2023-Q1-AI 15. DQN DDQN

 

15.1. Video / Materials, 24 Apr 2023, 18:00

Video: https://youtube.com/live/0xHdJC_6gyk?feature=share

Jamboard: https://jamboard.google.com/d/116aaOmwkG7Dvo0Tg63WduZhwyYQhQqXciZa7ohqTtCQ/edit?usp=sharing

Sagatavošanās materiāli: Rainbow DQN: https://arxiv.org/abs/1710.02298 https://storage.googleapis.com/deepmind-media/dqn/DQNNaturePaper.pdf https://pytorch.org/tutorials/intermediate/reinforcement_q_learning.html https://medium.freecodecamp.org/an-introduction-to-reinforcement-learning-4339519de419


 

^ Shared: stefan.dayneko@gmail.com

Youtube key: ea11-mrgb-4jg2-4ajc-d4hr


 

Iepriekšējā gada video

Video https://youtu.be/tiaoLNMWZUA

Jamboard: https://jamboard.google.com/d/18gFXn4E36cP9P25wSKpvGgEv1mlAnKfDQbB7fGTDPi0/viewer

 


 

13.2. Implementēt DQN

Balstoties uz 13.1. materiāliem un video implementēt DQN, izmantojot sagatavi.

Iesniegt kodu un ekrānšāviņus ar rezultātiem.

Template: http://share.yellowrobot.xyz/1628158950-vea-rtu-course-2020-q1/13_2_dqn_lunar_lander_unfinished.py


13.3. Implementēt priority replay memory

Balstoties uz 13.1. materiāliem un video implementēt "Priority replay memory", izmantojot sagatavi.

Iesniegt kodu un ekrānšāviņus ar rezultātiem.

Template: http://share.yellowrobot.xyz/1628158950-vea-rtu-course-2020-q1/13_5_priority_dqn_lunar_lander_unfinished.py


13.4. Implementēt DDQN

Implementēt DDQN, balstoties uz 13.3 uzdevuma sagatavi.

Iesniegt kodu un ekrānšāviņus ar rezultātiem.

Vienādojums: http://share.yellowrobot.xyz/1628158950-vea-rtu-course-2020-q1/ddqn.png


13.5. Mājasdarbs - Dueling DDQN + MountainCar

  1. Balstoties uz 13.4 kodu, implementēt jaunu vidi MountainCar: https://gym.openai.com/envs/MountainCar-v0

  2. Implementēt Dueling DDQN modeļa arhitektūru

  3. Iesniegt kodu un ekrānšāviņus ar rezultātiem.

Modeļa shēma: http://share.yellowrobot.xyz/1628158950-vea-rtu-course-2020-q1/dual.png

Modeļa apraksts: https://arxiv.org/abs/1511.06581


Materials

6CFCE675-1F3C-4980-8DF0-7A304ECBEC23

ECDB0E4E-5E4D-4BE5-A719-BBD0A29F9689-2

317CD11B-0003-4132-B1BD-DC4C90EA4852

E9105FA6-104A-49F9-AA3F-A56A4DA89C9C

F783C618-5E70-4477-BB75-ADF03ADF5264

CF51809D-04EE-4CDD-B340-A8122E59E04D

D416AFAD-E68B-4106-8BF5-F532C8DFD17E

D0F8F846-6308-404F-9172-48E3ED484C38

C95CFEC9-C2B8-48D4-BA46-8A16CD97A54A

B84E1B2A-C95F-40F1-9727-F9C33D880795

28454713-7BA1-4216-AE46-BC195CA49D21

07614319-25B9-4D8E-8A20-3A7CD86465F0

3780603F-CE1A-4004-89DE-7127000CF9DB

44428F96-1D4A-46BD-A635-8D1C51C295BE

18054E82-30E9-486B-9498-A00131F7BA3B

3761D110-C085-45E0-87BE-E84ED73595DF

385F5BB1-823B-43EF-ACFA-0E3829E0836D

45BE7E4B-55B6-49CD-A974-AEA4BEEA3AAA