AI Platform Overview Google Professional Data Engineer GCP

• The diagram shows high-level overview of ML workflow stages.
• The blue-filled boxes indicate where AI Platform provides managed services and APIs:

AI Platform can manage the following stages in the ML workflow:

• Train an ML model on data:
  • Train model
  • Evaluate model accuracy
  • Tune hyperparameters
• Deploy trained model.
• Send prediction requests to model:
  • Online prediction
  • Batch prediction (for TensorFlow only)
• Monitor the predictions on an ongoing basis.
• Manage models and model versions.

Components of AI Platform

• A Training service:
• allows you to train models using a wide range of different customization options.
• can select
  • many different machine types
  • power training jobs,
  • enable distributed training, use hyperparameter tuning,
  • accelerate with GPUs and TPUs.
• can select different ways to customize training application.
• can submit input data for AI Platform to train

Prediction service:

• allows to serve predictions based on a trained model,
• whether or not the model was trained on AI Platform.

Notebooks:

• enables you to create and manage VM instances
• Instances are pre-packaged with JupyterLab.
• Also has deep learning packages, like TensorFlow and PyTorch
• can configure either CPU-only or GPU-enabled instances
• instances are protected by Google Cloud authentication and authorization
• can easily sync notebook with a GitHub repository.

Data labeling service:

• lets you request human labeling for a dataset to use to train a custom machine learning model.
• You can submit a request to label video, image, or text data.
• provide a representative sample of labeled data, specify all possible labels for dataset, and provide some instructions for how to apply those labels.

Deep learning VM image:

• It lets you choose from a set of Debian 9-based machine images
• optimized for data science and machine learning tasks.
• All come with key ML frameworks and tools pre-installed
• can be used on instances with GPUs

AI Platform Deep Learning Containers

• are a set of Docker containers
• have data science frameworks, libraries, and tools pre-installed.

Deep Learning Containers images include the following:

Frameworks:

• TensorFlow
• TensorFlow 2.0
• PyTorch
• scikit-learn
• R

Python, including these packages:

• numpy
• sklearn
• scipy
• pandas
• nltk
• pillow
• many others

Nvidia packages with the latest Nvidia driver for GPU-enabled instances:

• CUDA 10.0
• CuDNN 7.*
• NCCL 2.*

JupyterLab

Tools to interact with AI Platform

Google Cloud Console:

• deploy models to the cloud and manage models, versions, and jobs on the Cloud Console.
• gives a UI for working with machine learning resources.

The gcloud command-line tool:

• manage models and versions, submit jobs, and accomplish other AI Platform tasks
• recommended for most AI Platform tasks

REST API:

• provides RESTful services for managing jobs, models, and versions, and for making predictions
• use the Google APIs Client Library for Python to access the APIs.