19th October 2020 Azure Skane held a meetup in Helsingborg for the first time, I wrote about some of our future plans in this post: Azure Skåne! Past, Present and Future – AzureFabric and this first visit to Helsingborg fits those plans.
This meetup was about Azure AI Parking Space Finder & CI/CD with Github Actions.
Find the recordings!
AI Assisted Parking space finder
Helsingborg stad is pushing some great innovation projects and initiatives in their innovation hub Hbg Works.
One of the PoCs was taking data from multiple resources and predict if there would be parking spaces available.
The problem statement this PoC was working with, states that 30% of traffic congestions is due to vehicles circling to find free parking spaces.
With the problem defined there are some goals also for the solution.
- Reduce Traffic congestion with 50%
- Reduce CO2 emissions
- Increase customer statisfaction
If I was planning a trip to central Helsingborg for some shopping, maybe not in theses Covid-19 times, I would not choose Glasbruket since the solution predicts I have better luck on the spots suggested near the address I put in.
You can try the app and solution here: Hem – Parkeringsprediktion (parkeringsprediktionhbg.azurewebsites.net)
Cool and simple solution setup
It is really cool what you can do with cloud resources at hand! So many options to fulfill your idea!
The solution is basically the below bullets.
- Get data from multiple sources (Azure Data Factory)
- SMHI (weather), Sensors in Parking houses, Event calendars, municipality calendars (schooldays) etc.
- Store it in Azure Data Lake
- Clean and arrange the data and store it in database (SQL DW/Synapse)
- Run ML on it
- Output data for the app.
Getting the data is mostly pulled via APIs using Azure Data Factory.
Basic level SKUs and free usage of services in this PoC is keeping the solution low cost, for example the Web app Free sku gives 60 min compute per day.
Auto ML was used to train the ML model, again no deeper ML knowledge needed. We have had this topic in the Azure Skåne meetups before and you can also check the recording: Solving 3 business problems with ML.NET in 45 minutes
The Machine Learning model is trained every day by an Azure Function that is triggered by the Azure Data factory. The model is stored in Azure Data Lake after each training, making it easy to roll back.
Next steps will be talking to municipalities, talk with AI Sweden etc and in parallel build the smarter city Labs! Helsingborg is an really interesting city, I am looking forward to see where they go with this and what they come up with next.
After a short lunch break and sticker table looting we got to listen to some GitHub actions experiences.
We where joined by Tibi Covaci from Microsoft. He showed us how GitHub actions could be used in software development and showed how Kollokvium | A free multi-party video conference for you and your friends! was setup with GitHub Actions
GitHub Actions is closing on feature parity with Azure DevOps and Tibi walked through the Dev experiences. Messages from Microsoft on the GitHub vs Azure DevOps is.
If you just starting the go GitHub Actions. If you have a big footprint in AzureDevops then keep working with it.
This was yet another night where we get to see small and smart implementations of “low code” AI solutions that gives almost instant value.
The solution almost cost nothing but can realize big values, it was built by a few people and no deep machine learning knowledge was needed. The tools delivers the abbreviation and speeds up everything.
Going forward I think this solution should be put into GitHub actions and that would automate the process beautifully and eventually get the ML Ops started for this project.