Here we're looking at the main interface for Oracle machine learning on autonomous database. We're going to select the quick action for AutoML to take us to the OML AutoML UI. This is where we can see the list of experiments we've worked on previously. We can create new experiments, edit existing ones, delete, duplicate, as well as start and stop experiments. Let's create a new experiment. We're going to call this, "Customer360" and we'll access the table from the "OMLUSER Schema." We're logged in as all OMLUSER03. We'll select the, "Customers360," table. We'd like to predict whether a customer should get an Affinity_Card as part of our campaign. The unique identifier will be the Customer ID. Because we have one record per case. The interface displays the features, and notes that our Affinity_Card is our target, can also see some statistics that are computed for the various features. From there, we can also specify additional settings, how many models we actually wish to have reported, so we'll reduce that to three, and specify a maximum run duration depending on the modeling that you have in mind. You can also change the database service level, and the model metric that you'd like to use for selecting the top models. From there, we can identify which algorithms we'd like to include. We can start the experiment and request either faster results or better accuracy. We're going to go for faster results. With the experiments starting, we'll bring up the, "Progress bar." We'll see that this phases, we're going to move through our algorithms selection, adaptive sampling, feature selection, model tuning, and feature prediction impact. The set of algorithms that have been selected include random forests and two support vector machine variants, one with the linear kernel and the other with Gaussian. We're now in the adaptive sampling stage. Now we move on to feature selection to identify which columns will be most useful for the various algorithms. Now we're onto model tuning. We start with random forest and we see the initial accuracy estimate of 0.74 showing up on our Leader Board, and as successive models are attempted, we see that the accuracy is improving. Next we move on to support vector machine with the linear kernel. Now we move on to support vector machine with the Gaussian kernel. Notice that the new accuracy for support vector machine Gaussian is showing even higher accuracy at this point. Also notice under the accuracy title, that we're seeing a progression of how the quality of the model has improved over time. The last part is feature prediction impact, which uses the model explainability to identify which features have the most impact on predictions. With this completed, we can look at specific models and see the details for the prediction impacts for that model. The svmgaussian has household_size and education and occupation as the top predictors and we also have a computer confusion matrix indicating how well the model performed and what types of errors were made. Selecting the top model, we can deploy our model. We'll give it a name that we'll recognize Customer360SVM, and the URI, a version, and a namespace. I'll say that we'd like to share this model with others. Now this is deployed to AutoML Services. We can go into the models listing and we see these are all the models that are available to us. But if we go to deployments, we see that we have our Customer360SVM model. It's shared the version, the namespace. If we click on the "Model name," will see the model metadata, and on the "URI," will see the open API specification. This model is now ready for use with rest endpoints. Let's go back to our AutoML experiments listing. Go back into Customer360. We can also just select a model and create a notebook for it. We're going to create a notebook called Customer360SVM, and if we go into our notebooks listing, we'll see our Customer360SVM notebook. We start up the notebook server and then load the notebook. We see that the code that has been generated is using the OML for Py interface. These are all Python paragraphs and they allow us to prepare the data, create the training data necessary for building the model. Here we see the exact settings that were used by AutoML UI to build our SVM_model. We create the SVM object and fit the model with the training data. From there we can see model details and generate predictions. We can also show the accuracy for the model. This concludes our demo, Oracle machine learning AutoML user interface.[NOISE]
