MEST Accra — April 2019
After a Machine Learning focused meetup sometime around July of 2018, I got a message from my friend John, who co-leads Developer Circles Accra. In the message, he mentioned that someone from the school he finished from (Meltwater Entrepreneurial School of Technology) was interested in speaking to me.
A few days later, I was speaking with Felix, who was interested in getting me to visit for a few days so I could teach some of the students (Entrepreneurs-in-Training, EiT) about Machine Learning. That way, they would have a much better foundation for building their startups.
We scheduled my session for two weeks in April, but I was unfortunately away for the first week which I used to attend Google Cloud Next ’19, and to take my certification examination for Professional Data Engineer.
Because the EiT at MEST is trained to build businesses, it was important for me that whatever they learn could be applied immediately. I prepared a curriculum that would teach them not only TensorFlow Estimators but also ML APIs on GCP, as well as Dialogflow and Firebase ML-Kit. I had grand plans.
One week to the start of my session, it occurred to me that some individuals might have difficulty signing up for accounts on the Google Cloud Platform. I sent a message to Google Developer Relations asking for Qwiklabs credits for the EiTs, and they gracefully provided those in the form of one month’s access.
On Saturday before my session was due to start, I got on a remote session with the EiTs, did an introduction to Qwiklabs and GCP, and left them with a quest to familiarize themselves with the platform.
I arrived Accra on the second week of my session and started teaching TensorFlow Estimators and Serving. By the third day, we went into some of the math and theory of Machine Learning. Afterwards, we did a lab on training and deploying using Cloud Machine Learning Engine, as well as a second lab on hyperparameter tuning. We then proceeded to look at ML APIs, and did a lab on AutoML Vision, in which they learnt to train a model to identify clouds.
On the fifth day, we attempted a lab on Dialogflow, but sum oAuth issues derailed us. We then pivoted and did a lab on Cloud IoT Core, which involved simulating IoT data, streaming into Pub/Sub, using Dataflow to write out the data to files on GCS, and finally using Dataprep to transform the files. This exercise served to teach the basics of data pipelines, and they got a better idea of how they might collect and process data for training Machine Learning models.
In the end, we did not cover everything that I would have wanted us to cover, but a lot of their questions were answered, and I left them with some course codes that they can use to take a specialization on Coursera.
This was a fulfilling exercise, and I hope the students had as much fun learning, as I did teaching.
