\uc778\uacf5\uc9c0\ub2a5 \ubc0f \uac15\ud654\ud559\uc2b5(Deep Reinforcement Learning, DRL) \ubd84\uc57c\uc5d0\uc11c \ubcf4\uc0c1 \ud568\uc218\ub294 \ub9e4\uc6b0 \uc911\uc694\ud55c \uc5ed\ud560\uc744 \ucc28\uc9c0\ud569\ub2c8\ub2e4. \uac15\ud654\ud559\uc2b5\uc758 \ud575\uc2ec \uc544\uc774\ub514\uc5b4\ub294 \uc5d0\uc774\uc804\ud2b8\uac00 \ud658\uacbd\uacfc \uc0c1\ud638\uc791\uc6a9\ud558\uba74\uc11c \ubcf4\uc0c1\uc744 \ucd5c\ub300\ud654\ud558\ub294 \ubc29\ubc95\uc744 \ubc30\uc6b0\ub294 \uac83\uc774\uba70, \uc774\ub54c \ubcf4\uc0c1 \ud568\uc218\ub294 \uc5d0\uc774\uc804\ud2b8\uac00 \uc5b4\ub5a4 \ud589\ub3d9\uc744 \ucde8\ud574\uc57c \ud560\uc9c0\ub97c \uacb0\uc815\ud558\ub294 \ub370 \ud544\uc218\uc801\uc778 \uae30\uc900\uc744 \uc81c\uacf5\ud569\ub2c8\ub2e4. \ubcf8 \uae00\uc5d0\uc11c\ub294 \ubcf4\uc0c1 \ud568\uc218\uc758 \uc911\uc694\uc131\uacfc \uadf8 \uc124\uacc4 \uc6d0\uce59, \uadf8\ub9ac\uace0 \uc774\ub97c \ud65c\uc6a9\ud55c \uc608\uc81c \ucf54\ub4dc\ub97c \ud3ec\ud568\ud558\uc5ec \uc790\uc138\ud788 \uc124\uba85\ud558\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
\ubcf4\uc0c1 \ud568\uc218\ub294 \uc5d0\uc774\uc804\ud2b8\uac00 \ud658\uacbd \uc0c1\ud0dc\uc5d0\uc11c \ud2b9\uc815 \ud589\ub3d9\uc744 \ucde8\ud560 \ub54c, \uadf8 \ud589\ub3d9\uc758 \uac00\uce58\ub97c \ud3c9\uac00\ud558\ub294 \ud568\uc218\uc785\ub2c8\ub2e4. \ubcf4\uc0c1\uc740 \uc77c\ubc18\uc801\uc73c\ub85c \uc22b\uc790\ub85c \ud45c\ud604\ub418\uba70, \uae0d\uc815\uc801\uc778 \ubcf4\uc0c1\uc740 \uc5d0\uc774\uc804\ud2b8\uac00 \ubc14\ub78c\uc9c1\ud55c \ud589\ub3d9\uc744 \ucde8\ud558\ub3c4\ub85d \uc720\ub3c4\ud558\uace0, \ubd80\uc815\uc801\uc778 \ubcf4\uc0c1\uc740 \ubc14\ub78c\uc9c1\ud558\uc9c0 \uc54a\uc740 \ud589\ub3d9\uc744 \ud53c\ud558\ub3c4\ub85d \ud569\ub2c8\ub2e4. \ubcf4\uc0c1 \ud568\uc218\ub294 \uac15\ud654\ud559\uc2b5\uc758 ‘\ubaa9\ud45c’\ub97c \uc124\uc815\ud558\ub294 \uadfc\ubcf8\uc801\uc778 \uc694\uc18c\ub85c, \uc5d0\uc774\uc804\ud2b8\uac00 \ud559\uc2b5\ud558\ub294 \uacfc\uc815\uc5d0\uc11c \ucd5c\uc801\uc758 \uc815\ucc45(policy)\uc744 \ucc3e\ub294 \ub370 \ub9e4\uc6b0 \uc911\uc694\ud55c \uc5ed\ud560\uc744 \ud569\ub2c8\ub2e4.<\/p>\n
\ubcf4\uc0c1 \ud568\uc218\ub294 \uc5ec\ub7ec \uac00\uc9c0 \uc911\uc694\ud55c \uc5ed\ud560\uc744 \uc218\ud589\ud569\ub2c8\ub2e4:<\/p>\n
\ubcf4\uc0c1 \ud568\uc218\uc758 \uc124\uacc4\ub294 \uac15\ud654\ud559\uc2b5\uc758 \uc131\uacf5 \uc5ec\ubd80\uc5d0 \ud070 \uc601\ud5a5\uc744 \ubbf8\uce69\ub2c8\ub2e4. \uc801\uc808\ud558\uac8c \uc124\uacc4\ub41c \ubcf4\uc0c1 \ud568\uc218\ub294 \ud559\uc2b5 \uc18d\ub3c4\ub97c \ub192\uc774\uace0, \ub354 \ub098\uc544\uac00 \uc6d0\ud558\ub294 \uacb0\uacfc\ub97c \uc5bb\ub294 \ub370 \uae30\uc5ec\ud569\ub2c8\ub2e4. \ubc18\uba74\uc5d0 \ubd88\uc644\uc804\ud558\uac70\ub098 \ubd80\uc815\ud655\ud55c \ubcf4\uc0c1 \ud568\uc218\ub294 \uc5d0\uc774\uc804\ud2b8\uac00 \uc798\ubabb\ub41c \ubc29\ud5a5\uc73c\ub85c \ud559\uc2b5\ud558\uac8c \ub9cc\ub4e4 \uc218 \uc788\uc73c\uba70, \uadf9\ub2e8\uc801\uc778 \uacbd\uc6b0\uc5d0\ub294 \uc544\uc608 \ud6a8\uacfc\uc801\uc778 \ud559\uc2b5\uc774 \uc774\ub8e8\uc5b4\uc9c0\uc9c0 \uc54a\uc744 \uc218\ub3c4 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\uc608\ub97c \ub4e4\uc5b4, \uc790\uc728\uc8fc\ud589\ucc28\uc758 \uacbd\uc6b0 \ub3c4\ub85c\uc5d0\uc11c \ubcf4\ud589\uc790\ub97c \uc778\uc2dd\ud558\uace0 \uc548\uc804\ud558\uac8c \uc8fc\ud589\ud558\ub294 \uac83\uc774 \ubaa9\ud45c\uc785\ub2c8\ub2e4. \ub9cc\uc57d \ubcf4\uc0c1 \ud568\uc218\uac00 \ubcf4\ud589\uc790\ub97c \uc778\uc2dd\ud560 \ub54c\ub9c8\ub2e4 \uc18c\uc815\uc758 \ubcf4\uc0c1\uc744 \uc8fc\uace0, \ucda9\ub3cc\ud560 \uacbd\uc6b0 \ud070 \ud328\ub110\ud2f0\ub97c \ubd80\uc5ec\ud558\ub3c4\ub85d \uc124\uacc4\ub418\uc5b4 \uc788\ub2e4\uba74, \uc790\uc728\uc8fc\ud589\ucc28\ub294 \ubcf4\ud589\uc790\ub97c \ud53c\ud558\uace0 \uc548\uc804\ud558\uac8c \uc8fc\ud589\ud558\ub294 \ubc29\ubc95\uc744 \ud559\uc2b5\ud558\uac8c \ub429\ub2c8\ub2e4.<\/p>\n
\ubcf4\uc0c1 \ud568\uc218\ub97c \uc124\uacc4\ud560 \ub54c \uace0\ub824\ud574\uc57c \ud560 \uba87 \uac00\uc9c0 \uc6d0\uce59\uc774 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\n
\ubcf4\uc0c1 \ud568\uc218\ub294 \ucd5c\ub300\ud55c \uba85\ud655\ud558\uace0 \uc9c1\uad00\uc801\uc73c\ub85c \uc124\uacc4\ud574\uc57c \ud569\ub2c8\ub2e4. \uc5d0\uc774\uc804\ud2b8\uac00 \ubcf4\uc0c1\uc744 \ud1b5\ud574 \ubaa9\ud45c\ub97c \uc774\ud574\ud560 \uc218 \uc788\ub3c4\ub85d \ud558\uae30 \uc704\ud574\uc11c\ub294 \ubcf4\uc0c1 \uad6c\uc870\uac00 \uba85\ud655\ud574\uc57c \ud569\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, \ud2b9\uc815 \uc791\uc5c5\uc744 \uc131\uacf5\uc801\uc73c\ub85c \uc218\ud589\ud588\uc744 \ub54c \ubcf4\uc0c1\uc774 \uc8fc\uc5b4\uc9c0\ub294 \uad6c\uc870\ub294 \uc9c1\uad00\uc801\uc785\ub2c8\ub2e4.<\/p>\n
\ubcf4\uc0c1 \ud568\uc218\ub294 \ud2b9\uc815 \ud589\ub3d9\uc774 \uc591\ud638\ud558\uac70\ub098 \uc88b\uc744 \ub54c \uadf8\uc5d0 \uc0c1\uc751\ud558\ub294 \ubcf4\uc0c1\uc744 \uc8fc\uc5b4\uc57c \ud558\uba70, \ubc18\ub300\ub85c \ub098\uc05c \ud589\ub3d9\uc5d0 \ub300\ud574\uc11c\ub294 \ubd80\uc815\uc801\uc778 \ubcf4\uc0c1\uc744 \uc8fc\uc5b4\uc57c \ud569\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ub300\uce6d\uc131\uc744 \uc720\uc9c0\ud568\uc73c\ub85c\uc368 \uc5d0\uc774\uc804\ud2b8\ub294 \uc815\ud655\ud558\uac8c \uc5b4\ub5a4 \ud589\ub3d9\uc774 \uc131\uacf5\uc801\uc774\uace0 \uc5b4\ub5a4 \ud589\ub3d9\uc774 \uc2e4\ud328\uc778\uc9c0 \uc774\ud574\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\ubcf4\uc0c1 \uac12\uc758 \uc2a4\ucf00\uc77c\uc740 \uc5d0\uc774\uc804\ud2b8\uc758 \ud559\uc2b5 \uc18d\ub3c4\uc5d0 \uc601\ud5a5\uc744 \ubbf8\uce60 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ubc94\uc704\uac00 \ub108\ubb34 \ub113\uac70\ub098 \uc881\uc73c\uba74 \uc5d0\uc774\uc804\ud2b8\uac00 \ubcf4\uc0c1 \uc2e0\ud638\ub97c \ud559\uc2b5\ud558\ub294 \ub370 \uc5b4\ub824\uc6c0\uc744 \uacaa\uc744 \uc218 \uc788\uc73c\ubbc0\ub85c, \uc801\uc808\ud55c \ubc94\uc704\uc758 \ubcf4\uc0c1\uc744 \uc124\uc815\ud558\ub294 \uac83\uc774 \uc911\uc694\ud569\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, \ub9e4\uc6b0 \uc791\uc740 \ubcf4\uc0c1\uc740 \ud559\uc2b5\uc744 \ubc29\ud574\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\uc5d0\uc774\uc804\ud2b8\ub294 \uc0c8\ub85c\uc6b4 \ud589\ub3d9\uc744 \uc2dc\ub3c4\ud574\uac00\uba70 \ubcf4\uc0c1\uc744 \ud0d0\ud5d8\ud574\uc57c \ud558\uba70, \ub3d9\uc2dc\uc5d0 \uc9c4\ud589 \uc911\uc778 \ud559\uc2b5 \uacb0\uacfc\ub97c \ud1a0\ub300\ub85c \uae30\uc874\uc758 \ud589\ub3d9\uc744 \ucc29\ucde8\ud574\uc57c \ud569\ub2c8\ub2e4. \ubcf4\uc0c1 \ud568\uc218\ub294 \uc774 \ub450 \uac00\uc9c0\uc758 \uade0\ud615\uc744 \ub9de\ucd9c \uc218 \uc788\ub3c4\ub85d \uc124\uacc4\ub418\uc5b4\uc57c \ud569\ub2c8\ub2e4.<\/p>\n
\uc774\uc81c \uac04\ub2e8\ud55c \uc608\uc81c\ub97c \ud1b5\ud574 \ubcf4\uc0c1 \ud568\uc218\ub97c \uc5b4\ub5bb\uac8c \uc124\uacc4\ud558\uace0 \uad6c\ud604\ud560 \uc218 \uc788\ub294\uc9c0 \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc774\ubc88 \uc608\uc5d0\uc11c\ub294 OpenAI\uc758 Gym \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc0ac\uc6a9\ud558\uc5ec \uac15\ud654\ud559\uc2b5 \uc5d0\uc774\uc804\ud2b8\ub97c \ud6c8\ub828\uc2dc\ud0ac \uac83\uc785\ub2c8\ub2e4.<\/p>\n
!pip install gym numpy<\/code><\/pre>\n5.2. \ud658\uacbd \uc0dd\uc131<\/h3>\nimport gym\nimport numpy as np\n\n# CartPole \ud658\uacbd \uc0dd\uc131\nenv = gym.make('CartPole-v1')\nstate = env.reset()\ndone = False\ntotal_reward = 0\n\nwhile not done:\n action = env.action_space.sample() # \ubb34\uc791\uc704 \ud589\ub3d9 \uc120\ud0dd\n state, reward, done, info = env.step(action) # \ud589\ub3d9 \uc218\ud589\n total_reward += reward\n\nprint(f\"Total reward: {total_reward}\")<\/code><\/pre>\n5.3. \ucee4\uc2a4\ud140 \ubcf4\uc0c1 \ud568\uc218 \uc791\uc131<\/h3>\n
\uae30\ubcf8\uc801\uc778 \ubcf4\uc0c1 \ud568\uc218\ub97c \ucee4\uc2a4\ud130\ub9c8\uc774\uc988\ud558\uc5ec \uc5d0\uc774\uc804\ud2b8\uc758 \ud559\uc2b5\uc744 \ub354\uc6b1 \ud6a8\uacfc\uc801\uc73c\ub85c \ub9cc\ub4e4 \uc218 \uc788\uc2b5\ub2c8\ub2e4. CartPole \uc608\uc81c\uc5d0\uc11c\ub294 \ub9c9\ub300\uac00 \uc138\uc6cc\uc9c4 \uc0c1\ud0dc\ub97c \uc720\uc9c0\ud558\ub3c4\ub85d \uc720\ub3c4\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
def custom_reward(state):\n angle = state[2] # \ub9c9\ub300\uc758 \uac01\ub3c4\n return 1 - abs(angle) \/ (np.pi \/ 2) # \uac01\ub3c4\ub97c \uae30\uc900\uc73c\ub85c \ubcf4\uc0c1 \uacc4\uc0b0<\/code><\/pre>\n5.4. \uc804\uccb4\uc801\uc778 \ud559\uc2b5 \ucf54\ub4dc<\/h3>\nimport gym\nimport numpy as np\n\ndef custom_reward(state):\n angle = state[2]\n return 1 - abs(angle) \/ (np.pi \/ 2) # \uac01\ub3c4\uc5d0 \ub530\ub978 \ubcf4\uc0c1\n\nenv = gym.make('CartPole-v1')\nnum_episodes = 1000\n\nfor episode in range(num_episodes):\n state = env.reset()\n total_reward = 0\n done = False\n \n while not done:\n action = env.action_space.sample() # \ubb34\uc791\uc704 \ud589\ub3d9 \uc120\ud0dd\n state, _, done, _ = env.step(action) # \ud589\ub3d9 \uc218\ud589\n reward = custom_reward(state) # \ucee4\uc2a4\ud140 \ubcf4\uc0c1 \ud568\uc218\n total_reward += reward\n \n print(f\"Episode {episode + 1}: Total Reward: {total_reward}\")<\/code><\/pre>\n6. \uacb0\ub860<\/h2>\n
\ubcf4\uc0c1 \ud568\uc218 \uc124\uacc4\ub294 \uac15\ud654\ud559\uc2b5\uc758 \uc131\uacf5\uc744 \uc88c\uc6b0\ud558\ub294 \uc911\uc694\ud55c \uc694\uc18c\uc785\ub2c8\ub2e4. \uc774\ubc88 \uae00\uc5d0\uc11c\ub294 \ubcf4\uc0c1 \ud568\uc218\uc758 \uc815\uc758, \uc5ed\ud560, \uadf8\ub9ac\uace0 \uc124\uacc4 \uc6d0\uce59\uc744 \uc0b4\ud3b4\ubcf4\uc558\uc2b5\ub2c8\ub2e4. \ub610\ud55c, \uc2e4\uc81c\ub85c \ucee4\uc2a4\ud140 \ubcf4\uc0c1 \ud568\uc218\ub97c \uc791\uc131\ud558\uc5ec \uc5d0\uc774\uc804\ud2b8\uac00 \ud658\uacbd\uc5d0\uc11c \ub354 \ud6a8\uacfc\uc801\uc73c\ub85c \ud559\uc2b5\ud558\ub3c4\ub85d \uc720\ub3c4\ud558\ub294 \ubc29\ubc95\uc744 \uc54c\uc544\ubcf4\uc558\uc2b5\ub2c8\ub2e4. \uac15\ud654\ud559\uc2b5\uc744 \ucc98\uc74c \uc2dc\uc791\ud558\ub294 \uac1c\ubc1c\uc790\ub4e4\uc740 \ubcf4\uc0c1 \ud568\uc218\ub97c \uc138\uc2ec\ud558\uac8c \uc124\uacc4\ud568\uc73c\ub85c\uc368 \ub354 \ub098\uc740 \ud559\uc2b5 \uc131\uacfc\ub97c \uc774\ub04c\uc5b4\ub0bc \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\ub2e4\uc74c \uae00\uc5d0\uc11c\ub294 \ub2e4\uc591\ud55c \ud658\uacbd\uc5d0\uc11c \ubcf4\uc0c1 \ud568\uc218\ub97c \ucd5c\uc801\ud654\ud558\uae30 \uc704\ud55c \ubc29\ubc95\ub860\uacfc \uc804\ub7b5\uc5d0 \ub300\ud574 \uc54c\uc544\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
<\/body><\/p>\n","protected":false},"excerpt":{"rendered":"