Vertex AI Tutorial: Unlocking the Power of Machine Learning

Vertex AI Tutorial: Unlocking the Power of Machine Learning: Welcome to the world of Vertex AI, where machine learning transcends boundaries, and innovation knows no limits. In this article, we’ll delve into the intricacies of Google’s Vertex AI, a powerful platform that will equip you with the skills to master machine learning effortlessly. Buckle up as we explore the diverse facets of this platform, from seamless data integration to robust model deployment, and everything in between.

Embracing Vertex AI: Your Gateway to Machine Learning Excellence

Vertex AI is your all-encompassing portal to the realm of machine learning. It empowers you to effortlessly train, deploy, and tailor machine learning models and AI applications. With Vertex AI, you can unlock the full potential of large language models (LLMs) and infuse AI into your projects with ease. This platform seamlessly brings together data engineering, data science, and ML engineering workflows, promoting collaboration among your teams. By providing a shared toolset, Vertex AI enables you to scale your applications, harnessing the unparalleled capabilities of Google Cloud.

Exploring Vertex AI’s Multifaceted Features

Let’s take a closer look at the plethora of options that Vertex AI offers for model training and deployment:

1. AutoML Brilliance: Effortless Training at Your Fingertips

AutoML in Vertex AI revolutionizes training by eliminating the need for coding. Whether you’re dealing with tabular, image, text, or video data, AutoML simplifies the process, bypassing the complexities of data preparation. It’s the perfect choice for those new to AI or individuals with limited coding experience.

2. Custom Mastery: Unleash Your Creativity

Enter the world of custom training, where you have absolute control. Your preferred ML framework shapes the journey, your code guides the process, and hyperparameter tuning options become your compass. With Vertex AI, customization knows no bounds.

3. Model Garden Exploration: Your Innovation Oasis

The Model Garden is an inviting oasis of discovery, testing, customization, and deployment. Here, Vertex AI seamlessly integrates with open-source models and assets, offering a rich playground for innovation. Explore, experiment, and evolve your models effortlessly.

4. Generative AI Horizon: Unleash Your Creativity

The horizon widens with Generative AI, granting access to Google’s expansive generative AI models. Text, code, images, and speech become your canvas, ready to be tuned and deployed within your AI-powered realm. Get ready to embark on a journey of creative AI exploration.

Mastering Vertex AI: A Step-by-Step Tutorial

Now that we’ve unveiled the beauty of Vertex AI let’s dive into a step-by-step tutorial to get you started on your machine learning journey.

Getting Started: Prerequisites

To successfully complete this tutorial, ensure you have an active Google Cloud subscription and the Google Cloud SDK installed on your workstation. These tools will be your companions as you venture into the exciting world of AI and machine learning.

Step 1: Dataset Creation

Creating a dataset is the foundation of building an effective AI model. In this step, we’ll focus on uploading images and labeling them for training purposes. With Google Vertex AI AutoML, the training process requires minimal intervention, making it ideal for those new to AI or with limited coding experience.

For this tutorial, we’ll use a dataset of images containing faces with and without masks, created by Prajna Bhandary. She employed image augmentation techniques to generate over 600 images for each class, providing sufficient data for AutoML to work its magic.

Uploading Images to Google Cloud Storage

Begin by creating a Google Cloud Storage bucket to store your dataset. Use the following commands to initiate the bucket creation process:

gsutil mb -l $REGION -c STANDARD gs://$BUCKET

Remember to adjust the bucket name and region as needed. Note that at the time of writing, Vertex AI AutoML is available in the US-CENTRAL1 (Iowa) and EUROPE-WEST4 (Netherlands) regions.

Next, clone the GitHub repository containing the dataset to your local machine:

git clone

Navigate to the data directory and run the following commands to upload images to the bucket:

gsutil cp -r with_mask gs://$BUCKET
gsutil cp -r without_mask gs://$BUCKET

Ensure you execute these commands in separate terminal windows to upload images from both directories simultaneously.

Step 2: Generating a CSV File

To create a CSV file containing the path and label of each image, we’ll use a simple BASH script. Run the following commands to generate entries for images with masks:

for filename in with_mask/.jpg; do
  [ -e "$filename" ] || continue
  echo "gs://$BUCKET/$filename,mask" >> mask-ds.csv

Repeat the process for images without masks:

for filename in without_mask/.jpg; do
  [ -e "$filename" ] || continue
  echo "gs://$BUCKET/$filename,no-mask" >> mask-ds.csv

Upload the generated CSV file to the bucket:

gsutil cp mask-ds.csv gs://$BUCKET

Step 3: Creating a Vertex AI Dataset

With your data ready, it’s time to create a Vertex AI dataset. Follow these steps:

  1. Access the Vertex AI Dashboard in the Google Cloud Console and enable the API.
  2. Choose your desired region and click on “Create Dataset.”
  3. Provide a name for the dataset and select “Image Classification” with a single label.
  4. Import files from Cloud Storage by selecting the CSV file you uploaded earlier.

The import process will take a few minutes. Once completed, you’ll see the labeled and unlabeled images in the dataset.

Leveraging Vertex AI’s Flexibility

Vertex AI provides a spectrum of tools, including Python’s SDK, to empower your machine learning journey. Whether you prefer the Vertex AI Workbench, Google Cloud Console, or the gcloud command line, Vertex AI offers diverse interfaces for maximum adaptability.

How Vertex AI Fuels Model Construction and Deployment

In the context of model creation and deployment, let’s explore the machine learning workflow facilitated by Vertex AI:

1. Data Preparation

Commence with data extraction and refinement, conducting exploratory data analysis (EDA) to understand the anticipated data schema and required traits for the ML model. Employ data alterations and feature engineering, and segment the data into training, validation, and test sets. Explore and visualize data seamlessly with Vertex AI Workbench notebooks, where integration with Cloud Storage and BigQuery accelerates data access and processing. For larger datasets, streamline processing with Dataproc Serverless Spark through Vertex AI Workbench, eliminating the need for managing individual Dataproc clusters.

2. Model Training

Delve into training methodologies, choosing the one suited for your model’s training needs. For those averse to coding, the AutoML overview offers a code-free model training path. AutoML supports diverse data types like tabular, image, text, and video. Alternatively, embrace custom training by crafting your training code within your favored ML framework. Tailor your models by optimizing hyperparameters through custom tuning jobs or leverage Vertex AI Vizier’s expertise in tuning complex ML models. Exploit Vertex AI Experiments to experiment with distinct ML techniques and gauge their outcomes.

3. Model Evaluation and Iteration

Shift your focus to model evaluation, refining your model based on evaluation metrics and iterative enhancements. Utilize metrics like precision and recall to assess model performance, leveraging Vertex AI Model Registry for versioning and production transitions. Whether through Vertex AI Model Registry or Vertex AI Pipelines workflow, evaluations can be seamlessly integrated.

4. Model Serving

Transition to model deployment with production-ready predictions. Deploy custom-trained models using prebuilt or customized containers, offering real-time online predictions. Facilitate asynchronous batch predictions without endpoint deployment. The optimized TensorFlow runtime further reduces latency and cost, outperforming open-source TensorFlow Serving containers. In scenarios involving tabular models, the Vertex AI Feature Store serves as a repository, facilitating feature provisioning and monitoring feature health. Vertex Explainable AI aids comprehension by attributing features to model predictions, while also identifying mislabeled training dataset instances.

5. Model Monitoring

With your deployed model in operation, vigilantly monitor its performance. Leverage Vertex AI Model Monitoring to scrutinize training-serving skew and prediction drift, promptly notifying you when incoming prediction data deviates significantly from the established baseline.


In summary, Vertex AI offers a diverse array of tutorials covering various AI workflows, encompassing tasks like creating datasets from tabular data, training classification models using AutoML, and deploying models for online predictions. To kickstart your Vertex AI journey, you can establish a project, set up a development environment, install the Vertex AI SDK for Python, and explore AI models and APIs. Moreover, you can build custom ML models with Vertex AI through step-by-step instructions in YouTube videos or create tailored Vertex AI pipelines for MLOps using a comprehensive tutorial from Towards Data Science.

Overall, Vertex AI serves as an integrated platform merging data engineering, data science, and ML engineering, fostering collaboration among teams through shared tools and harnessing the advantages of Google Cloud’s capabilities. It’s time to unlock the true potential of machine learning with Vertex AI.

Leave a Comment