Hacking Machine learning in Microsoft OpenHack 2018, Dubai

Microsoft has organized an OpenHack event in Dubai for the first time in the region. I was glad to participate in this event and thought of sharing my experience here.

We started our day knowing nothing about what is coming up. I had the chance to network with amazing, like-minded persons from different organization and that was the first Benefit from this event. At that stage the common feeling among us was wondering what will we do in this event?

Well, The event started and the speaker on stage welcomed us warmly and mentioned that Microsoft is doing this kind of events for the first time in the region, and he explained that there will no one on stage to lecture anything, today is all about learn by doing. Also in a brief he explained the agenda of the day as there will be some 101 help optional sessions to explain some topics like how machine learning working, python basics, deep learning, …etc

So the teams were formed cleverly based on the background of each member to have the maximum diversity. On each table we have a coach who have deep knowledge about the technologies we will use in the challenges. However they will not give you the solution to any challenge because they were instructed not to do so, and simply because there is no single solution to those challenges. However the coach role was very important in this event to provide the guidance and give the answers to the technical challenges in using python or azure machine learning workspace. 

The challenges were to help an organization specialized in outdoor and climbing gear retail to understand their customer behavior through machine learning. 

As the first speaker warned us, the first two challenges were frustrating as we had to jump into python code directly without previous experience in writing python code (at least for me). We had to start doing data preparation for some datasets of images that the organization has in their database. This was frustrating since it took us a lot of time to do basic things in python like for loops while most of us are senior programmers with more than 10 years of doing code. 

However, this itself was the second major benefit we got from this event. We were out of our comfort zone to learn new language and use it immediately. For that we had to search the internet and read a lot of documentations. 

Although preparing the data is not fun, data scientists spend 80% of their time there because of its importance in any model you want to build. Small mistakes in the data preparation might lead to totally wrong recommendations by the AI models. 

So the challenges got more exciting as after we have prepared the datasets we were able to develop machine learning model using scikit-learn library where we had to train the model based on the input and output datasets. Then we have gone deep learning by developing a model using Keras library without the need to human intervention to tell the model what the output should look like.

After creating and training the model, and when we were satisfied with the results, it was time to deploy the model to be used by any other application. And that’s what we have done using Microsoft Azure machine learning workbench on the cloud. the workbench was super easy and efficient to use.

Finally, we have used custom vision to platform to train a model to analyze the images and identify parts of the image. This information was so powerful and easy to use to drive the marketing and safety strategy for the (virtual) organization we are trying to help. Then it was super easy to expose that model to an API and consume it in any application. 

In conclusion, in this unique learning experience, we were able to hack a lot of new topics and successfully complete a lot of the challenges. Although it is not the easiest way of learning, but definitely it is an efficient way of learning that worth trying in our organizations.