Path1.Mod2.a - Explore Workspace Developer Tools - ML Studio Flashcards
The three tools for Azure ML and which to use…
- ML Studio
- Python SDK
- Azure CLI
ALL are a personal preference…so use whatever
Ideal use cases for ML Studio
- When you want to use a no-code approach
- Quickly review your work and its results
- Submit jobs and manage models from a Jupyter notebook
- Ideal for data scientists.
Ideal use cases for Azure ML Python SDK
- Prefer the Python code approach and automating repetative work
- Submit jobs and manage models from a Jupyter notebook
- Ideal for data scientists
Ideal use cases for Azure CLI
- Prefer the code approach and automating repetative work (uses the Azure Machine Learning extension)
- Ideal for infrastructure automation
When to use the Azure ML v2 and some of its features
Whenever you’re starting a new Machine Learning project or workflow!!
V2 has new features:
* Managed Inferencing
* Reusable pipeline components
* Improved pipeline scheduling
* Responsible AI Dashboard
* Assets Registry
Azure ML v2-created Workspaces cannot reuse Azure ML v1-created entities (Workspaces, Compute, Models, Environments) due to incompatibility (T/F)
False. v2 can use any of them.
Au-j As-a Co-c
The three main Menus (left-hand side) in Azure ML Studio and what they consist of wrt Models
- Authoring - Create new jobs to train and track an ML Model
- Assets - Create and review assets used for training models
- Manage - Create and manage compute resources needed to train Models
D AML
The two Authoring tools for creating a new Job in the Studio
- Designer - Drag n Drop interface for creating Pipelines with prebuilt (custom) Components
- Auto Machine Learning (AutoML) - Wizard interface that lets you train a model using a combination of algorithms and data preproessing techniques to find the best Model for your data.
Notebook Kernels
Type of Compute ideal for development and why …
Compute Instances are ideal for dev because they are more scalable and cost efficient than local training
Type of Compute ideal for training models and why…
Compute Clusters are ideal for training because the Cluster will dynamically resize with its number of nodes in order to run a training job, then go back to zero nodes once the Cluster isn’t needed anymore