Using PyTorch for High-Performance Machine Learning with GPU on Azure
Question
During your machine learning experiments, you need to use the PyTorch framework for training models and you need GPU for high performance.
One of your colleagues comes up with the following code, telling you that it could save you manual work:
# connect to workspace from azureml.core import Workspace, Environment ws = Workspace.from_config() ... # set environment my_env = Environment.get(workspace=ws,name="AzureML-PyTorch-1.1-GPU") ...Does it really make your life easier or not?
Answers
Explanations
Click on the arrows to vote for the correct answer
A. B.Answer: A.
Option A is CORRECT because, besides enabling you to define your own run environments from scratch, Azure ML comes with a bunch of pre-built (“curated”) environments for typical ML scenarios.
You can use them easily, by simply using them in your “Environment” definition.
"AzureML-PyTorch-1.1-GPU" is one of them.
Option B is incorrect because Azure provides a lot of pre-defined environments for typical ML scenarios.
If your requirements fit for these “curated” specifications, you can readily use them, instead of specifying manually.
Reference:
The code provided by the colleague helps to connect to an Azure Machine Learning Workspace and set an environment for PyTorch version 1.1 with GPU support. It seems to simplify the task of setting up the environment and reduces manual work.
However, to determine whether the code actually makes life easier, we need to consider a few things:
Is the PyTorch framework appropriate for the machine learning experiments? If PyTorch is not the appropriate framework for the experiments, then using the provided code would not be helpful.
Is GPU necessary for high performance in the experiments? If the machine learning experiments do not require the use of GPU for high performance, then using the provided code would not be helpful.
Is the PyTorch environment provided by the code suitable for the experiments? If the PyTorch environment provided by the code does not meet the requirements of the experiments, then using the provided code would not be helpful.
Assuming that PyTorch is the appropriate framework, GPU is necessary for high performance, and the PyTorch environment provided by the code meets the requirements of the experiments, then the code does make life easier by simplifying the task of setting up the environment.
Therefore, the answer is A. Yes, provided that the PyTorch framework is appropriate, GPU is necessary for high performance, and the PyTorch environment provided by the code meets the requirements of the experiments.
