Looking back on learning with Azure Machine Learning Studio
Introduction
This is a review of using Azure Machine Learning Studio to study Udemy's machine learning-related courses that were held from the end of last year to the spring of this year.
Skill level
- Python beginner Udemy first learned about "jupyter notebook"
- Linux beginners who know a little command
Why did you decide to do it with Azure Machine Learning?
- My PC is too weak to use GPU.
- At Ignite in December 2019, I was interested in the story of Azure Machine Learning.
- Azure had a usable environment.
- I didn't know about other services.
how to use
- Log in below.
https://ml.azure.com/
- First create a VM to run from compute.
The created machine will be automatically started.
To save money, stop the VM when it is not in use.
If you want to use the GPU, change your computer resources here.
At that time, I had to select CPU/GPU etc. from the screen to create a normal AZURE virtual machine, but now it is easier to select for the following screens.
- In Notebooks, create a new one and select from the created computing resources to use.
What I had a hard time at that time
- The version of Tensorflow of learning content was 1.14, but the environment on Azure was 2.0.
→ Open the terminal from "Open Terminal", and with unreliable knowledge
I was able to install Tensorflow
pip install tensorflow==1.14
→ Command execution was also possible from that terminal.
- The operation of the service itself was unstable.
Compute's Start and Stop often failed due to an error.
→ I was logging in again or executing after a while.
- When I tried to edit with Jupyter in a notebook, it became 504 Geteway Time-out and could not be used.
→ Log out from AzuleML and discard all logins with multiple accounts. The browser also restarts.
Furthermore, when I restarted the created virtual machine, it remained restarting, so I also recreated the virtual machine.
- The size of the VM that can be created has been changed in the settings on the Azure side, and the size I want to create has disappeared.
→ There is no choice but to use a different size.
- When I was creating a VM with various conditions, I couldn't create a VM.
→ There was an error in the upper limit of the total number of cores of VMs that can be created. Removed infrequently used VMs as it was 20 per region.
What impressed me at that time and what was good.
- I was able to use the GPU, and I was able to realize the power of the GPU.
With 60,000 test data, when I tried CNN, it took 120 seconds for the CPU and 8 seconds for the GPU.
- It's natural because it's a cloud environment, but I was able to execute it without worrying about the location or the PC used.
I could use it on my office PC or home PC as long as I could connect to the Internet and access ml.azure. It was a very thankful environment at that time.
- It was convenient because various Python programs that I learned and created can also be saved on Azure Machine Learning.
in conclusion
Since spring, I have taken the coursera's "Machine learning" course for machine learning, but Azure Machine Learning has not been available since spring. You can also use Azure ML
I hope I can proceed. In the future, I would like to implement ML using the data of the business DB.
(Although the story is completely different, the above couesera course was very meaningful.)
- We would appreciate it if you could point out any errors or inappropriate wording in the description.
Thank you very much.