当前位置：首页 > news >正文

Python动态规划：从基础到高阶优化的全面指南(3)

news 2025/7/29 16:40:01

七、动态规划性能优化实战

7.1 矩阵快速幂优化

def matrix_mult(A, B):"""矩阵乘法"""n = len(A)m = len(B[0])p = len(B)C = [[0]*m for _ in range(n)]for i in range(n):for k in range(p):if A[i][k]:for j in range(m):C[i][j] += A[i][k] * B[k][j]return Cdef matrix_power(matrix, power):"""矩阵快速幂"""n = len(matrix)# 初始化单位矩阵result = [[1 if i==j else 0 for j in range(n)] for i in range(n)]base = matrixwhile power:if power & 1:result = matrix_mult(result, base)base = matrix_mult(base, base)power //= 2return resultdef fib_matrix(n):"""O(log n)斐波那契数列"""if n < 2: return n# 转移矩阵T = [[1, 1],[1, 0]]# 初始状态 [F(1), F(0)]init = [[1], [0]]# 计算 T^(n-1)T_exp = matrix_power(T, n-1)result = matrix_mult(T_exp, init)return result[0][0]# 测试
print(fib_matrix(100))  # 输出：354224848179261915075

7.2 四边形不等式优化

def optimal_bst(keys, freq):"""最优二叉搜索树 - 四边形不等式优化"""n = len(keys)# dp[i][j] = keys[i..j]的最小搜索代价dp = [[0]*(n+1) for _ in range(n+1)]# 根节点记录root = [[0]*(n) for _ in range(n)]# 前缀和prefix = [0]*(n+1)for i in range(n):prefix[i+1] = prefix[i] + freq[i]# 初始化单个节点for i in range(n):dp[i][i] = freq[i]root[i][i] = i# L为子树长度for L in range(2, n+1):for i in range(n-L+1):j = i+L-1dp[i][j] = float('inf')# 利用四边形不等式缩小范围low = root[i][j-1] if j-1 >= i else ihigh = root[i+1][j] if i+1 <= j else jfor r in range(low, high+1):cost = dp[i][r-1] if r > i else 0cost += dp[r+1][j] if r < j else 0cost += prefix[j+1] - prefix[i]if cost < dp[i][j]:dp[i][j] = costroot[i][j] = rreturn dp[0][n-1]# 测试
keys = [10, 12, 20]
freq = [34, 8, 50]
print(optimal_bst(keys, freq))  # 输出：142

八、动态规划工程实践指南

8.1 测试框架设计

import unittestclass TestDP(unittest.TestCase):def test_knapsack(self):weights = [1, 2, 3]values = [6, 10, 12]capacity = 5self.assertEqual(knapsack_01(weights, values, capacity), 22)def test_edit_distance(self):self.assertEqual(min_edit_distance("kitten", "sitting"), 3)def test_stock_profit(self):prices = [7, 1, 5, 3, 6, 4]self.assertEqual(max_profit_k_transactions(prices, 2), 7)# 性能测试def test_performance(self):import timestart = time.time()result = fib_matrix(10000)duration = time.time() - startself.assertTrue(duration < 0.1, f"耗时过长: {duration:.4f}s")if __name__ == "__main__":unittest.main()

8.2 可视化调试工具

def visualize_dp_table(dp, title="DP Table"):"""可视化DP表格"""import matplotlib.pyplot as pltimport seaborn as snsplt.figure(figsize=(10, 8))sns.heatmap(dp, annot=True, fmt=".0f", cmap="YlGnBu")plt.title(title)plt.xlabel("Column")plt.ylabel("Row")plt.show()# 示例：网格最小路径和
grid = [[1,3,1],[1,5,1],[4,2,1]]
dp = min_path_sum(grid)  # 修改函数返回dp表
visualize_dp_table(dp)

8.3 动态规划代码生成器

def dp_code_generator(problem_type, params):"""动态规划代码模板生成器"""templates = {"knapsack": """
def knapsack(weights, values, capacity):n = len(weights)dp = [0] * (capacity+1)for i in range(n):for w in range(capacity, weights[i]-1, -1):dp[w] = max(dp[w], dp[w-weights[i]] + values[i])return dp[capacity]""","lcs": """
def longest_common_subsequence(text1, text2):m, n = len(text1), len(text2)dp = [[0]*(n+1) for _ in range(m+1)]for i in range(1, m+1):for j in range(1, n+1):if text1[i-1] == text2[j-1]:dp[i][j] = dp[i-1][j-1] + 1else:dp[i][j] = max(dp[i-1][j], dp[i][j-1])return dp[m][n]"""}return templates.get(problem_type, "# 未支持的动态规划类型")# 使用示例
print(dp_code_generator("knapsack", {}))

九、动态规划在AI领域的应用

9.1 强化学习中的值迭代

def value_iteration(grid, discount=0.9, theta=1e-3):"""网格世界值迭代算法"""actions = [(0,1), (1,0), (0,-1), (-1,0)]  # 上右下左states = [(i,j) for i in range(len(grid)) for j in range(len(grid[0]))]V = {s: 0 for s in states}while True:delta = 0for s in states:if grid[s[0]][s[1]] == 'G':  # 目标状态continuev = V[s]max_value = float('-inf')for a in actions:next_i, next_j = s[0]+a[0], s[1]+a[1]# 边界检查if not (0 <= next_i < len(grid) and 0 <= next_j < len(grid[0])):next_s = selse:next_s = (next_i, next_j)# 状态转移：确定环境reward = -1if grid[next_s[0]][next_s[1]] == 'G':reward = 100elif grid[next_s[0]][next_s[1]] == 'X':  # 障碍reward = -10next_s = svalue = reward + discount * V[next_s]if value > max_value:max_value = valueV[s] = max_valuedelta = max(delta, abs(v - V[s]))if delta < theta:breakreturn V# 测试网格
grid = [['S', ' ', ' ', 'X'],[' ', 'X', ' ', ' '],[' ', ' ', ' ', 'G']
]
values = value_iteration(grid)
for s, v in values.items():print(f"状态 {s}: 值 {v:.1f}")

9.2 序列预测中的维特比算法

def viterbi(obs, states, start_p, trans_p, emit_p):"""维特比算法 - 隐马尔可夫模型解码"""V = [{}]  # 路径概率表path = {}  # 路径记录表# 初始化初始状态for st in states:V[0][st] = start_p[st] * emit_p[st][obs[0]]path[st] = [st]# 递推计算for t in range(1, len(obs)):V.append({})new_path = {}for curr in states:# 计算最大概率路径(prob, state) = max((V[t-1][prev] * trans_p[prev][curr] * emit_p[curr][obs[t]], prev)for prev in states)V[t][curr] = probnew_path[curr] = path[state] + [curr]path = new_path# 回溯最优路径(prob, state) = max((V[len(obs)-1][st], st) for st in states)return prob, path[state]# 示例：天气预测
states = ('Sunny', 'Rainy')
obs = ('walk', 'shop', 'clean')  # 观测序列
start_p = {'Sunny': 0.6, 'Rainy': 0.4}
trans_p = {'Sunny': {'Sunny': 0.7, 'Rainy': 0.3},'Rainy': {'Sunny': 0.4, 'Rainy': 0.6}
}
emit_p = {'Sunny': {'walk': 0.1, 'shop': 0.4, 'clean': 0.5},'Rainy': {'walk': 0.6, 'shop': 0.3, 'clean': 0.1}
}prob, path = viterbi(obs, states, start_p, trans_p, emit_p)
print(f"最可能状态序列: {path}, 概率: {prob:.6f}")

十、动态规划的局限与挑战

维度灾难：
- 状态空间随维度指数增长
- 解决方案：近似动态规划、蒙特卡洛方法
连续状态空间：
- 离散化处理导致精度损失
- 解决方案：函数逼近、神经网络
模型不确定性：
- 状态转移概率未知
- 解决方案：Q学习、模型预测控制
实时性要求：
- 复杂问题计算时间长
- 解决方案：启发式剪枝、并行计算

# 应对维度灾难的近似方法示例
def approximate_dp(problem, state_repr, epsilon=0.1):"""近似动态规划框架"""value_fn = {}  # 值函数近似while True:max_diff = 0for s in problem.states:old_value = value_fn.get(state_repr(s), 0)# 计算新值new_value = max(problem.reward(s, a) + problem.discount * value_fn.get(state_repr(problem.transition(s, a)), 0)for a in problem.actions(s))# 更新值函数value_fn[state_repr(s)] = (1-epsilon)*old_value + epsilon*new_valuemax_diff = max(max_diff, abs(new_value - old_value))if max_diff < problem.theta:breakreturn value_fn