Machine learning workflows are complex. Building each part of the machine learning workflow can overwhelm developers and cause failure. Learn how to build, train, and deploy machine learning with AWS SageMaker avoiding the pitfalls of unmanaged solutions.
- [Instructor] AWS SageMaker is a the root of all machine learning, or ML solutions, deployed through AWS. SageMaker equips each piece of the stack of requirements for a machine learning solution. All machine learning solutions require articulated goals, data, algorithms that set the process of how to learn, and the deployment of the results of the learning. SageMaker supplies each of these pieces of the puzzle. Their supply pieces are also available for adjustment if the solution provider needs a different learning set or organizational goal. Let's jump to SageMaker and see how it puts it all together for rapid deployment of machine learning projects. Back in the AWS management console, I will type SageMaker. There it is, select that, and hit enter. So we haven't built any training models yet, but you will be presented with, on the left-hand column, the pieces required to get to a trained and deployed machine learning model. Let's go to the dashboard and see how this looks. The overview gives us each piece of a machine learning training model. It gives us the labeling so that we can decide what the data means. The notebooks, inside of which are SDKs, algorithms, and labeling occur. The training, in which our machine learning algorithms actually do the job with the data. And the inference, in which it is the output of the answer, or the result, or function of the machine learning algorithm. Further down the page, we have recent activity, which we have none. Options to learn and to understand SageMaker better. Let's go over to labeling jobs in the left-hand column. Let's create a labeling job, understanding that this is our ability to know the dataset and its purpose as we continue. The name of the job is example learning. I want to check I want to specify a label attribute name different than the labeling job name. The attribute name is going to be example learning myself. This is where I could put a dataset that I already had if I had an S3 bucket that had my list of attributes. When I was born, my height, my wife's name, my address. But I'll be listed in that S3 bucket. And once the algorithm ran across that data, I could define where that was located as a result. I can set up the task type, these are wonderful examples out of the box for SageMaker, and apply tags. 'Cause we're not actually going to run an ML model, I'm going to cancel this and go back to the dashboard. For an example of notebook instances, let's select notebook instances and create a notebook instance. We'll give the instance name example notebook. Let the instance type be a medium machine learning T2 environment. Scroll down. Notice that it wants us to give us and then enable us root user access to the notebook. That is why I am in the root user. We can then decide that the notebook exists on the network, or is located in the get repository. Those are optional settings. Having a review of the notebook, let's go back to the SageMaker homepage. Training algorithms. I'm not going to set up an algorithm, but this is a fantastic view of finding an algorithm to let you learn how to create an ML result. Finally, in the inference, compiling jobs, model packages, and models. Those will be the output of the machine learning algorithm. So these will be in the functions and end results of your machine learning process. I'm going to select AWS marketplace for the final piece of this. Depending on what your defined goals are, there might be a requirement of hardware that is already associated with AWS infrastructure. The marketplace allows you where to find that hardware and deliver it to an AWS machine learning library. Now let's jump over to AWS Deep Learning and understand what else we have to offer.
- Comparing AWS to previous IT solutions
- Benefits of AWS
- AWS security and pricing
- Planning big data solutions on AWS
- Cloud storage and databases
- Data warehousing
- Backing up data
- AWS for mobile services
- IoT approaches
- Augmented reality (AR) and virtual reality (VR) on AWS
- Planning AWS blockchain solutions
- Deep learning and robotics on AWS