Jekyll2022-09-20T11:19:25-05:00https://minhlong94.github.io/blog/feed.xmlAn easy to use blogging platform with support for Jupyter Notebooks.2022-02-10T00:00:00-06:002022-02-10T00:00:00-06:00https://minhlong94.github.io/blog/tricks/2022/02/10/Pycharm%20and%20Linux%20server

Introduction I am not a fan of long wording so let’s jump straight into the problem.

2022-01-04T00:00:00-06:002022-01-04T00:00:00-06:00https://minhlong94.github.io/blog/drl/shitpost/2022/01/04/RL%20Optimal%20Example%20With%20Faker

Recently I tried to explain a struggle in a project to a friend, who had zero knowledge in Reinforcement Learning, that was: my friend set up a suitable action space for an agent to choose to “neglect” a chest in a gridworld, and she thought: “if the agent is smart enough, it should learn how to neglect these chests if necessary.” The key word here is “if necessary”: she did not know what cases should be considered “necessary”. In short, she did not know the optimal policy and under what conditions would make the policy optimal. Say, for example, if you have 10 steps left in the gridworld, and you can both get the chest and the goal in 10 steps, why should you neglect the chest, assume that it gives bonus reward and the number of steps left does not contribute to the reward? Transfer to our problem, in this case, she wonders why the agent does not neglect the chest (expectation), but she does not know how many steps are left, and where is the agent on the gridworld (reality).

2021-09-25T00:00:00-05:002021-09-25T00:00:00-05:00https://minhlong94.github.io/blog/drl/2021/09/25/Making%20your%20Deep%20RL%20matters

A collection of implementation tricks, hyperparameter sensitivity, and others in Deep RL which I gave a presentation in my research group.

2021-09-25T00:00:00-05:002021-09-25T00:00:00-05:00https://minhlong94.github.io/blog/drl/2021/09/25/Action%20space%20shaping%20in%20Deep%20Reinforcement%20Learning

This is a presentation of the paper “Action space shaping in Deep Reinforcement Learning”, by Anssi Kanervisto et al., 2020, in IEEE Conference on Games 2020.

2021-09-10T00:00:00-05:002021-09-10T00:00:00-05:00https://minhlong94.github.io/blog/shitpost/2021/09/10/A-Neural-Network-To-Print-From-1-to-50

Recently, a user named logo asked a question on the RL Discord server: how to print numbers from 1 to 50 in python? Little did I know, I was about to engage in one of the funniest conversation. Pure Python approach for i in range(50): print(i) A very simple and straightforward solution. Wait, why can’t we just print(1,2,3,4,5,6,7,8,9,10...,50)?