\n \uac15\ud654 \ud559\uc2b5(Reinforcement Learning, RL)\uc740 \uc5d0\uc774\uc804\ud2b8\uac00 \ud658\uacbd\uacfc \uc0c1\ud638\uc791\uc6a9\ud558\uba70 \ucd5c\uc801\uc758 \ud589\ub3d9\uc744 \ud559\uc2b5\ud558\ub294 \uae30\uacc4 \ud559\uc2b5\uc758 \ud55c \ubd84\uc57c\uc785\ub2c8\ub2e4. \uc774\ubc88 \uae00\uc5d0\uc11c\ub294 \n \ubc30\uace0\ud514 \uad00\ub9ac \uc2dc\uc2a4\ud15c\uc740 \uc0dd\ubb3c\ud559\uc801 \uc694\uc18c\uc640 \uc678\ubd80 \ud658\uacbd\uc758 \ubcc0\ud654\uc5d0 \uae30\ubc18\ud558\uc5ec \uc5d0\uc774\uc804\ud2b8\uac00 \uc2dd\uc0ac\ub97c \uad00\ub9ac\ud558\ub294 \uac83\uc744 \ubaa9\ud45c\ub85c \ud569\ub2c8\ub2e4. \uc774\ub7ec\ud55c \uc2dc\uc2a4\ud15c\uc740 \n \ubc30\uace0\ud514 \uad00\ub9ac \uc5d0\uc774\uc804\ud2b8\uc758 \uc0c1\ud0dc \uacf5\uac04\uc740 \uc5d0\uc774\uc804\ud2b8\uc758 \ud604\uc7ac \uc0c1\ud0dc\ub97c \ub098\ud0c0\ub0b4\uba70, \uc774\ub294 \ub2e4\uc74c\uacfc \uac19\uc740 \uc694\uc18c\ub85c \uad6c\uc131\ub420 \uc218 \uc788\uc2b5\ub2c8\ub2e4:\n <\/p>\n \n \uc5d0\uc774\uc804\ud2b8\ub294 \ub2e4\uc74c\uacfc \uac19\uc740 \ud589\ub3d9\uc744 \ucde8\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4:\n <\/p>\n \n \uc0c1\ud0dc\uc640 \ud589\ub3d9 \uacf5\uac04\uc744 \uc815\uc758\ud568\uc73c\ub85c\uc368 \uc5d0\uc774\uc804\ud2b8\ub294 \uae30\ub2a5\uc801\uc774\uace0 \uc758\ubbf8 \uc788\ub294 \uc120\ud0dd\uc744 \ud560 \uc218 \uc788\uc73c\uba70, \ud658\uacbd\uacfc\uc758 \uc0c1\ud638\uc791\uc6a9\uc744 \ud1b5\ud574 \n \ubc30\uace0\ud508 \uad00\ub9ac \uc5d0\uc774\uc804\ud2b8\uc758 \ud658\uacbd\uc740 \uac15\ud654 \ud559\uc2b5\uc758 OpenAI Gym \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc774 \ud658\uacbd\uc740 \uc5d0\uc774\uc804\ud2b8\uac00 \uc0c1\ud0dc\ub97c \uad00\ucc30\ud558\uace0 \ud589\ub3d9\uc744 \ucde8\ud560 \uc218 \uc788\ub294 \n \uc5d0\uc774\uc804\ud2b8\ub97c \uc124\uacc4\ud558\uae30 \uc704\ud574 \uac15\ud654 \ud559\uc2b5 \uc54c\uace0\ub9ac\uc998 \uc911 \ud558\ub098\uc778 Q-Learning<\/strong>\uc744 \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. Q-Learning \uc54c\uace0\ub9ac\uc998\uc740 \n \uc5d0\uc774\uc804\ud2b8\uac00 \ud658\uacbd\uacfc \uc0c1\ud638\uc791\uc6a9\ud558\uba74\uc11c \ud559\uc2b5\ud558\ub3c4\ub85d \ud6c8\ub828\ud558\ub294 \ucf54\ub4dc\ub97c \uc791\uc131\ud574 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \n \uc5d0\uc774\uc804\ud2b8 \ud559\uc2b5\uc774 \uc644\ub8cc\ub41c \ud6c4, \uc131\ub2a5\uc744 \ud3c9\uac00\ud558\uae30 \uc704\ud574 \uc5d0\uc774\uc804\ud2b8\ub97c \uc5ec\ub7ec \ubc88 \uc2e4\ud589\ud558\uc5ec \ubc30\uace0\ud514 \uc218\uc900\uc774 0\uc5d0 \ub3c4\ub2ec\ud558\ub294 \n \ubc30\uace0\ud514 \uad00\ub9ac \uc5d0\uc774\uc804\ud2b8\ub97c \uac1c\ubc1c\ud558\uba74\uc11c \uac15\ud654 \ud559\uc2b5\uc758 \uae30\ubcf8 \uac1c\ub150\uacfc \ud658\uacbd \uc124\uacc4, \uc5d0\uc774\uc804\ud2b8 \uc124\uacc4, <\/body><\/p>\n","protected":false},"excerpt":{"rendered":"
\n \ubc30\uace0\ud514 \uad00\ub9ac \uc5d0\uc774\uc804\ud2b8<\/strong>\ub97c \ub9cc\ub4e4\uae30 \uc704\ud55c \uc0ac\ub840\ub97c \uc18c\uac1c\ud569\ub2c8\ub2e4. \uc774 \uc5d0\uc774\uc804\ud2b8\ub294 \ud604\uc2e4 \uc138\uacc4\uc5d0\uc11c\uc758 \ubcf5\uc7a1\ud55c \ud589\ub3d9 \ubb38\uc81c\ub97c \ud574\uacb0\ud558\uae30 \uc704\ud574 \uac15\ud654 \ud559\uc2b5\uc744
\n \ud65c\uc6a9\ud558\uba70, \uc0c1\ud0dc\uc640 \ud589\ub3d9 \uacf5\uac04\uc744 \ud6a8\uacfc\uc801\uc73c\ub85c \uc124\uacc4\ud558\uc5ec \ubaa9\ud45c\ub97c \ub2ec\uc131\ud560 \uc218 \uc788\ub3c4\ub85d \ud569\ub2c8\ub2e4.\n <\/p>\n\ubc30\uacbd \uc9c0\uc2dd<\/h2>\n
\n \ud2b9\ud788 \uc778\uac04\uacfc \uac19\uc740 \ubcf5\uc7a1\ud55c \ud589\ub3d9\uc744 \ubaa8\ub378\ub9c1\ud558\ub294 \ub370 \uc801\ud569\ud569\ub2c8\ub2e4. \ud589\ub3d9 \ud559\uc2b5\uc5d0\uc11c \uace0\ub824\ud574\uc57c \ud560 \uc8fc\uc694 \uc694\uc18c\ub294
\n \uc0c1\ud0dc(state)<\/strong>\uc640 \ud589\ub3d9(action)<\/strong>\uc785\ub2c8\ub2e4. \uc5d0\uc774\uc804\ud2b8\ub294 \uc8fc\uc5b4\uc9c4 \uc0c1\ud0dc\uc5d0\uc11c \uc5b4\ub5a4 \ud589\ub3d9\uc744 \uc120\ud0dd\ud558\uace0,
\n \uadf8 \ud589\ub3d9\uc774 \uc120\ud0dd\ub41c \ud6c4\uc5d0 \uc0c8\ub85c\uc6b4 \uc0c1\ud0dc\ub85c \uc804\uc774\ub418\ub294 \ud658\uacbd\uc5d0\uc11c \ud559\uc2b5\ud558\uac8c \ub429\ub2c8\ub2e4.\n <\/p>\n\uc0c1\ud0dc\uc640 \ud589\ub3d9 \uacf5\uac04 \uc124\uacc4<\/h2>\n
\n
\n
\n \ucd5c\uc801\uc758 \uc815\ucc45(.policy)\uc744 \ud559\uc2b5\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n\ud658\uacbd \uc124\uc815<\/h2>\n
\n \uc778\ud130\ud398\uc774\uc2a4\ub97c \uc81c\uacf5\ud569\ub2c8\ub2e4. \uc774\ubc88\uc5d0 \uad6c\ucd95\ud560 \ud658\uacbd\uc758 \ucd08\uae30 \ucf54\ub4dc\ub97c \uc544\ub798\uc640 \uac19\uc774 \uc124\uc815\ud574 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4.\n <\/p>\nimport gym\nfrom gym import spaces\nimport numpy as np\n\nclass HungerManagementEnv(gym.Env):\n def __init__(self):\n super(HungerManagementEnv, self).__init__()\n self.action_space = spaces.Discrete(3) # 3\uac00\uc9c0 \ud589\ub3d9: \uc74c\uc2dd\uc744 \uba39\ub2e4, \uc6b4\ub3d9\ud558\ub2e4, \uc26c\ub2e4\n self.observation_space = spaces.Box(low=0, high=10, shape=(3,), dtype=np.float32) # \ubc30\uace0\ud514 \uc218\uc900, \uc5d0\ub108\uc9c0\uc6d0, \ud65c\ub3d9 \uc218\uc900\n self.state = None\n\n def reset(self):\n self.state = np.array([10, 0, 0], dtype=np.float32) # \ucd08\uae43\uac12: \ub9e4\uc6b0 \ubc30\uace0\ud514\n return self.state\n\n def step(self, action):\n hunger_level, energy_source, activity_level = self.state\n if action == 0: # \uc74c\uc2dd\uc744 \uba39\ub2e4\n hunger_level = max(0, hunger_level - 5) # \ubc30\uace0\ud514 \uac10\uc18c\n energy_source += 1 # \uc5d0\ub108\uc9c0\uc6d0 \uc99d\uac00\n elif action == 1: # \uc6b4\ub3d9\ud558\ub2e4\n hunger_level = min(10, hunger_level + 2) # \ubc30\uace0\ud514 \uc99d\uac00\n activity_level += 1 # \ud65c\ub3d9 \uc218\uc900 \uc99d\uac00\n elif action == 2: # \uc26c\ub2e4\n hunger_level = min(10, hunger_level + 1) # \ubc30\uace0\ud514 \uc644\ud654\n\n self.state = np.array([hunger_level, energy_source, activity_level], dtype=np.float32)\n reward = -hunger_level # \ubc30\uace0\ud514\uc774 \ub0ae\uc744\uc218\ub85d \uc88b\uc740 \ubcf4\uc0c1\n done = hunger_level <= 0 # \ubc30\uace0\ud514 \uc218\uc900\uc774 0\uc774\uba74 \uc885\ub8cc\n\n return self.state, reward, done, {}\n\n def render(self, mode='human'):\n print(f\"Hunger Level: {self.state[0]}, Energy Source: {self.state[1]}, Activity Level: {self.state[2]}\") \n <\/code><\/pre>\n\uc5d0\uc774\uc804\ud2b8 \uc124\uacc4<\/h2>\n
\n \uac01 \uc0c1\ud0dc\uc5d0\uc11c \uac00\ub2a5\ud55c \ud589\ub3d9\uc5d0 \ub300\ud55c \uac00\uce58\ub97c \ud559\uc2b5\ud558\uc5ec \uc5d0\uc774\uc804\ud2b8\uac00 \ucd5c\uc801\uc758 \uc815\ucc45\uc744 \uc120\ud0dd\ud558\ub3c4\ub85d \ud569\ub2c8\ub2e4. \uc5d0\uc774\uc804\ud2b8\uc758 \uae30\ubcf8 \uad6c\uc870\ub294 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4.\n <\/p>\nimport random\n\nclass QLearningAgent:\n def __init__(self, action_space):\n self.action_space = action_space\n self.q_table = {}\n self.learning_rate = 0.1\n self.discount_factor = 0.95\n self.exploration_rate = 1.0\n self.exploration_decay = 0.99\n\n def get_action(self, state):\n if random.uniform(0, 1) < self.exploration_rate:\n return self.action_space.sample() # \ubb34\uc791\uc704 \ud589\ub3d9 \uc120\ud0dd\n return np.argmax(self.q_table.get(tuple(state), np.zeros(self.action_space.n))) # \ucd5c\uc801 \ud589\ub3d9 \uc120\ud0dd\n\n def learn(self, state, action, reward, next_state):\n current_q = self.q_table.get(tuple(state), np.zeros(self.action_space.n))[action]\n max_future_q = np.max(self.q_table.get(tuple(next_state), np.zeros(self.action_space.n)))\n new_q = (1 - self.learning_rate) * current_q + self.learning_rate * (reward + self.discount_factor * max_future_q)\n if tuple(state) not in self.q_table:\n self.q_table[tuple(state)] = np.zeros(self.action_space.n)\n self.q_table[tuple(state)][action] = new_q\n\n self.exploration_rate *= self.exploration_decay # \ud0d0\uc0c9\ub960 \uac10\uc18c\n <\/code><\/pre>\n\ud6c8\ub828 \ubc0f \ud559\uc2b5<\/h2>\n
\n \uc5d0\ud53c\uc18c\ub4dc\ub97c \ubc18\ubcf5\ud558\uba74\uc11c \uc5d0\uc774\uc804\ud2b8\uac00 \ud589\ub3d9\uc744 \ud558\ub3c4\ub85d \ud558\uace0, \ubcf4\uc0c1\uc744 \uc81c\uacf5\ud558\uba70 Q-\ud14c\uc774\ube14\uc744 \uc5c5\ub370\uc774\ud2b8\ud569\ub2c8\ub2e4.\n <\/p>\ndef train_agent(env, agent, episodes):\n for episode in range(episodes):\n state = env.reset()\n done = False\n while not done:\n action = agent.get_action(state)\n next_state, reward, done, _ = env.step(action)\n agent.learn(state, action, reward, next_state)\n state = next_state\n if (episode + 1) % 100 == 0:\n print(f\"Episode: {episode + 1}, Q-Table Size: {len(agent.q_table)}\")\n <\/code><\/pre>\n\uacb0\uacfc \ubc0f \uc131\ub2a5 \ud3c9\uac00<\/h2>\n
\n \ud3c9\uade0 \uc5d0\ud53c\uc18c\ub4dc \uc218\ub97c \uce21\uc815\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc544\ub798 \ucf54\ub4dc\ub294 \uc131\ub2a5 \ud3c9\uac00\ub97c \uc704\ud55c \ubc29\ubc95\uc785\ub2c8\ub2e4.\n <\/p>\ndef evaluate_agent(env, agent, episodes):\n total_steps = 0\n for episode in range(episodes):\n state = env.reset()\n done = False\n steps = 0\n while not done:\n action = agent.get_action(state)\n next_state, reward, done, _ = env.step(action)\n state = next_state\n steps += 1\n total_steps += steps\n return total_steps \/ episodes\n\naverage_steps = evaluate_agent(env, agent, 100)\nprint(f\"Average Steps to Hunger Level 0: {average_steps}\")\n <\/code><\/pre>\n\uacb0\ub860<\/h2>\n
\n \uadf8\ub9ac\uace0 Q-Learning \uc54c\uace0\ub9ac\uc998\uc744 \ud1b5\ud574 \ubcf5\uc7a1\ud55c \ud589\ub3d9 \ubb38\uc81c\ub97c \ud574\uacb0\ud558\ub294 \ubc29\ubc95\uc744 \uc0b4\ud3b4\ubcf4\uc558\uc2b5\ub2c8\ub2e4.
\n \uc774 \uae00\uc5d0\uc11c \ub17c\uc758\ud55c \ub0b4\uc6a9\uc744 \ubc14\ud0d5\uc73c\ub85c \ub354 \ubcf5\uc7a1\ud55c \ud658\uacbd\uacfc \uc5d0\uc774\uc804\ud2b8\ub97c \uc124\uacc4\ud558\uace0 \uac1c\uc120\ud560 \uc218 \uc788\ub294
\n \uae30\ud68c\ub97c \ub9cc\ub4e4\uc5b4 \ubcf4\uae30\ub97c \ubc14\ub78d\ub2c8\ub2e4. \uac15\ud654 \ud559\uc2b5\uc740 \ud6cc\ub96d\ud55c \uac00\ub2a5\uc131\uc744 \uac00\uc9c4 \ubd84\uc57c\uc774\uba70, \ub2e4\uc591\ud55c \ubb38\uc81c\ub97c \ud574\uacb0\ud558\ub294 \ub370
\n \ud65c\uc6a9\ub420 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n