Smart Bots 2023: Journey to Super 8
Smart Bots 2023, the highly anticipated four-month-long tech tournament, has finally come to an end of its Super 8 phase. The journey till now has been a great success, with many participants learning new things and pushing their limits in the competition. We saw an influx of tech enthusiasts and students who were constantly improving and tweaking their codes.
This year’s Smart Bots Coding Challenge is being covered by various tech-news partners, TechSaathi, TechLekh, ICT Samachar, ICT Frame and GadgetByte. A huge thanks to all for being part of the Smart Bots Family.
Alongwith Shequal Foundation, we visited almost all engineering and IT colleges of Kathmandu, to tell students about the Smart Bots as an opportunity for them to excel and explore coding dynamics. However, it was also quite a learning experience for us too.
- We observed that many students were unsure of what a "bot" was, with half of them envisioning the construction of mechanical bots.
- The word "Bot" alone led them to assume that they couldn’t do it themselves.
A "Bot," on the other hand, is merely a simple application with a set of functions. It gets certain inputs and returns some value (outputs) as per rules.
Nevertheless, we had a great time interacting with the students from different colleges and learning about their views regarding the competition.
The qualifier phase kicked off with a series of learning sessions, where participants got the chance to brush up on their skills. These sessions were led by some of the industry's leading experts, who shared their wealth of knowledge and experience with the participants. The participants were able to gain a deeper understanding of various algorithms as well as get hands-on experience building a Tic-Tac-Toe game from scratch.
Monte Carlo Tree Search Algorithm by Ranjan Shrestha
On a Saturday afternoon, with a houseful of around 70 tech enthusiasts at King’s College, we started our first learning session on "Monte Carlo Tree Search". The fact that the winner from the previous year used the MCTS algorithm was widely known, and we feel that this increased interest in the concept.
Like last year, the speaker for the session was Mr. Ranjan Shrestha, Chief Technology Officer at Bhoos Games. The Monte Carlo Tree Search is like a curious explorer traversing a branching maze, trying random paths until it discovers the most promising route to its treasure. The session was quite interactive and fruitful. After the session, participants decoded their doubts with the help of our Smart Bots Team.
Reinforcement Learning by Rojesh Shikhrakar
The speaker for the session was Mr. Rojesh M. Shikhrakar, a data scientist and machine learning engineer from Fusemachines. He introduced the concepts of Q-learning to the participants, and helped them build a Tic-Tac-Toe bot from scratch.
What is reinforcement learning?
In simple terms, reinforcement learning is like a student learning from trial and error by receiving rewards for their successes and penalties for their mistakes.
The session was scheduled for mid of the week, at Bhoos itself. We thought weekdays would attract a smaller number of participants; however, we witnessed around 50 participants in this session.
Developing AI Game bot from Scratch by Saurav Dhungana
Our final session for the Smart Bots Challenge concluded with a fundamental topic on creating game bots from scratch. The speaker for the session was Mr. Saurav Dhungana, the founder of Inductiv Science.
The session took place at the LOCUS Event at the Pulchowk Campus. We saw around 35 participants in the session. (We know most of them were busy exploring the LOCUS event.)
This session proved to be an eye-opener for all those who were stuck on how to get started with building a bot. We know this should have been the first session, but we also know that our participants love to hustle at the last hour. :)
Gift Hampers to teams who scored above 50%
It's quite an accomplishment to build a 29-point game bot that even outperforms Bhoos bots. This year, there were about 31 teams that were successful in defeating the Bhoos bots by scoring more than 50%.
Our Smart Team gave gift hampers to all such teams. These gift hampers were a token of appreciation for their hard work and dedication. The participants were thrilled to receive them, and it added to the excitement of the tournament.
We had a customized dashboard this year, much like the year before. As it would inform players of their bot performance, other teams’ rankings, progression of the competition, and had a ton of resources, it was highly reflective.
Here’s in detail about it.
The highlight of the qualifier phase was the fight to be on the top spot of the leaderboard. The competition was fierce, with participants working around the clock to come up with the best rankings. The leaderboard was constantly updated, with participants battling it out for the top spot.
This year, we incorporated two kinds of ranking systems.
One where a team’s bot played against Bhoos bots. Based on the winnings and other factors, they received a score. Unfortunately, using the "Bhoos bot" alone was not the most effective or reliable strategy. A few points of variation no longer accurately reflect the performance difference as the scores began to saturate.
Hence, here’s how Peer-to-Peer Mode Ranking came to play. P2P mode uses the TrueSkill algorithm developed by Microsoft. Here’s about it.
In the qualifying phase, over 27 thousand matches ( 27, 209 matches) and 11 million games (11,356,141 games) for 29 points were played in total. The competition was intense and challenging, and we were blown away by the level of skill and creativity demonstrated by all the participants. However, we selected our SUPER 8 for the road to winning Smart Bots-2023.
Congratulations to all the SUPER 8 teams, and thank you to all the participants for being a part of the Smart Bots 2023 family. We wish all the future code-experts be able to write clean codes and solve problems at hand very quickly.