From the course: Automated ML.NET Training, Metrics, and Accuracy

Demo: AutoML

- [Instructor] Now let's further drill down into AutoML by utilizing a demo. Looking in my solution explorer, you'll see that I have imported the NuGet package, Microsoft.ml.AutoML, and this allows me to use the AutoML within my project. As with most machine learning projects, I instantiate my ML context and then I specify the data path. And let's discuss the three major steps that we talked about in the previous video. First, we're going to infer the column information, and we're going to be using columnInference, which will result in the columnInference results. And here we're inferring the various columns from that data set. And more specifically, we're identifying which one will be our label. In this case it will be the fare amount. Then we'll go ahead and load the data. And also, as mentioned, we're going to be using the train test split method. And here as specified, we're going to be utilizing 20% as our test data and the remaining 80% will be used for training. The second major step will be defining the data pipeline. And here we're going to be utilizing the Featurizer method as we had discussed. And if you'll notice, we are accessing the machine learning context object, but more specifically, we're using the auto method within it to access the AutoML features. And within that is the Featurizer method where we specify the data set that we had loaded previously, as well as all the other column information that goes along with it that's previously defined. The third step is creating the experiment. So you'll see we're using the AutoMLExperiment data type, and we are creating the experiment using the CreateExperiment method. We then proceed by configuring the experiment, by setting the pipeline which was previously defined. We're also going to be looking at the training time in seconds and the training of the data set that we had set previously. In addition, there's another step that we're going to be utilizing regression metrics. We are then continuing to log the results of each experiment and displaying that. And when everything is said and done and we're ready to run our experiment, we utilize the run method within our experiment object. Once the run has completed, we then save the model, because at this stage it has produced the best available model given the dataset, and now we can just simply save that model utilizing the save method. That concludes the demo.

Contents