• Data Science BKK #8 powered by Wisesight

    บริษัท ไวซ์ไซท์ (ประเทศไทย) จำกัด

    Shoutout to our venue sponsor Wisesight, the leader in social listening in Thailand and Southeast Asia. RSVP at: https://www.meetup.com/Bangkok-School-of-AI/events/262409333 True to our philosophy of no-nonsense data science, this coming Thursday we have two speakers, with varying degrees of credentials. One thing they can both claim, however, is that they have created a data product that has real impact. Meet Chawan, winner of KBTG nationwide TechJam 2018, and Tor, the Data Research Manager at one of Southeast Asia's largest social listening companies Wisesight. Free beer 🍻(2 barrels) and pizzas on a first-come-first-serve basis. Talks: 1. Lesson from TechJam 2018 - Using Deep Learning in Big Data Analysis by Chawan Piansaddhayanon, Student at Chulalongkorn University / Winner of TechJam 2018 / Intern at Oxygen.AI For predictive model, it is common that deep learning often outperforms traditional machine learning when there is a large amount of data. However, the use of deep learning is not that popular as it is well-known for having low interpretability compared to the traditional one. Yet this might not always be true as now, there are tools for visualizing the model and some modern architecture, such as Autoencoder can give us some understanding about the data. Therefore, the advantage of using deep learning, the way to get some insight from the data and understand will be discussed in this talk by using the data from 2018 techjam competition, which is time series data, as an example. 2. [Talk in Thai] Learning from Social Data and Thailand Elections by Puttasak Tantisuttivet (Tor), Data Research Manager at Wisesight The long-awaited general election in Thailand has been one of the most controversial and actively discussed one in our history. This is in no small part fueled by social media. Tor leads Wisesight data unit in one of the region's largest and most sophisticated social listening operations. Let him walk you through the steps of churning massive social media data into valuable insights including how to deal with sampling, sourcing, and noise. Schedule: - Door opens at 18:30 - Talk with a small break in between - Networking ends around 20:30 * Schedule might have minor changes according to circumstances. Stay tuned. See you all at the meetup! About Us Data Science BKK is a no-nonsense, no-agenda meetup for data science and data engineering practitioners in Thailand. We welcome speakers and participants from all companies and industries. We believe in a place where we can truly share ideas and best practices without commercial interests--data people to data people.

    4
  • Recommendation System Workshop with SnapLogic Snap Dataset

    True Innovation Center @ Siam Soi 3 (Opposite K-Bank) Level 4

    Welcome to the One-shot Workshop Series by Bangkok School of AI. The purpose of these workshops is to get you to learn ONE skill after two hours of a tutorial session, then allow you to put that skill to use in the next two-hour mini-hackathon. This time we will learn to create a recommendation system from scratch using tree-based models with scikit-learn and deep learning models using PyTorch. The dataset contains 25,331 pipelines used internally at SnapLogic. Each pipeline contains a series of snaps--a block of codes that perform various functions such as data ingestion, manipulation, and even machine learning. No pipeline in this dataset belongs to SnapLogic customers. Venue sponsor: True Innovation Center @ Siam Soi 3 (Opposite K-Bank) Level 4 Dataset sponsor: SnapLogic is one of the world's leading data integration companies based in San Mateo, CA. They provide excellent machine learning use cases using their platform in this repository. See details here: https://github.com/datatouille/snaplogic_snap_recommendation We expect participants to be comfortable with Python; machine learning knowledge is a plus. Other than that, you are all set to come. All codes can be run by Google Colab so you only need to bring your laptop, Internet access and curiosity. Our mission is to offer a world-class AI education to anyone on Earth for free. This workshop, therefore, is FREE, although we will de-prioritize you in the next meetups for no shows.

    5
  • RL Workshop #6: Continuous Control with Deep Deterministic Policy Gradient

    This is the last session in RL workshop series. If you have stuck with us this long, you're amazing. So far, we have dealt with almost every mainstream RL and Deep RL algorithms such as Monte Carlo Methods, (Deep) Q-learning, and Policy Gradient. We can solve almost any reinforcement learning with these tools except one case: continuous actions. In this session, you will learn to use Deep Deterministic Policy Gradient (DDPG), a type of algorithm similar to the actor-critic family, to solve an environment with both continuous states AND actions. Our venue is sponsored by CRYPTONIST - Empowering Blockchain Community. See their meetup at https://www.meetup.com/CRYPTONIST-Meetup/ We recommend that you read up on the materials we have gone through so far before attending the session. You will first use a pre-configured agent to see how it works, and then we will deep-dive into how each and every element works, including all mathematical equations and codes (in the least painless way for humans possible). We leave no stone unturned; you will get the taste of both the hype and the reality. See details here: https://github.com/Datatouille/rl-workshop We expect participants to be comfortable with Python and a healthy level of familiarity with machine learning algorithms such as classification, regression and clustering. Other than that, you are all set to come. All codes can be run by Google Colab so you only need to bring your laptop and curiosity. Our mission is to offer a world-class AI education to anyone on Earth for free. This workshop, therefore, is FREE, although we will de-prioritize you in the next meetups for no shows.

    1
  • Data Science BKK #7

    KX Knowledge Exchange Center

    Shoutout to our venue sponsor CRYPTONIST - Empowering Blockchain Community. See their meetup at https://www.meetup.com/CRYPTONIST-Meetup/ Happy holidays everyone! Bangkok's least bullsh*t-tolerant data science meetup is back. We are extremely excited to have two practitioners with track records from two multinational corporations to come and tell us about their work. Mo is a geologist at the energy firm Chevron who taught himself data science just a year ago to help with oil and gas exploration, and recently came first at the company's global data science competition - not Thailand, not Asia, best in the world. Alan is an ex-Lazada data scientist and co-founder of Data Science BKK who jumped shi... went on a quest to revolutionize the music industry as a knowledge transfer associate at University of Leeds. He is bringing data science to one of UK's largest record label the same way Billy Beane brought data science to the Boston Redsox. See you at the meetup! Talks: 1. Data Science from Scratch: How to Learn Data Science for non-IT People by Mo Pongtachchai, Geologist at Chevron One of the most frequently asked questions in Data Science BKK is "where do I start". Mo will guide you through the process he took during the past year to go from zero to hero (literally) in data science. This talk is for those who are non-coders and those who forgot their high school math. 2. Let The Artists Do Their Work! How Data Will Help Music Industry? by Alan Choicharoon, Data Scientist at University of Leeds Music industry is quite adaptive to change forced upon it by advancement in technology. From the time of vinyl to cassette, cassette to CD, and CD to current streaming platform. It has learn to change with time. However, that's not to said that it has been easy nor that there are still no room for data to improve work process in the industry. I am going to talk about my work with experts in music industry to find how data will help them works more efficiently and productively such as providing a more robust benchmark, revealing hidden connection amongst artist, and automating task that are mundane, but also that there are still no ignoring the human factor in this industry. Schedule: - Registration starts at 18:30 - Talk by Mo and Alan with a small break in between - Networking ends around 20:30 * Schedule might have minor changes according to circumstances. Stay tuned. See you all at the meetup! About Us Data Science BKK is a no-nonsense, no-agenda meetup for data science and data engineering practitioners in Thailand. We welcome speakers and participants from all companies and industries. We believe in a place where we can truly share ideas and best practices without commercial interests--data people to data people.

    5
  • RL Workshop #5: Deep Q-Learning to Drive MountainCar

    After getting rid of the Q dictionary and replacing it with a stochastic policy powered by neural networks, this time we revisit the idea of value-based learning - with a twist. In this session, you will learn to use neural networks as your Q-value estimator to play the game of MountainCar-v0 and (as a take-home challenge) MountainCarContinuous-v0. For those who want to learn more about GET or are interested to be a tester of its Beta App, you can visit www.getthailand.com. We recommend that you read up on the materials we have gone through so far before attending the session. You will first use a pre-configured agent to see how it works, and then we will deep-dive into how each and every element works, including all mathematical equations and codes (in the least painless way for humans possible). We leave no stone unturned; you will get the taste of both the hype and the reality. See details here: https://github.com/Datatouille/rl-workshop We expect participants to be comfortable with Python and a healthy level of familiarity with machine learning algorithms such as classification, regression and clustering. Other than that, you are all set to come. All codes can be run by Google Colab so you only need to bring your laptop and curiosity. Our mission is to offer a world-class AI education to anyone on Earth for free. This workshop, therefore, is FREE, although we will de-prioritize you in the next meetups for no shows.

    1
  • RL Workshop #4: Policy-gradient Pong

    GET! Office

    We are moving closer to state-of-the-art model that is deep reinforcement learning. We will start with something that looks like Monte Carlo methods but not quite the same: policy-gradient methods. The model will play one of the all-time favorite games: Atari Pong. For those who want to learn more about GET or are interested to be a tester of its Beta App, you can visit www.getthailand.com. We recommend that you read up on the materials we have gone through so far before attending the session. You will first use a pre-configured agent to see how it works, and then we will deep-dive into how each and every element works, including all mathematical equations and codes (in the least painless way for humans possible). We leave no stone unturned; you will get the taste of both the hype and the reality. See details here: https://github.com/Datatouille/rl-workshop We expect participants to be comfortable with Python and a healthy level of familiarity with machine learning algorithms such as classification, regression and clustering. Other than that, you are all set to come. All codes can be run by Google Colab so you only need to bring your laptop and curiosity. Our mission is to offer a world-class AI education to anyone on Earth for free. This workshop, therefore, is FREE, although we will de-prioritize you in the next meetups for no shows.

    2
  • Reinforcement Learning Workshop #3b: Neural Networks in Pytorch

    So far we have learned two value-based methods, Monte Carlo and temporal difference learning, in discrete state and action spaces. This has allowed us to solve problems whose state and action representations are categorical such as grid locations in Grid World, hit/stay in Black Jack, and taxi-passenger combinations in Taxi-v2 by storing our learned action values in a Q dictionary. But as you might have noticed, our world is a continuous one. The state representations we observe usually come in continuous numbers such as temperature in Celcius, colors in RGB, and sounds in decibels. And if we were to build a Q dictionary out of all the fine-grained numbers, we would be ending up with a very, very huge one. That is why after this session, we will replace dictionaries with neural networks. This session is meant for those who are not familiar with neural networks at all but have used modeling packages such as caret in R, sklearn in python, or Keras. We will be learning about the building blocks of a neural network, how to make one, and end with an example that explains why we even need one instead of just a linear model. Our choice of framework is Pytorch due to flexibility and ease of use but the same principles will apply across all frameworks. For those who want to learn more about our fantastic venue sponsor GET or are interested to be a tester of their Beta App, you can visit www.getthailand.com. We recommend that you read up on the materials we have gone through so far before attending the session. You will first use a pre-configured agent to see how it works, and then we will deep-dive into how each and every element works, including all mathematical equations and codes (in the least painless way for humans possible). We leave no stone unturned; you will get the taste of both the hype and the reality. See details here: https://github.com/Datatouille/rl-workshop We expect participants to be comfortable with Python and a healthy level of familiarity with machine learning algorithms such as classification, regression and clustering. Other than that, you are all set to come. All codes can be run by Google Colab so you only need to bring your laptop and curiosity. Our mission is to offer a world-class AI education to anyone on Earth for free. This workshop, therefore, is FREE, although we will de-prioritize you in the next meetups for no shows.

    4
  • Reinforcement Learning Workshop #3: GET a Taxi with Temporal Difference Learning

    We are happy to announce our new sponsor venue for November, GET! - the new ride-hailing service. They are providing a gorgeous training room for us to do the workshop. For those who want to learn more about GET or are interested to be a tester of its Beta App, you can visit www.getthailand.com. In the spirit of GET!, we will be solving another classic reinforcement learning and real-life problem of taxi driving using OpenAI's Taxi-v2 environment. Some of your might have tried solving RL problems with Monte Carlo from last session. In this session, we will be comparing that approach with another family of algorithms which include SARSA and Q-learning. As a take-home challenge, you will be solving one of OpenAI Gym or Unity's environment of your choosing. We recommend that you read up on the materials we have gone through so far before attending the session. You will first use a pre-configured agent to see how it works, and then we will deep-dive into how each and every element works, including all mathematical equations and codes (in the least painless way for humans possible). We leave no stone unturned; you will get the taste of both the hype and the reality. See details here: https://github.com/Datatouille/rl-workshop We expect participants to be comfortable with Python and a healthy level of familiarity with machine learning algorithms such as classification, regression and clustering. Other than that, you are all set to come. All codes can be run by Google Colab so you only need to bring your laptop and curiosity. Our mission is to offer a world-class AI education to anyone on Earth for free. This workshop, therefore, is FREE, although we will de-prioritize you in the next meetups for no shows.

    5
  • Reinforcement Learning Workshop #2: Win Big at Monte Carlo

    Big Co-Working Space

    Last session, you guys have been amazing and really enthusiastic to learn the basics of reinforcement learning through a very simple GridWorld example. Let's build on that and learn one of the simplest yet useful tricks in solving the control problem--Monte Carlo Methods--by trying to win at Black Jack! In this session, we will learn how to actually solve a reinforcement learning problem with discrete states and actions by estimating the action values and choosing the best policy using Monte Carlo methods. At the end of the session, you are expected to be able to create an agent that is better than anyone in the world--human or robotic--at playing Black Jack. As a take-home challenge, you will be solving one of OpenAI Gym or Unity's environment of your choosing. This session of the workshop is sponsored by Humanize, the company that helps your business grow with AI. Humanize develops all AI-powered products such as enterprise-grade chatbots for your specific needs. You will first use a pre-configured agent to see how it works, and then we will deep-dive into how each and every element works, including all mathematical equations and codes (in the least painless way for humans possible). We leave no stone unturned; you will get the taste of both the hype and the reality. See details here: https://github.com/Datatouille/rl-workshop We expect participants to be comfortable with Python and a healthy level of familiarity with machine learning algorithms such as classification, regression and clustering. Other than that, you are all set to come. All codes can be run by Google Colab so you only need to bring your laptop and curiosity. Our mission is to offer a world-class AI education to anyone on Earth for free. This workshop, therefore, is FREE, although we will de-prioritize you in the next meetups for no shows.

    2
  • Reinforcement Learning Workshop #1: Escaping GridWorld with Simple RL Agents

    Let's kick off Bangkok School of AI by the hypest of the hyped branch of approaches: reinforcement learning aka AlphaGo, aka OpenAI Five DotA bots, aka pancake-flipping robots, aka all the cool "real AI" stuff. In this session, you will learn the basics of reinforcement learning that is Markov Decision Process. Our goal is to make a blind agent escape a 2D maze full of traps and unpredictable winds called GridWorld. You will first use a pre-configured agent to see how it works, and then we will deep-dive into how each and every element works, including all mathematical equations and codes (in the least painless way for humans possible). We leave no stone unturned; you will get the taste of both the hype and the reality. See details here: https://github.com/Datatouille/rl-workshop We expect participants to be comfortable with Python and a healthy level of familiarity with machine learning algorithms such as classification, regression and clustering. Other than that, you are all set to come. All codes can be run by Google Colab so you only need to bring your laptop and curiosity. Our mission is to offer a world-class AI education to anyone on Earth for free. This workshop, therefore, is FREE, although we will de-prioritize you in the next meetups for no shows.

    9