Juni 8, 2018 3 min to read
Chuck Norris Roundhouse Kick AKA Deep Learning on Mallorca
Category : Artificial Intelligence, Data Science, Deep Learning, Teaching
Namaste! I just came back from the beautiful mediterranean island of Mallorca. Business-trip. Vacation will be soon. Part of my Deep Learning activities is educating people about Artificial Intelligence, its tools and its applications. I got invited by Mayflower to hold a workshop. This astonishing company from Würzburg did its yearly barcamp in Spain. What was it all about? Reflecting on their tools and knowledge. Exchanging ideas and doing brainstorming. And of course learning about the hottest topics. Time for Deep Learning on Mallorca!
Learning by kaggl’ing.
Kaggle is awesome! For those of you who do not know it yet: Kaggle is an excellent platform for competitions. Especially for Machine Learning and Deep Learning. It is positively overflowing with datasets that you can use for practice. And hey! You can even win quite some money if you manage to be among the top-3.
If you want to get started with Deep Learning, go to Kaggle. With you want to get into a data domain that is new to you, got to Kaggle. If you want to compete with others, while learning a lot, go to Kaggle. I guess you get me.
For our workshop we decided on a deep dive. I gave a straight-to-business introduction to Deep Learning. MNIST with Keras in no-time. Clearing the Artificial Intelligence rumor house. Showing what technology is now available. Deep Learning on Mallorca made it clear that it is kinda easy to get started. Even with limited resources. Such as bad tourist-WiFi 😀 And after that, right before noon, we started with a Kaggle competition.
The Chuck Norris Roundhouse Kick of Machine Learning
For the team, the workshop was a Chuck Norris roundhouse kick (registered trademark?). The outstanding developers of Mayflower had a chance to refresh their Python skills. What did they learn? You can rule the place with Jupyter-notebooks. Matplotlib is a powertool when it comes to visualizations. Numpy is your weapon of choice for highly efficient data manipulations. And Pandas is perfect when it comes to named dictionaries, also known as table data.
Having all tools at hand it was easy to get a first grip on Artificial Neural Networks. Specifying Neural Nets with Keras is impressive. But only in the sense that it is both elegant and straightforward.
The bottom line was simple… In order to have a good start with Deep Learning you do not have to read all those huge books. Learn a handful of tools and principles and you are off to success. After that you can mindfully strengthen your skills by digging deeper.
Home Credit Default Risk. A very nice challenge.
The Home Credit Default Risk challenge is a classic classification problem. You get a huge relational database with informations about customers. Financial data, historical data and beyond. The challenge is about coming up with a system that predicts credit-worthiness with a high accuracy. As you are reading this blog, the top-score is 80,1%. There is still some space for improvement!
For me it is always important to get a quick and not-so-dirty prototype up and running. Establish the pipeline first and then expand upon it. In our case, this meant getting a rudimentary Neural Net to train on some data from the dataset. Aiming high but starting low. It is not necessary to go for a great and complex solution in the beginning. Start small. Train something small and learn from it. And take your learnings and come up with a next prototype. Repeat until keypressed.
I admit I am a little jet-lagged. Traveling was a little longer than expected. Nevertheless I feel very, very accomplished. Again an highly motivated group knows how to sing the Deep Learning song. I love teaching. Especially if it is Deep Learning on Mallorca. Thanks Mayflower for making this possible!
The feedback that I got was very positive. This honors me. Those eight hours passed by like a Imperial Star Destroyer on lightspeed.
And a final thing… I guess our friends at IBM found something nice about Deep Learning and relational databases!
Stay in touch.
I hope you liked the article. Why not stay in touch? You will find me at LinkedIn, XING and Facebook. Please add me if you like and feel free to like, comment and share my humble contributions to the world of AI. Thank you! I am looking forward to talking to you!