TRAINING | April 24, 2019
Last week on April 20th, Data Science Battleground Round 7 was successfully hosted with over 20 participants. Like in other previous rounds of Data Science Battleground, our participants are empowered to acquire fundamental Data Science knowledge as well as topic-related Data Science skills through high-quality lectures prepared by our mentors. What’s more, they can practice all the skills and code they have learnt right after the lecture by teaming up to compete against each other in our special battleground. We will pick the winner based on the final evaluation score. Different topics will have different evaluation style.
Tokyo Techies welcomed over 20 participants to Round 7 of Data Science Battleground
In Data Science Battleground #7, our topic was on stock price prediction using the New York Stock Exchange Market. We kicked off our event, with a warm welcome from our mentor Kien Le, who is a senior software engineer from Microsoft. After the introduction, we went through the concept of sequential data. Time series data, such as stock dataset, are usually complicated and challenging due to many irregular components, cyclic components, trend component and season component.
Mentor Kien Le instructed the fundamental knowledge of Data Science to our participants
However, having prepared carefully a premade code in our jupyter notebook, Phong Nguyen, Head of Tokyo Techies Data Science & AI Department, managed to explain all the technical terms and model so easily that even Yu Cheng – a stock market researcher who has no experience with Data Science – said that she could enjoy learning more about stock by “playing with” the model we have provided. Remarkably, our participants were introduced to a Tokyo Techies signature method for stock prediction, which is called: “Ground Truth”. We use this method to measure the accuracy of the training set, the inaccuracies in the ground truth will correlate to inaccuracies in stock prices prediction.
Mentor Phong Nguyen and his signature stock evaluation method – Ground Truth
After the lecture, we were divided into four teams to compete. Each team spent one hour tweaking the codes to find the best numbers for our three evaluation: (1) MSE, (2) Correlation between ground truth and prediction, and (3) Ratio of direction. All teams had a productive one-hour to improve the premade code, but one team bested them all with 0.27 correlation score. What they did is adding a bidirectional LSTM which helps run your inputs in two ways and increase the accuracy. Along with the atmosphere, we congratulated the winning team with vouchers taking two free classes with Tokyo Techies.
Four teams had an exciting session in which they competed with each other to improve the premade code and predict stock
In the end, we had a wonderful networking session. Everyone was surprised with each other background. Some are investment bankers. One is working as a space engineer. Some are already machine learning engineers. If you were in that room, you could feel the passion for knowledge from every participant!
If you are yearning for more knowledge and new friends, let’s join the Data Science Battleground round 8: MovieLens – Recommender System on May 11th, 2019. Let’s learn, compete and make friends!
Want to know more about our AI services?