What is Vertex AI and why does a unified platform matter?
1. What is Vertex AI and why does a unified platform matter?
Welcome to the second section of the Machine Learning Operations, or MLOps, Fundamentals course. In this part of the course, you’ll: Be introduced to Vertex AI, Google’s unified AI platform. Be introduced to MLOps on Vertex AI. And explore how Vertex AI helps with the MLOps workflow. If you completed the Machine Learning on Google Cloud course, you’ll recall that building an ML model and taking it to production requires expertise in both the workflow and the products required. This includes tools and services to build, package, deploy, and monitor a model. Let’s recall a few key constructs in machine learning. First, you create datasets by ingesting the data, analyzing the data, and cleaning it up. There are different processes for creating datasets, such as extract, transform, and load or ETL, and extract, load, and transform, or ELT. In the next step called model training, you train a model. This includes experimentation with feature processing, model architecture, and hyperparameter tuning. You also revert or iterate the model when there is new data, when the code changes, or based on a schedule. You then evaluate and compare the model to existing model versions. And finally, you deploy the model and use it for online and batch predictions. Recall that this end-to-end process is called “MLOps,” as defined in the previous section. It’s important to note that the type of data used to train a model will affect the rest of the pipeline. For example, are your data JPEG files or TensorFlow records? Do you store your data in Cloud Storage or in BigQuery? Also, the way you deploy a TensorFlow model is different from how you deploy a PyTorch model. And even the model deployment process might differ based on whether the model is created with AutoML or a custom TensorFlow model. With so many variables to consider when you create a machine learning model, an end-to-end ML platform like Vertex AI brings many benefits. Vertex AI brings together all the Google Cloud services for building ML and AI in one unified platform, which helps enterprises realize more value with their data and accelerate time to value. Vertex AI lets you unify certain parts of an ML workflow: A dataset can be either structured or unstructured. It can have managed metadata, including annotations, and can be stored anywhere on Google Cloud, This currently includes Cloud Storage and BigQuery. A training pipeline consists of steps to train an ML model by using a dataset. The containerization helps with generalization, reproducibility, and auditability. An ML model, which consists of metadata, can be built with a training pipeline or it can be directly loaded from other resources, as long as it is in a compatible format. An endpoint can be invoked by users for online predictions and explanations. It can have one or more models, and one or more versions of those models, with disambiguation carried out based on the request. The main idea is that these endpoints are the same regardless of the dataset type, training pipeline or model. It’s all mix and match. After you create a dataset, you can use it for different models. Please check Vertex AI documentation to learn more about Vertex AI. Because Vertex AI is a unified platform, the user interface can be used to directly manage the following stages in the ML workflow: Creating a dataset and uploading data. Training an ML model on your data which includes: Training the model. Evaluating model accuracy. Tuning hyperparameters, for custom training only. Uploading and storing your model in Vertex AI. Deploying your trained model to an endpoint for serving predictions. Sending prediction requests to your endpoint. Specifying a prediction traffic split in your endpoint. And managing your models and endpoints. Vertex AI is flexible. You can choose either AutoML, which lets you create and train a model with minimal technical effort, or custom training, which lets you create a training application that is optimized for your targeted outcome. In summary, Vertex AI offers fast experimentation, accelerated deployment, and simplified model management to achieve your ML goals. With Vertex AI, you have a platform to experiment with ML model development and quickly deploy solutions that will help meet your machine learning goals. For more information about Vertex AI, explore more courses in the ML training catalog at cloud.google.com/training.2. Let's practice!
Create Your Free Account
or
By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.