Describing Virtual Machines in Azure
-
Azure offers its own implementation of virtual machines
- These virtual machines are hosted by Azure
- This service is called an Azure Data Science Virtual Machine
-
Azure Data Science Virtual Machines are virtual machines with pre-installed data science packages
- It comes with Python and R
- And Python and R specific packages
- We can build models using Python and R code in our local IDE on the virtual machine
-
Azure Data Science Virtual Machines are used for:
- Performing analysis with visualizations
- Building ad-hoc models
- Registering models
- etc.
Describing Data Science Images in Azure
- We can see our built images in the Azure Machine Learning service
- We can also see any running containers based on those images
-
This image can consist of the following components:
- A model represented as a pickle file
-
A scoring script
- The scoring script refers to score.py
- The scoring script is responsible for consuming the model
-
The scoring script only has two functions:
- An init function, which loads the model
- A run function, which does the inference (i.e. model.predict(data))
-
An environment file represented as a YAML file
-
The environment file declares the dependencies for:
- The model
- Scoring script
- Application
-
- For example, we could specify numpy or scikit-learn in our YAML file
Details about Data Science Images
- Once a model has been trained and registered, an image will be built for deployment
- We then deploy our model by running a container (Azure container, Docker container, etc.) based on this image
- Specifically, we refer to these running containers as deployed web services, if the container is a AKS or FPGA container
- On the other hand, we refer to these running containers as an IoT module, if the container is a Docker container
- Then, we can use Python to access the deployed model in our container
References
Previous
Next