Join Jonathan Fernandes for an in-depth discussion in this video Machine learning overview, part of AWS Machine Learning by Example.
- [Narrator] So, let's look at machine learning. Machine learning can use historical data to make better future predictions. Machine learning algorithms, or models that we use, determine patterns in the data. So with binary classification, we will predict only two outcomes. With a multiclass classification, predictions can be in multiple classes or bins. And finally in regression, we're looking to predict a numerical value. So, when should we be using machine learning? There are many case when you can determine the predicted value by applying a formula or a series of steps.
Not every problem can be solved by machine learning, contrary to what you see in the news. There are instances when we will want to use machine learning. This is, for example, when the rules depend on several factors and many of these rules can overlap or need to be tuned very finely. If you need to scale your solution from a couple of hundred to several million, then machine learning solutions are effective at handling large-scale problems. One of the key goals of machine learning modeling is to ensure that your model can generalize beyond the data that it was trained on.
A model that can make accurate predictions only on data it was trained on is not very useful. What is common in the machine learning community is to take 70% of your data and use that to train your model. You then use the remaining 30% to evaluate your model. Finally, if you want to see how well your model can generalize, you can then use test data. And this is data that has not been used in training your model. In the next video, we will look at learning algorithms and hypoparameters.
- Learning algorithms and hyperparameters
- Preparing data for AWS
- Using binary, multiclass, and regression techniques
- Creating a datasource
- Generating predictions
- Creating and interpreting batch predictions
- Additional AWS capabilities