We used the bank marketing dataset to train, configure and deploy a machine learning model using Azure Machine Learning Studio.
Below is the architectural diagram that we will follow
The steps are as follows :
- Load your dataset to azure ML studio
- Run the AutoML model using classification/li>
- Select the best model and deploy using azure container instance(ACI)
- Enable insights to keep track of what is happening.
- Use the REST endpoint to interact with the deployed model with sample data and check predictions.
- Using python SDK create a pipeline selecting the best AutoML model and publish it.
Create an experiment using Automated ML, configured a compute cluster, and used that cluster to run the experiment. Below screenshots highlight the steps. The best model was the voting ensemble with an approximate accuracy of 91.9%
We deployed the best model as captured below :
- Enable logs
- We choose the best model for deployment and enable "Authentication" while deploying the model using Azure Container Instance (ACI). The executed code in logs.py enables Application Insights. "Application Insights enabled" is disabled before executing logs.py.
- Use swagger to document anf consume the RESTful web services
Run the endpoint.py file which contains rest endpoint url to interact with the deployed model. Below is the output of the model.
Run the benchmark.shas captured below.
- pipeline scheduled run details
- pipeline created successfully
- pipeline REST endpoint in active state
- Tuning the model parameters to enhance performance. Hyperparameter tuning help to get the best parameters for our model. In future maybe to expand the search space. (https://docs.microsoft.com/en-us/azure/machine-learning/how-to-tune-hyperparameters)
- We could also add some additional steps to tthe pipeline like feature engineering before running the AutoML step. This would have helped in getting better results
Below is the link to the video recording
