Please enable JavaScript.

Coggle requires JavaScript to display documents.

ML for QA (Problem we want to solve (Smoke test, performance tracking,…

- - - - exploration then exploitation
      - plus approprié pour des grilles
      - not dead + 1
    - - Handle the exploration/exploitation directly
      - longer to train
      - slower converge, local maximum
      - policy score function
        
        average reward per time step
        
        average value
        
        Total reward from start to end
      - Deep Reinforcement Learning: Pong from Pixels
      - better than q leraning
    - - much less traning than policy gradient
      - :star:ACKTR
        
        :star: Best so far
        
        Better than A2c
        
        OpenAI Baselines: ACKTR & A2C
      - Actor Advantage Critic (A2C) model
        
        Intuitive RL: Intro to Advantage-Actor-Critic (A2C)
      - TRPO
      - PPO
      - A3C
        
        Asynchronous Actor-Critic Agents (A3C)
    - - compute qtable with NN
      - Better suit than Q Leaning for large environment
    - - Intro and Definition
      - Can agents learn inside of their own dreams?
  - - - Gym Retro
    - - Q-learning
        
        Q-Learning with Tables and Neural Networks
      - A3C
        
        Asynchronous Actor-Critic Agents (A3C)
      - ML
        
        https://www.youtube.com/watch?v=Se9ByBnKb0o&list=PLXO45tsB95cJHXaDKpbwr5fC_CCYylw1f
      - Policy Gradient
        
        An introduction to Policy Gradients with Cartpole and Doom
      - Deep Reinforcement Learning
        
        Deep Deterministic Policy Gradients in TensorFlow
      - Implementations of Reinforcement Learning Models in Tensorflow
      - Simple Reinforcement Learning in Tensorflow
    - - q-learning
        
        Keras plays catch, a single file Reinforcement Learning example
        
        Deep Q-Learning with Keras and Gym
      - Actor-Critic
        
        Reinforcement Learning w/ Keras + OpenAI: Actor-Critic Models
        
        :star:Reinforcement Learning w/ Keras+OpenAI: The Basics
        
        :star:Intro Intuitive RL: Intro to Advantage-Actor-Critic (A2C)
    - - A2C
        
        simple-A2C/3_A2C TUTORIAL
    - - Policy gradient
        
        Reinforcement learning with policy gradient
      - a3c
        
        Improved RL with Parallel Advantage Actor-Critic
- - - - A policy defines the learning agent's way of behaving at a given time.
- - - - more, simpler example than ACKTR
- - - - Create Observation Space
      - Create Reward logic
      - Create Reset and Step interface
      - Create Action Space
    - - Training
        
        For nb game
        
        reset env
        
        while !done
        
        step forward
        
        Train
    - - Create environment input
        
        create numerical representation of the environment
        
        Convert screen to input
        
        remove art noise
      - Create model