-ibm_1.0-py3.6            8d863266-7927-4d1e-97d7-56a7f4c0a19b  base
-----------------------------  ------------------------------------  ----
Note: Only first 50 records were displayed. To display more use 'limit' parameter.

5) Setting the python environment:

software_spec_uid = client.software_specifications.get_uid_by_name("default_py3.8")

6) Storing the model in space:

model_details = client.repository.store_model(model=bestAdaModFitted2, meta_props={
client.repository.ModelMetaNames.NAME: "Customer churn prediction",
client.repository.ModelMetaNames.TYPE: "scikit-learn_0.23",
client.repository.ModelMetaNames.SOFTWARE_SPEC_UID: software_spec_uid})

model_id = client.repository.get_model_uid(model_details)

7) Model deployment:

The model can be deployed in 2 ways:

Using deploy option:

• go to deployments -> spaces -> select you space name -> select the deploy option infront of model

Online model deployment method:

deployment_props = {
    client.deployments.ConfigurationMetaNames.NAME:"Sklearn Deployment", 
    client.deployments.ConfigurationMetaNames.ONLINE: {}
}

# Deploy
deployment = client.deployments.create(
    artifact_uid=model_id, 
    meta_props=deployment_props 
)
# Output result
deployment

8) Prediction using deployed model:

For prediction also we can use 2 ways:

Using UID:

payload = {"input_data":
           [
               {"fields":X_test[0:1].columns.to_numpy().tolist(), "values":X_test[0:1].to_numpy().tolist()}
           ]
          }

deployment_uid = client.deployments.get_uid(deployment)
result = client.deployments.score(deployment_uid, payload)
result

{'predictions': [{'fields': ['prediction', 'probability'],
   'values': [[0, [0.5306418238572117, 0.46935817614278824]]]}]}

Result:

pred_values = np.squeeze(result['predictions'][0]['values']);
pred_values[0]

result is : 0
Which means customer won't churn.

Using IBM Watson studio API reference:

Check the Notebook for implementation.

After deploying the model IBM watson studio generates an API reference URL for accessing the deployed model.

API_KEY = ""
token_response = requests.post('' data={"apikey": API_KEY, "grant_type": 'urn:ibm:params:oauth:grant-type:apikey'})
mltoken = token_response.json()["access_token"]

header = {'Content-Type': 'application/json', 'Authorization': 'Bearer ' + mltoken}

payload_scoring = {"input_data": [{'fields': ['gender',
    'SeniorCitizen',
    'Partner',
    'Dependents',
    'tenure',
    'PhoneService',
    'MultipleLines',
    'InternetService',
    'OnlineSecurity',
    'OnlineBackup',
    'DeviceProtection',
    'TechSupport',
    'StreamingTV',
    'StreamingMovies',
    'Contract',
    'PaperlessBilling',
    'PaymentMethod',
    'MonthlyCharges',
    'TotalCharges'],
    "values": [[0.0,
     0.0,
     0.0,
     0.0,
     -0.629446027814582,
     1.0,
     2.0,
     2.0,
     1.0,
     1.0,
     1.0,
     1.0,
     1.0,
     1.0,
     0.0,
     0.0,
     3.0,
     -1.322051671307397,
     -0.8000098038460207]]}]}

response_scoring = requests.post('<url>', json=payload_scoring, headers={'Authorization': 'Bearer ' + mltoken})
print("Scoring response")
print(response_scoring.json())

Scoring response
{'predictions': [{'fields': ['prediction', 'probability'], 'values': [[0, [0.5306418238572117, 0.46935817614278824]]]}]}

Customer will likely to continue with her subscription (won't churn)

Licences:

MIT License

Learnings:

Model Deployment using IBM Watson Studio

References:

IBM Cloud
Watson Machine Learning
Watson Studio
Example deploying with Cloud Pak for Data

Feedback

If you have any feedback, please reach out at pradnyapatil671@gmail.com

🚀 About Me

Hi, I'm Pradnya! 👋

I am an AI Enthusiast and Data science & ML practitioner