\uac15\ud654\ud559\uc2b5(Reinforcement Learning, RL)\uc740 \uc5d0\uc774\uc804\ud2b8\uac00 \ud2b9\uc815 \ud658\uacbd\uc5d0\uc11c \uc0c1\uc744 \uadf9\ub300\ud654\ud558\uae30 \uc704\ud574 \ud559\uc2b5\ud558\ub294 \ubc29\ubc95\ub860\uc785\ub2c8\ub2e4. \uae30\ubcf8\uc801\uc73c\ub85c RL\uc740 \uc5d0\uc774\uc804\ud2b8\uac00 \uc8fc\uc5b4\uc9c4 \ud658\uacbd(state)\uc5d0\uc11c \ud589\ub3d9(action)\uc744 \uc120\ud0dd\ud558\uace0, \uadf8 \uacb0\uacfc\ub85c \ubcf4\uc0c1(reward)\uc744 \ubc1b\uc73c\uba70 \ud559\uc2b5\uc774 \uc9c4\ud589\ub429\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc5d0\uc11c \uc5d0\uc774\uc804\ud2b8\ub294 \ud658\uacbd\uc5d0 \ub300\ud55c \uc774\ud574\uc640 \ud589\ub3d9\uc5d0 \ub300\ud55c \ucd5c\uc801 \uc804\ub7b5\uc744 \uac1c\ubc1c\ud558\uac8c \ub429\ub2c8\ub2e4. \uadf8\ub7ec\ub098 \uac15\ud654\ud559\uc2b5\uc758 \uc801\uc6a9 \ubd84\uc57c\uac00 \ud655\ub300\ub428\uc5d0 \ub530\ub77c \ud658\uacbd\uc744 \uc5b4\ub5bb\uac8c \ud655\uc7a5\ud560 \uac83\uc778\uc9c0, \uadf8\ub9ac\uace0 \uc804\uc774 \ud559\uc2b5\uc744 \ud1b5\ud574 \uc774\uc804 \ud559\uc2b5 \uacbd\ud5d8\uc744 \uc5b4\ub5bb\uac8c \ud65c\uc6a9\ud560 \uac83\uc778\uc9c0\uc5d0 \ub300\ud55c \ud544\uc694\uc131\uc774 \uc911\uc694\ud574\uc9c0\uace0 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\ud658\uacbd \ud655\uc7a5\uc774\ub780, \uae30\uc874\uc758 \uac15\ud654\ud559\uc2b5 \ud658\uacbd\uc744 \uae30\ubc18\uc73c\ub85c \uc0c8\ub85c\uc6b4 \uc694\uc18c\ub97c \ucd94\uac00\ud558\uc5ec \ub354 \ubcf5\uc7a1\ud55c \ubb38\uc81c\ub97c \ub2e4\ub8e8\uac70\ub098, \uc5d0\uc774\uc804\ud2b8\uc758 \uc778\uc9c0 \ub2a5\ub825\uc744 \ud5a5\uc0c1\uc2dc\ud0a4\uae30 \uc704\ud55c \ub2e4\uc591\ud55c \ubcc0\uc218\ub97c \ub3c4\uc785\ud558\ub294 \uacfc\uc815\uc744 \uc758\ubbf8\ud569\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, \ub85c\ubd07 \uc190\uc758 \uac15\ud654\ud559\uc2b5\uc744 \uace0\ub824\ud574 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \ucd08\uae30 \ud658\uacbd\uc5d0\uc11c\ub294 \ub2e8\uc21c\ud788 \ubb3c\uccb4\ub97c \uc7a1\ub294 \uac83\uc5d0 \ucd08\uc810\uc744 \ub9de\ucd9c \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uadf8\ub7ec\ub098 \ud658\uacbd\uc744 \ud655\uc7a5\ud558\uc5ec \ubb3c\uccb4\uc758 \uc9c8\uac10, \ud06c\uae30, \ud615\ud0dc \ub4f1 \ub2e4\uc591\ud55c \ud2b9\uc131\uc744 \uace0\ub824\ud558\ub3c4\ub85d \ud558\uc5ec \ub85c\ubd07\uc774 \ubcf4\ub2e4 \ub2e4\uc591\ud558\uace0 \ubcf5\uc7a1\ud55c \uc791\uc5c5\uc744 \uc218\ud589\ud558\ub3c4\ub85d \ud6c8\ub828\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\uc804\uc774 \ud559\uc2b5(Transfer Learning)\uc740 \uc5d0\uc774\uc804\ud2b8\uac00 \uc774\uc804 \ud658\uacbd\uc5d0\uc11c \uc2b5\ub4dd\ud55c \uc9c0\uc2dd\uc774\ub098 \uae30\uc220\uc744 \uc0c8\ub85c\uc6b4 \ud658\uacbd\uc5d0 \uc801\uc6a9\ud558\ub294 \uacfc\uc815\uc744 \uc758\ubbf8\ud569\ub2c8\ub2e4. \uc774 \ubc29\ubc95\uc740 \uae30\uc874\uc758 \ud559\uc2b5 \ub370\uc774\ud130\ub97c \ud65c\uc6a9\ud558\uc5ec \uc0c8\ub85c\uc6b4 \ud658\uacbd\uc5d0\uc11c\uc758 \ud559\uc2b5 \uc18d\ub3c4\ub97c \ud5a5\uc0c1\uc2dc\ud0a4\uace0, \ub370\uc774\ud130 \ubd80\uc871 \ubb38\uc81c\ub97c \ud574\uacb0\ud558\ub294 \ub370 \uc720\uc6a9\ud569\ub2c8\ub2e4. \uc804\uc774 \ud559\uc2b5\uc740 \uc8fc\ub85c \ub450 \uac00\uc9c0 \ubc29\uc2dd\uc73c\ub85c \uc774\ub8e8\uc5b4\uc9d1\ub2c8\ub2e4: \ud30c\ub77c\ubbf8\ud130 \uc804\uc774\uc640 \uad6c\uc870 \uc804\uc774\uc785\ub2c8\ub2e4.<\/p>\n
\uc774\uc81c \uad6c\uccb4\uc801\uc778 \uc0ac\ub840\ub97c \ub4e4\uc5b4 \ud658\uacbd \ud655\uc7a5\uacfc \uc804\uc774 \ud559\uc2b5\uc774 \uc5b4\ub5bb\uac8c \ud65c\uc6a9\ub420 \uc218 \uc788\ub294\uc9c0 \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
\ub85c\ubd07 \ud314\uc744 \ud6c8\ub828\ud560 \ub54c, \ucd08\uae30 \ud658\uacbd\uc5d0\uc11c\ub294 \ub2e8\uc21c\ud55c \ubb3c\uccb4\ub97c \uc7a1\ub294 \ubaa9\ud45c\ub9cc \uc788\uc5c8\uc744 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uadf8\ub7ec\ub098 \ud658\uacbd\uc744 \ud655\uc7a5\ud558\uc5ec \ub2e4\uc591\ud55c \ubb3c\uccb4\uc758 \uc704\uce58, \uac01\ub3c4, \ud06c\uae30 \ubc0f \ud615\ud0dc\ub97c \ud3ec\ud568\ud558\ub294 \ubcf5\uc7a1\ud55c \uc2dc\ub098\ub9ac\uc624\ub97c \ud3ec\ud568\ud55c\ub2e4\uba74, \ub85c\ubd07 \ud314\uc740 \ubcf4\ub2e4 \uc2e4\uc81c \uc0c1\ud669\uc5d0\uc11c\uc758 \uc0ac\uc6a9 \uac00\ub2a5\uc131\uc744 \ub192\uc77c \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub807\uac8c \ubcf5\uc7a1\ud55c \ud658\uacbd\uc5d0\uc11c \ud6c8\ub828 \ud55c \ud6c4, \ub85c\ubd07 \ud314\uc740 \uc804\uc774 \ud559\uc2b5\uc744 \ud1b5\ud574 \ub2e4\ub978 \uc885\ub958\uc758 \ubb3c\uccb4\ub97c \uc870\uc791\ud558\ub294 \ub370 \ud544\uc694\ud55c \uae30\uc220\uc744 \uc27d\uac8c \ud559\uc2b5\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\uc790\uc728\uc8fc\ud589\ucc28 \uac1c\ubc1c\uc5d0\uc11c\ub3c4 \ud658\uacbd \ud655\uc7a5\uacfc \uc804\uc774 \ud559\uc2b5\uc774 \uc911\uc694\ud569\ub2c8\ub2e4. \uc790\uc728\uc8fc\ud589\ucc28\ub294 \ub2e4\uc591\ud55c \ub3c4\ub85c \uad50\ud1b5 \uc0c1\ud669\uc5d0 \uc9c1\uba74\ud558\uac8c \ub429\ub2c8\ub2e4. \ucd08\uae30 \ud559\uc2b5 \ub2e8\uacc4\uc5d0\uc11c \uc790\uc728\uc8fc\ud589\ucc28\ub294 \ub2e8\uc21c\ud55c \ub3c4\ub85c \ud658\uacbd\uc5d0\uc11c \ud6c8\ub828\ub418\uc5c8\uc9c0\ub9cc, \ud658\uacbd\uc744 \ud655\uc7a5\ud558\uc5ec \ub2e4\ub978 \uae30\ud6c4 \uc870\uac74, \uc57c\uac04 \ub4dc\ub77c\uc774\ube59, \ub2e4\uc591\ud55c \ucc28\ub7c9 \ubc0f \ubcf4\ud589\uc790 \ub4f1\uc744 \ud3ec\ud568\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub807\uac8c \ud658\uacbd\uc744 \ud655\uc7a5\ud55c \ud6c4, \uc790\uc728\uc8fc\ud589\ucc28\ub294 \uc0c8\ub85c\uc6b4 \ud658\uacbd\uc5d0\uc11c\ub3c4 \ud6a8\uacfc\uc801\uc73c\ub85c \ub300\uc751\ud560 \uc218 \uc788\ub294 \ub2a5\ub825\uc744 \ud0a4\uc6b8 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\uc774\uc81c \uac04\ub2e8\ud55c \ud658\uacbd \ud655\uc7a5 \ubc0f \uc804\uc774 \ud559\uc2b5\uc758 \uc608\uc81c\ub97c \ud1b5\ud574 \uc2e4\uc81c \uad6c\ud604\uc744 \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc5ec\uae30\uc11c\ub294 OpenAI\uc758 \uac15\ud654\ud559\uc2b5\uc5d0\uc11c\uc758 \ud658\uacbd \ud655\uc7a5 \ubc0f \uc804\uc774 \ud559\uc2b5\uc740 \uc5d0\uc774\uc804\ud2b8\uac00 \ubcf4\ub2e4 \ubcf5\uc7a1\ud55c \ubb38\uc81c\ub97c \ud574\uacb0\ud558\uace0, \ub2e4\uc591\ud55c \uc2dc\ub098\ub9ac\uc624\uc5d0\uc11c \ud559\uc2b5\ud55c \uacbd\ud5d8\uc744 \ud6a8\uacfc\uc801\uc73c\ub85c \uc7ac\uc0ac\uc6a9\ud558\ub294 \ub370 \ub9ce\uc740 \ub3c4\uc6c0\uc774 \ub429\ub2c8\ub2e4. \uc774\ub294 \ud2b9\ud788 \ud604\uc2e4 \uc138\uacc4\uc758 \ubcf5\uc7a1\uc131\uc744 \ubc18\uc601\ud558\uc5ec RL \ubaa8\ub378\uc744 \ub354 robust\ud558\uac8c \ub9cc\ub4e4\uace0, \ub370\uc774\ud130 \ud6a8\uc728\uc131\uc744 \ub192\uc774\ub294 \ub370 \uae30\uc5ec\ud569\ub2c8\ub2e4. \uc55e\uc73c\ub85c\ub3c4 \ub2e4\uc591\ud55c \ud658\uacbd\uc5d0\uc11c \uc774\ub7ec\ud55c \uae30\ubc95\uc744 \uc801\uc6a9\ud558\uc5ec \ubcf4\ub2e4 \ubc1c\uc804\ub41c AI \uc2dc\uc2a4\ud15c\uc744 \ub9cc\ub4e4 \uc218 \uc788\ub3c4\ub85d \ud574\uc57c \ud560 \uac83\uc785\ub2c8\ub2e4.<\/p>\n <\/body><\/p>\n","protected":false},"excerpt":{"rendered":"gym<\/code> \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc0ac\uc6a9\ud558\uc5ec \ud658\uacbd\uc744 \ud655\uc7a5\ud558\uace0, \uc774\uc804\uc5d0 \ud559\uc2b5\ud55c \uc815\ucc45\uc744 \uc804\uc774\ud558\ub294 \ubc29\uc2dd\uc744 \uc124\uba85\ud569\ub2c8\ub2e4.<\/p>\n\ucf54\ub4dc \uc608\uc81c<\/h3>\n
\n# \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac \uc784\ud3ec\ud2b8\nimport gym\nimport numpy as np\nimport random\n\n# \ud658\uacbd \uc0dd\uc131\nclass CustomEnv(gym.Env):\n def __init__(self):\n super(CustomEnv, self).__init__()\n self.action_space = gym.spaces.Discrete(2) # \ub450 \uac00\uc9c0 \ud589\ub3d9 \uc120\ud0dd (\uac00\uc18d, \uac10\uc18d)\n self.observation_space = gym.spaces.Discrete(5) # 5\uac1c\uc758 \uc0c1\ud0dc (\uc18d\ub3c4 \ubc94\uc704)\n self.state = 0\n self.done = False\n\n def reset(self):\n self.state = 0\n self.done = False\n return self.state\n\n def step(self, action):\n if action == 0: # \uac00\uc18d\n self.state = min(self.state + 1, 4)\n elif action == 1: # \uac10\uc18d\n self.state = max(self.state - 1, 0)\n\n reward = 1 if self.state == 4 else -1 # \ubaa9\ud45c \uc18d\ub3c4\uc5d0 \ub3c4\ub2ec\ud558\uba74 \ubcf4\uc0c1\n self.done = self.state == 4\n return self.state, reward, self.done, {}\n\n# \uc5d0\uc774\uc804\ud2b8 \ud074\ub798\uc2a4\nclass Agent:\n def __init__(self):\n self.q_table = np.zeros((5, 2)) # \uc0c1\ud0dc * \ud589\ub3d9\uc758 Q \ud14c\uc774\ube14\n\n def choose_action(self, state):\n return np.argmax(self.q_table[state]) # \ucd5c\ub300 Q \uac12\uc5d0 \ub530\ub978 \ud589\ub3d9 \uc120\ud0dd\n\n def update_q(self, state, action, reward, next_state):\n self.q_table[state, action] += 0.1 * (reward + 0.9 * np.max(self.q_table[next_state]) - self.q_table[state, action])\n\n# \ud658\uacbd \uc0dd\uc131 \ubc0f \ud559\uc2b5\nenv = CustomEnv()\nagent = Agent()\n\nfor episode in range(1000):\n state = env.reset()\n done = False\n\n while not done:\n action = agent.choose_action(state)\n next_state, reward, done, _ = env.step(action)\n agent.update_q(state, action, reward, next_state)\n state = next_state\n\n# \uae30\uc874 \ud658\uacbd\uc5d0\uc11c\uc758 \ud559\uc2b5 \ud6c4, \ud658\uacbd \ud655\uc7a5\nclass ExtendedEnv(CustomEnv):\n def __init__(self):\n super(ExtendedEnv, self).__init__()\n self.action_space = gym.spaces.Discrete(2) # \ud589\ub3d9\uc740 \uac19\uc74c\n self.observation_space = gym.spaces.Discrete(10) # \uc0c1\ud0dc\uac00 10\uc73c\ub85c \ud655\uc7a5\ub428\n self.state = 0\n\n def reset(self):\n return super().reset() # \uae30\ubcf8 reset \ud638\ucd9c\n\n def step(self, action):\n # \uc0c8\ub85c\uc6b4 \uc0c1\ud0dc \uc815\uc758 \ub85c\uc9c1 \ucd94\uac00\n pass # \uc0c1\ud0dc \uc804\uc774 \ub85c\uc9c1\uc744 \ud655\uc7a5\ud558\uc5ec \uad6c\ud604\ud574\uc57c \ud568\n\n# \uc0c8\ub85c\uc6b4 \ud658\uacbd\uc5d0\uc11c \uc774\uc804 \uc5d0\uc774\uc804\ud2b8\uac00 \uc804\uc774 \ud559\uc2b5\uc744 \uc218\ud589\ud558\ub294 \ucf54\ub4dc\nextended_env = ExtendedEnv()\n# \uc5ec\uae30\uc5d0 \uc804\uc774 \ud559\uc2b5 \uc218\ud589\n <\/code><\/pre>\n\uacb0\ub860<\/h2>\n