\uac15\ud654\ud559\uc2b5(Reinforcement Learning)\uc740 \uc5d0\uc774\uc804\ud2b8\uac00 \ud658\uacbd\uacfc\uc758 \uc0c1\ud638\uc791\uc6a9\uc744 \ud1b5\ud574 \ucd5c\uc801\uc758 \ud589\ub3d9\uc744 \ud559\uc2b5\ud558\ub294 \uae30\uacc4\ud559\uc2b5 \ubd84\uc57c\uc785\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc5d0\uc11c \uc911\uc694\ud55c \uac1c\ub150 \uc911 \ud558\ub098\uac00 \ud589\ub3d9 \uacf5\uac04(Action Space)<\/strong>\uc785\ub2c8\ub2e4. \ud589\ub3d9 \uacf5\uac04\uc740 \uc5d0\uc774\uc804\ud2b8\uac00 \uc218\ud589\ud560 \uc218 \uc788\ub294 \ubaa8\ub4e0 \uac00\ub2a5\ud55c \ud589\ub3d9\uc758 \uc9d1\ud569\uc744 \ub098\ud0c0\ub0b4\uba70, \uc774\ub97c \uc815\uc758\ud558\uace0 \ucd5c\uc801\ud654\ud558\ub294 \uacfc\uc815\uc740 \uac15\ud654\ud559\uc2b5 \uc54c\uace0\ub9ac\uc998\uc758 \uc131\ub2a5\uc5d0 \ud070 \uc601\ud5a5\uc744 \ubbf8\uce69\ub2c8\ub2e4.<\/p>\n \ud589\ub3d9 \uacf5\uac04\uc740 \uc5d0\uc774\uc804\ud2b8\uac00 \uc120\ud0dd\ud560 \uc218 \uc788\ub294 \ubaa8\ub4e0 \ud589\ub3d9\uc758 \uc9d1\ud569\uc785\ub2c8\ub2e4. \uc774 \uacf5\uac04\uc758 \uc885\ub958\uc5d0 \ub530\ub77c\uc11c \ub450 \uac00\uc9c0 \uc8fc\ub41c \uce74\ud14c\uace0\ub9ac\ub85c \ub098\ub20c \uc218 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\n \uc774\uc0b0\uc801 \ud589\ub3d9 \uacf5\uac04\uc740 \uc81c\ud55c\ub41c \uc218\uc758 \ud589\ub3d9\uc73c\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4. \uac01 \ud589\ub3d9\uc740 \ub3c5\ub9bd\uc801\uc778 \uc120\ud0dd\uc73c\ub85c \uac04\uc8fc\ub418\uba70, \uc608\ub97c \ub4e4\uc5b4 \ub2e4\uc74c\uacfc \uac19\uc740 \uc0c1\ud669\uc5d0\uc11c \ub098\ud0c0\ub0a0 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n \uc774\ub7f0 \uc0c1\ud669\uc5d0\uc11c \uc5d0\uc774\uc804\ud2b8\ub294 \uc120\ud0dd \uac00\ub2a5\ud55c \ubaa8\ub4e0 \ud589\ub3d9\uc744 \ubc30\uc5f4\ub85c \uc815\uc758\ud558\uace0, \uac01 \ud589\ub3d9\uc5d0 \ub300\ud55c \ubcf4\uc0c1\uc744 \uae30\ubc18\uc73c\ub85c \ucd5c\uc801\uc758 \ud589\ub3d9\uc744 \uacb0\uc815\ud569\ub2c8\ub2e4.<\/p>\n \uc5f0\uc18d\uc801 \ud589\ub3d9 \uacf5\uac04\uc5d0\uc11c\ub294 \ud589\ub3d9\uc774 \ubb34\ud55c\ud788 \ub9ce\uc740 \uac12\uc744 \uac00\uc9c8 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub294 \uc0c1\ud669\uc5d0 \ub530\ub77c \uc5bc\ub9c8\ub098 \ud589\ub3d9\uc744 \ubcc0\uacbd\ud560 \uc218 \uc788\ub294\uc9c0\uac00 \ub2e4\ub97c \uc218 \uc788\uc74c\uc744 \uc758\ubbf8\ud569\ub2c8\ub2e4.<\/p>\n \uc608\ub97c \ub4e4\uc5b4:<\/p>\n \uc774\ub7f0 \uacbd\uc6b0, \ud589\ub3d9 \uacf5\uac04\uc740 \uc2e4\uc218 \uac12\uc758 \ubc94\uc704\ub85c \uc815\uc758\ub418\uba70, \uac01 \ud589\ub3d9\uc5d0 \ub300\ud55c Q-value\ub97c \uacc4\uc0b0\ud560 \ub54c\ub294 \uc801\uc808\ud55c \ubc29\ubc95\uc744 \uc0ac\uc6a9\ud574\uc57c \ud569\ub2c8\ub2e4. \uc774\ub7ec\ud55c \uc5f0\uc18d\uc801 \ud589\ub3d9 \uacf5\uac04\uc5d0\uc11c \uc5d0\uc774\uc804\ud2b8\ub97c \ud559\uc2b5\uc2dc\ud0a4\uae30 \uc704\ud574 \uc77c\ubc18\uc801\uc73c\ub85c \uc0ac\uc6a9\ub418\ub294 \uc54c\uace0\ub9ac\uc998\uc740 \uc815\ucc45 \uacbd\uc0ac\ubc95 (Policy Gradient Methods)<\/strong>\uc785\ub2c8\ub2e4.<\/p>\n \ud589\ub3d9 \uacf5\uac04\uc744 \uc815\uc758\ud558\uace0 \ucd5c\uc801\ud654\ud558\ub294 \uacfc\uc815\uc740 \uac15\ud654\ud559\uc2b5\uc5d0\uc11c \uc5d0\uc774\uc804\ud2b8\uc758 \uc131\ub2a5\uc744 \ud06c\uac8c \uc88c\uc6b0\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ucd5c\uc801\ud654\ub41c \ud589\ub3d9 \uacf5\uac04\uc744 \uad6c\ucd95\ud558\uae30 \uc704\ud574 \uace0\ub824\ud574\uc57c \ud560 \uba87 \uac00\uc9c0 \ubc29\ubc95\uc740 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4:<\/p>\n \ud589\ub3d9 \uacf5\uac04\uc758 \uc815\uc758\uc640 \ucd5c\uc801\ud654\ub294 \uac15\ud654 \ud559\uc2b5\uc5d0\uc11c \ub9e4\uc6b0 \uc911\uc694\ud55c \uc694\uc18c\uc785\ub2c8\ub2e4. \uc774\uc0b0\uc801 \ud589\ub3d9 \uacf5\uac04\uacfc \uc5f0\uc18d\uc801 \ud589\ub3d9 \uacf5\uac04\uc5d0 \ub300\ud55c \uc774\ud574\ub294 \uc5ec\ub7ec \uac15\ud654\ud559\uc2b5 \uc54c\uace0\ub9ac\uc998\uc758 \uae30\ucd08\uac00 \ub418\uba70, \uac01 \uacf5\uac04\uc5d0 \uc801\ud569\ud55c \ubc29\ubc95\ub860\uc744 \ud1b5\ud574 \uc5d0\uc774\uc804\ud2b8\uc758 \uc131\ub2a5\uc744 \uadf9\ub300\ud654\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n \uc774 \uae00\uc5d0\uc11c\ub294 \ud589\ub3d9 \uacf5\uac04\uc758 \uac1c\ub150\uc744 \ud655\ub9bd\ud558\uace0, \uac01\uac01\uc758 \uc720\ud615\uc5d0\uc11c \uc5d0\uc774\uc804\ud2b8\uc758 \uad6c\ud604 \uc608\uc2dc\ub97c \ud1b5\ud574 \uc774\ud574\ub97c \ub3d5\uae30 \uc704\ud574 \ub178\ub825\ud588\uc2b5\ub2c8\ub2e4. \uc774\uc81c \ub3c5\uc790 \uc5ec\ub7ec\ubd84\uc774 \uc790\uc2e0\ub9cc\uc758 \uac15\ud654\ud559\uc2b5 \uc54c\uace0\ub9ac\uc998\uc744 \uac1c\ubc1c\ud558\ub294 \ub370 \ud544\uc694\ud55c \uae30\ubcf8 \uc9c0\uc2dd\uc744 \uac16\ucd94\uc5c8\uae30\ub97c \ubc14\ub78d\ub2c8\ub2e4.<\/p>\n <\/body><\/p>\n","protected":false},"excerpt":{"rendered":"\ud589\ub3d9 \uacf5\uac04(Action Space)\ub780 \ubb34\uc5c7\uc778\uac00?<\/h2>\n
\n
1. \uc774\uc0b0\uc801 \ud589\ub3d9 \uacf5\uac04 (Discrete Action Space)<\/h2>\n
\n
\uc608\uc81c: \uc774\uc0b0\uc801 \ud589\ub3d9 \uacf5\uac04\uc744 \uc0ac\uc6a9\ud558\ub294 \uac04\ub2e8\ud55c \uac15\ud654\ud559\uc2b5 \uc5d0\uc774\uc804\ud2b8<\/h3>\n
import numpy as np\nimport random\n\nclass SimpleDiscreteAgent:\n def __init__(self, actions):\n self.actions = actions\n self.Q = np.zeros(len(actions)) # Q-value \ucd08\uae30\ud654\n\n def select_action(self):\n return np.argmax(self.Q) # Q-value\uac00 \ucd5c\ub300\uc778 \ud589\ub3d9 \uc120\ud0dd\n\n def update_Q(self, action, reward):\n self.Q[action] += reward # \ubcf4\uc0c1\uc5d0 \ub530\ub77c Q-value \uc5c5\ub370\uc774\ud2b8\n\n# \ud589\ub3d9 \uacf5\uac04 \uc815\uc758\nactions = [\"\uc67c\ucabd\", \"\uc624\ub978\ucabd\", \"\uc704\", \"\uc544\ub798\"]\nagent = SimpleDiscreteAgent(actions)\n\n# \uc5d0\uc774\uc804\ud2b8 \ud589\ub3d9 \uc120\ud0dd \ubc0f Q-value \uc5c5\ub370\uc774\ud2b8 \uc608\uc2dc\nfor _ in range(10):\n action = agent.select_action()\n reward = random.choice([1, 0]) # \ub79c\ub364 \ubcf4\uc0c1\n agent.update_Q(action, reward)\n print(f\"Selected action: {actions[action]}, Updated Q-values: {agent.Q}\")\n<\/code><\/pre>\n2. \uc5f0\uc18d\uc801 \ud589\ub3d9 \uacf5\uac04 (Continuous Action Space)<\/h2>\n
\n
\uc608\uc81c: \uc5f0\uc18d\uc801 \ud589\ub3d9 \uacf5\uac04\uc744 \uc0ac\uc6a9\ud558\ub294 \uac15\ud654\ud559\uc2b5 \uc5d0\uc774\uc804\ud2b8<\/h3>\n
import numpy as np\n\nclass SimpleContinuousAgent:\n def __init__(self):\n self.theta = 0.0 # \ucd08\uae30 \uac01\ub3c4\n\n def select_action(self):\n return self.theta # \ud604\uc7ac \uac01\ub3c4\ub97c \ubc18\ud658\n\n def update_theta(self, action, reward):\n self.theta += 0.1 * reward # \ubcf4\uc0c1\uc5d0 \ub530\ub77c \uac01\ub3c4 \uc5c5\ub370\uc774\ud2b8\n\nagent = SimpleContinuousAgent()\n\n# \uc5d0\uc774\uc804\ud2b8 \ud589\ub3d9 \uc120\ud0dd \ubc0f \uc5c5\ub370\uc774\ud2b8 \uc608\uc2dc\nfor _ in range(10):\n action = agent.select_action()\n reward = np.random.randn() # \ub79c\ub364 \ubcf4\uc0c1\n agent.update_theta(action, reward)\n print(f\"Selected action (angle): {action}, Updated theta: {agent.theta}\")\n<\/code><\/pre>\n\ud589\ub3d9 \uacf5\uac04 \ucd5c\uc801\ud654<\/h2>\n
\n
\uacb0\ub860<\/h2>\n