Reinforcement Learning: Deep Q learning
#day2 of #100daysofcode
Today I learned about Deep Q Learning which comes in Reinforcement Learning. It is very fascinating and solves the problem of the memory issue of the Q — table. I watched these videos by sentdex.
Basically we use a Deep network instead of a Q-table. Here we will pass the images of the current state of the game. I have not picked a problem yet to solve it. I have not completed it today, therefore, tomorrow I will continue it.
Tomorrow I will complete its code and also implement it on a game. Maybe CartPole game. It is an amazing concept and I need to go deeper into it.
So today my most of the time was spent on learning the concept and I wrote little less code. But tomorrow I will compensate for it. As the concept will become totally clear than the coding part is straight forward and will be interesting to play with. We are going to play some amazing games now using our agent.
Stay with me for more exploration. Today I am not explaining anything but tomorrow or maybe day after tomorrow with an application of it, I will explain the concept to you.
The brief concept is that yesterday we used a Q-table we store values for each action for each state. And if the number of states and actions increases then we need a very large table which is not a good idea and also not efficient. To tackle this problem we use a Deep network.
Github link
