Creating an AutoML pipeline
We’ve now seen the two extremes of setting up a model training run – on the one hand, we could rely on AutoML experiments to manage the details of training for us and make the best guesses about our data and intentions. Doing this gives up control and may result in needing more time to find an optimal model.
At the other extreme, we can train a model manually, where we must select the exact model trainer and optimal hyperparameters – plus set up any needed transforms to get our data ready for that trainer.
Between these two options, we have another form of AutoML : AutoML with pipelines. By using pipelines like a manual model training process, we retain control over how our data is transformed and interpreted, but we still get the use of AutoML’s power for finding the optimal model trainer and hyperparameters.
AutoML pipelines are more code than the AutoML experiments we’ve seen previously, but they also give you...
