\n \uac15\ud654\ud559\uc2b5(Reinforcement Learning, RL)\uc740 \uba38\uc2e0\ub7ec\ub2dd\uc758 \ud55c \ubd84\uc57c\ub85c, \uc5d0\uc774\uc804\ud2b8\uac00 \ud658\uacbd\uacfc \uc0c1\ud638\uc791\uc6a9\ud558\uc5ec \ucd5c\uc801\uc758 \ud589\ub3d9\uc744 \ud559\uc2b5\ud558\ub3c4\ub85d \ud558\ub294 \ubc29\ubc95\ub860\uc785\ub2c8\ub2e4. \uac15\ud654\ud559\uc2b5\uc740 \uc8fc\uc5b4\uc9c4 \uc0c1\ud0dc\uc5d0\uc11c \ud2b9\uc815 \ud589\ub3d9\uc744 \uc120\ud0dd\ud558\uace0, \uadf8 \ud589\ub3d9\uc5d0 \ub300\ud55c \ubcf4\uc0c1\uc744 \ubc1b\uace0, \uc774\ud6c4\uc758 \uc120\ud0dd\uc5d0 \ubc18\uc601\ud574 \ub098\uac00\ub294 iterative\ud55c \uacfc\uc815\uc73c\ub85c \uc791\ub3d9\ud569\ub2c8\ub2e4. \uac15\ud654\ud559\uc2b5\uc758 \ud575\uc2ec\uc740 \ub204\uc801 \ubcf4\uc0c1\uc744 \uadf9\ub300\ud654\ud558\ub294 \uac83\uc785\ub2c8\ub2e4.\n<\/p>\n
\n \uac15\ud654\ud559\uc2b5 \uc2dc\uc2a4\ud15c\uc740 \uc8fc\ub85c \ub2e4\uc74c\uacfc \uac19\uc740 \uad6c\uc131 \uc694\uc18c\ub85c \uc774\ub8e8\uc5b4\uc838 \uc788\uc2b5\ub2c8\ub2e4:\n<\/p>\n
\n \uc5d0\uc774\uc804\ud2b8\ub294 \uac15\ud654\ud559\uc2b5\uc758 \uc8fc\uccb4\ub85c\uc11c \ud658\uacbd \ub0b4\uc5d0\uc11c \uc758\uc0ac\uacb0\uc815\uc744 \uc218\ud589\ud558\ub294 \uc874\uc7ac\uc785\ub2c8\ub2e4. \uc5d0\uc774\uc804\ud2b8\ub294 \ucc98\uc74c\uc5d0\ub294 \uc544\ubb34\ub7f0 \uc9c0\uc2dd\uc774 \uc5c6\ub294 \uc0c1\ud0dc\uc5d0\uc11c \uc2dc\uc791\ud558\uba70, \ud658\uacbd\uacfc\uc758 \uc0c1\ud638\uc791\uc6a9\uc744 \ud1b5\ud574 \uc815\ubcf4\ub97c \ud559\uc2b5\ud569\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc5d0\uc11c \uc5d0\uc774\uc804\ud2b8\ub294 \uc790\uc2e0\uc758 \uc815\ucc45\uc744 \uc5c5\ub370\uc774\ud2b8\ud558\uace0, \uc774\ub97c \ud1b5\ud574 \ub354 \ub098\uc740 \ud589\ub3d9\uc744 \uc120\ud0dd\ud560 \uc218 \uc788\uac8c \ub429\ub2c8\ub2e4. \uc5d0\uc774\uc804\ud2b8\ub294 \uc0c1\ud0dc\uc640 \ud589\ub3d9 \uc870\ud569\uc5d0 \ub300\ud55c \ud559\uc2b5\uc744 \ud1b5\ud574 \ud589\ub3d9\uc758 \uac00\uce58\ub97c \ud3c9\uac00\ud558\uac8c \ub429\ub2c8\ub2e4.\n<\/p>\n
\n \ud658\uacbd\uc740 \uc5d0\uc774\uc804\ud2b8\uac00 \uc0c1\ud638\uc791\uc6a9\ud558\ub294 \uc8fc\uccb4\ub85c, \uc5d0\uc774\uc804\ud2b8\uc758 \ud589\ub3d9\uc5d0 \ub530\ub77c \ubcc0\ud654\ud558\ub294 \uc2dc\uc2a4\ud15c\uc785\ub2c8\ub2e4. \ud658\uacbd\uc740 \uc0c1\ud0dc\uc640 \ubcf4\uc0c1\uc744 \uacb0\uc815\ud558\ub294 \uc5ed\ud560\uc744 \ud558\uba70, \uc5d0\uc774\uc804\ud2b8\uac00 \uc0c1\ud669\uc744 \uc778\uc2dd\ud558\uace0 \ud589\ub3d9\uc744 \uc120\ud0dd\ud558\ub294 \uae30\ubc18\uc774 \ub429\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, \uc790\uc728\uc8fc\ud589\ucc28\uc758 \uacbd\uc6b0 \ub3c4\ub85c, \uad50\ud1b5 \uc2e0\ud638, \ubcf4\ud589\uc790 \ub4f1\uc774 \ud658\uacbd\uc5d0 \ud574\ub2f9\ud569\ub2c8\ub2e4. \uc5d0\uc774\uc804\ud2b8\ub294 \uc774\ub7ec\ud55c \ud658\uacbd\uc5d0\uc11c \uc8fc\uc5b4\uc9c4 \uc0c1\ud0dc\ub97c \ubc14\ud0d5\uc73c\ub85c \ud589\ub3d9\uc744 \uc120\ud0dd\ud558\uace0, \uadf8 \uacb0\uacfc\uc5d0 \ub530\ub77c \uc0c8\ub85c\uc6b4 \uc0c1\ud0dc\uc640 \ubcf4\uc0c1\uc744 \ubc1b\uac8c \ub429\ub2c8\ub2e4.\n<\/p>\n
\n \uc0c1\ud0dc\ub294 \uc5d0\uc774\uc804\ud2b8\uac00 \ud658\uacbd\uc5d0\uc11c \ud604\uc7ac \uc704\uce58\ud558\uace0 \uc788\ub294 \ud2b9\uc815 \uc21c\uac04\uc758 \uc815\ubcf4\ub97c \uc758\ubbf8\ud569\ub2c8\ub2e4. \uc0c1\ud0dc\ub294 \ud658\uacbd\uc758 \ubcc0\uc218\ub85c\ubd80\ud130 \ub3c4\ucd9c\ub420 \uc218 \uc788\uc73c\uba70, \uac01 \uc0c1\ud0dc\ub294 \uc5d0\uc774\uc804\ud2b8\uc5d0\uac8c \uc81c\uacf5\ub418\ub294 \uc815\ubcf4\uc758 \uc9d1\ud569\uc785\ub2c8\ub2e4. \uac15\ud654\ud559\uc2b5\uc5d0\uc11c\ub294 \uc0c1\ud0dc\ub97c \uc815\uc758\ud558\ub294 \uac83\uc774 \ub9e4\uc6b0 \uc911\uc694\ud558\uba70, \uc0c1\ud0dc\uc758 \uc815\uc758\uac00 \uc798\ubabb\ub418\uba74 \uc5d0\uc774\uc804\ud2b8\ub294 \ucd5c\uc801\uc758 \uc804\ub7b5\uc744 \ud559\uc2b5\ud558\uc9c0 \ubabb\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, \uccb4\uc2a4\ub97c \ub450\ub294 \uc5d0\uc774\uc804\ud2b8\uc758 \uacbd\uc6b0, \uac8c\uc784\ud310\uc758 \ubaa8\ub4e0 \uae30\ubb3c\uc758 \ubc30\uce58\uac00 \uc0c1\ud0dc\ub85c \uc815\uc758\ub420 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n<\/p>\n
\n \ud589\ub3d9\uc740 \uc5d0\uc774\uc804\ud2b8\uac00 \ud2b9\uc815 \uc0c1\ud0dc\uc5d0\uc11c \uc120\ud0dd\ud560 \uc218 \uc788\ub294 \ub2e4\uc591\ud55c \ub300\uc548\uc785\ub2c8\ub2e4. \uac01 \ud589\ub3d9\uc740 \uc5d0\uc774\uc804\ud2b8\uc5d0\uac8c \uc601\ud5a5\uc744 \ubbf8\uce58\ubbc0\ub85c, \uc62c\ubc14\ub978 \ud589\ub3d9\uc744 \uc120\ud0dd\ud558\ub294 \uac83\uc774 \uc911\uc694\ud569\ub2c8\ub2e4. \uac15\ud654\ud559\uc2b5\uc5d0\uc11c \ud589\ub3d9\uc740 \uc5f0\uc18d\uc801\uc77c \uc218\ub3c4, \uc774\uc0b0\uc801\uc77c \uc218\ub3c4 \uc788\uc73c\uba70, \uc77c\ubc18\uc801\uc73c\ub85c \uc5d0\uc774\uc804\ud2b8\uc758 \uc815\ucc45\uc5d0 \ub530\ub77c \uacb0\uc815\ub429\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, \uac8c\uc784\uc5d0\uc11c \uc9c4\ud589\ud560 \uc218 \uc788\ub294 \ubaa8\ub4e0 \uc218\ub97c \ud589\ub3d9\uc73c\ub85c \uc815\uc758\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n<\/p>\n
\n \ubcf4\uc0c1\uc740 \uc5d0\uc774\uc804\ud2b8\uac00 \ud658\uacbd\uc5d0\uc11c \ud2b9\uc815 \ud589\ub3d9\uc744 \ucde8\ud55c \uacb0\uacfc\ub85c \ubc1b\ub294 \ud3c9\uac00\uc758 \ucc99\ub3c4\uc785\ub2c8\ub2e4. \ubcf4\uc0c1\uc740 \uc591\uc218\uc77c \uc218\ub3c4 \uc788\uace0 \uc74c\uc218\uc77c \uc218\ub3c4 \uc788\uc73c\uba70, \uc5d0\uc774\uc804\ud2b8\ub294 \ubcf4\uc0c1\uc744 \ucd5c\ub300\ud654\ud558\uae30 \uc704\ud574 \ud589\ub3d9\uc744 \uc9c0\uc18d\uc801\uc73c\ub85c \uc870\uc815\ud569\ub2c8\ub2e4. \ubcf4\uc0c1\uc740 \ud559\uc2b5\uc758 \ubaa9\ud45c\uac00 \ub418\uba70, \uc5d0\uc774\uc804\ud2b8\uac00 \uc120\ud0dd\ud55c \ud589\ub3d9\uc758 \uacb0\uacfc\ub97c \ud3c9\uac00\ud558\ub294 \uc911\uc694\ud55c \uc694\uc18c\uc785\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, \uac8c\uc784\uc5d0\uc11c \uc2b9\ub9ac\ud588\uc744 \uacbd\uc6b0 \uae0d\uc815\uc801\uc778 \ubcf4\uc0c1\uc744 \uc8fc\uace0, \ud328\ubc30\ud560 \uacbd\uc6b0 \ubd80\uc815\uc801\uc778 \ubcf4\uc0c1\uc744 \uc8fc\ub294 \uc2dd\uc785\ub2c8\ub2e4.\n<\/p>\n
\n \uac15\ud654\ud559\uc2b5\uc758 \uacfc\uc815\uc740 \ub2e4\uc74c\uacfc \uac19\uc740 \uc21c\uc11c\ub85c \uc774\ub8e8\uc5b4\uc9d1\ub2c8\ub2e4:\n<\/p>\n
\n \ub2e4\uc74c\uc740 \ud30c\uc774\uc36c\uc744 \uc0ac\uc6a9\ud558\uc5ec \ubbf8\ub85c\uc5d0\uc11c \uc5d0\uc774\uc804\ud2b8\uac00 \ubaa9\ud45c\uc9c0\uc810\uc5d0 \ub3c4\ub2ec\ud558\ub3c4\ub85d \uac15\ud654\ud559\uc2b5\uc744 \uc801\uc6a9\ud558\ub294 \uac04\ub2e8\ud55c \uc608\uc81c\uc785\ub2c8\ub2e4. \uc774 \uc608\uc81c\uc5d0\uc11c\ub294 Q-learning \uc54c\uace0\ub9ac\uc998\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc5d0\uc774\uc804\ud2b8\uac00 \ucd5c\uc801\uc758 \ud589\ub3d9\uc744 \ud559\uc2b5\ud569\ub2c8\ub2e4.\n<\/p>\n
import numpy as np\nimport random\n\n# \ubbf8\ub85c \ud658\uacbd \uc815\uc758\nclass Maze:\n def __init__(self, maze):\n self.maze = maze\n self.start = (0, 0)\n self.goal = (len(maze) - 1, len(maze[0]) - 1)\n self.state = self.start\n\n def reset(self):\n self.state = self.start\n return self.state\n\n def step(self, action):\n if action == 0: # \uc0c1\n new_state = (max(self.state[0] - 1, 0), self.state[1])\n elif action == 1: # \ud558\n new_state = (min(self.state[0] + 1, len(self.maze) - 1), self.state[1])\n elif action == 2: # \uc88c\n new_state = (self.state[0], max(self.state[1] - 1, 0))\n else: # \uc6b0\n new_state = (self.state[0], min(self.state[1] + 1, len(self.maze[0]) - 1))\n\n if self.maze[new_state[0]][new_state[1]] == 1: # \ubcbd\n return self.state, -1, False # \ubcf4\uc0c1\uc740 -1, \uc0c1\ud0dc \ubcc0\ud558\uc9c0 \uc54a\uc74c\n\n self.state = new_state\n\n if self.state == self.goal:\n return self.state, 10, True # \ubaa9\ud45c \ub3c4\ub2ec, \ubcf4\uc0c1 10\n\n return self.state, -1, False # \uc77c\ubc18\uc801\uc778 \uacbd\uc6b0, \ubcf4\uc0c1 -1\n\n# Q-learning \uc5d0\uc774\uc804\ud2b8 \uc815\uc758\nclass QLearningAgent:\n def __init__(self, actions, alpha=0.1, gamma=0.9, epsilon=0.1):\n self.q_table = {}\n self.actions = actions\n self.alpha = alpha\n self.gamma = gamma\n self.epsilon = epsilon\n\n def get_action(self, state):\n if state not in self.q_table:\n self.q_table[state] = np.zeros(len(self.actions))\n\n if random.random() < self.epsilon:\n return random.choice(self.actions) # \ub79c\ub364 \uc561\uc158 \uc120\ud0dd\n else:\n return np.argmax(self.q_table[state]) # \ucd5c\uc801 \uc561\uc158 \uc120\ud0dd\n\n def update(self, state, action, reward, next_state):\n if next_state not in self.q_table:\n self.q_table[next_state] = np.zeros(len(self.actions))\n\n # Q-\uac12 \uc5c5\ub370\uc774\ud2b8\n best_next_action = np.argmax(self.q_table[next_state])\n td_target = reward + self.gamma * self.q_table[next_state][best_next_action]\n self.q_table[state][action] += self.alpha * (td_target - self.q_table[state][action])\n\n\nif __name__ == '__main__':\n # \ubbf8\ub85c \uc815\uc758 (0: \ud1b5\ub85c, 1: \ubcbd)\n maze = [\n [0, 0, 1, 0],\n [1, 0, 1, 0],\n [0, 0, 0, 0],\n [0, 1, 1, 0],\n ]\n env = Maze(maze)\n agent = QLearningAgent(actions=[0, 1, 2, 3]) # \uc0c1, \ud558, \uc88c, \uc6b0\n\n episodes = 1000\n for episode in range(episodes):\n state = env.reset()\n done = False\n\n while not done:\n action = agent.get_action(state)\n next_state, reward, done = env.step(action)\n agent.update(state, action, reward, next_state)\n state = next_state\n\n # Q-\ud14c\uc774\ube14 \ucd9c\ub825\n for state, q_values in agent.q_table.items():\n print(f\"\uc0c1\ud0dc: {state}, Q-\uac12: {q_values}\")\n<\/code><\/pre>\n\uacb0\ub860<\/h2>\n
\n \uac15\ud654\ud559\uc2b5\uc740 \uc5d0\uc774\uc804\ud2b8\uc640 \ud658\uacbd \uac04\uc758 \uc0c1\ud638\uc791\uc6a9\uc744 \ud1b5\ud574 \ucd5c\uc801\uc758 \ud589\ub3d9\uc744 \ud559\uc2b5\ud558\ub294 \uac15\ub825\ud55c \uae30\ubc95\uc785\ub2c8\ub2e4. \uc5d0\uc774\uc804\ud2b8\ub294 \ubcf4\uc0c1\uc744 \ucd5c\ub300\ud654\ud558\uae30 \uc704\ud574 \ud589\ub3d9\uc744 \uc870\uc815\ud558\uba70, \uc774 \uacfc\uc815\uc5d0\uc11c \ub2e4\uc591\ud55c \uc54c\uace0\ub9ac\uc998\uacfc \uae30\uc220\uc774 \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \ubcf8 \uae00\uc5d0\uc11c\ub294 \uac15\ud654\ud559\uc2b5\uc758 \uc8fc\uc694 \uc6a9\uc5b4\uc640 \uad6c\uc131 \uc694\uc18c\uc5d0 \ub300\ud574 \uc0b4\ud3b4\ubcf4\uc558\uc73c\uba70, \ub610\ud55c \uac04\ub2e8\ud55c Q-learning \uc608\uc81c\ub97c \ud1b5\ud574 \uc774\ub97c \uc774\ud574\ud558\ub294 \ub370 \ub3c4\uc6c0\uc774 \ub418\uc5c8\uae30\ub97c \ubc14\ub78d\ub2c8\ub2e4.\n<\/p>\n
<\/body><\/p>\n","protected":false},"excerpt":{"rendered":"