使用一些最基本的机器学习模型来做回归预测（材料能耗问题）.zip

#!/usr/bin/env python # -*- coding: utf-8 -*- """ @Project ：ML_project @Product_name ：PyCharm @File ：ann_aeo.py @Author ：RockJim @Date ：2023/8/4 10:22 @Description ：使用AEO优化算法来查找ANN的最优参数 @Version ：1.0 """ import os import numpy as np import torch from torch import optim from config.log_config import log from config.config import DATASET_ROOT_PATH, BASE_DATASET_NAME from dataset.data_load import load_data, data_processing from evaluate.evaluate import evaluate_prediction from model.ann.ann import ANNModel from model.optimizer.aeo import aeo_algorithm_original logger = log().getLogger(__name__) # 检查是否有可用的GPU，如果有则使用第一个GPU，否则使用CPU device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") # 根据参数范围生成初始参数向量 def generate_initial_params(param_ranges_): params = {} for param_name, (min_val, max_val) in param_ranges_.items(): if param_name == 'activation': params[param_name] = np.random.choice(max_val) elif isinstance(min_val, int) and isinstance(max_val, int): params[param_name] = np.random.randint(min_val, max_val + 1) elif isinstance(min_val, float) and isinstance(max_val, float): params[param_name] = np.random.uniform(min_val, max_val) elif isinstance(min_val, list): params[param_name] = [np.random.randint(min_val[i], max_val[i] + 1) for i in range(len(min_val))] elif isinstance(min_val, str): params[param_name] = np.random.choice(max_val) return params # 将参数向量映射回参数值 def map_params_back(params, param_ranges_): mapped_params = {} for param_name, (min_val, max_val) in param_ranges_.items(): if isinstance(min_val, int) and isinstance(max_val, int): mapped_params[param_name] = int(params[param_name]) elif isinstance(min_val, float) and isinstance(max_val, float): mapped_params[param_name] = float(params[param_name]) elif isinstance(min_val, list): mapped_params[param_name] = [int(val) for val in params[param_name]] elif isinstance(min_val, str): mapped_params[param_name] = params[param_name] return mapped_params class ANN_AEO: def __init__(self): logger.info( "初始化ANN（Artificial Neural Network）+AEO（Artificial Ecosystem-Based Optimization）模型，并加载训练数据集和测试集数据") # 加载数据集 x, y = load_data(os.path.join(os.getcwd(), DATASET_ROOT_PATH, BASE_DATASET_NAME)) # 数据预处理 x_train, x_test, y_train, y_test, scaler = data_processing(x, y) # 定义数据部分 self.x_train = x_train self.x_test = x_test self.y_train = y_train self.y_test = y_test # 输入输出维度部分 self.input_dim = self.x_train.shape[1] # self.output_dim = self.y_train.shape[1] self.output_dim = 1 logger.info("输入维度为：\t{}，输出维度为：\t{}".format(self.input_dim, self.output_dim)) # 参数的可选范围和格式定义 self.param_ranges = { 'learning_rate': (0.001, 0.1), # 学习率范围 'batch_size': (8, 128), # 批大小范围 'num_hidden_layers': (1, 6), # 隐藏层数范围 'activation': ['relu', 'tanh'], # 使用整数编码 激活函数选项 'hidden_dims': (8, 64), # 隐藏层维度范围，每层的最小和最大维度 } logger.info("参数的可选范围和格式定义：\t{}".format(self.param_ranges)) def objective_function(self, params, x, y): print("当前参数为：\t{}".format(params)) logger.info("当前参数为：\t{}".format(params)) learning_rate = params[0] # 'learning_rate' batch_size = params[1] # 'batch_size' num_hidden_layers = params[2] # 转换为整数 'num_hidden_layers' activation = self.param_ranges['activation'][params[3]] # hidden_dims = [int(np.interp(params[i], [0, 1], self.param_ranges[f'hidden_dims_{i - 4}'])) for i in # range(4, num_hidden_layers + 4)] hidden_dims = params[4] model = ANNModel(self.input_dim, hidden_dims, self.output_dim, activation).to(device) criterion = torch.nn.MSELoss() optimizer = optim.SGD(model.parameters(), lr=learning_rate) for epoch in range(1000): optimizer.zero_grad() inputs = torch.tensor(x, dtype=torch.float32, device=device) targets = torch.tensor(y, dtype=torch.float32, device=device) outputs = model(inputs) # loss = custom_loss(outputs.detach(), targets) loss = criterion(outputs, targets) loss.backward() optimizer.step() # 打印日志输出 logger.info(f'Epoch [{epoch + 1}/1000], Loss: {loss.item()}') print(f'Epoch [{epoch + 1}/1000], Loss: {loss.item()}') return loss.item() def forward(self): # 定义AEO算法参数 num_generations = 5 competition_factor = 0.5 logger.info("AEO算法迭代\t{}\t代，竞争因子为：\t{}".format(num_generations, competition_factor)) # 使用AEO算法来优化神经网络超参数 logger.info("使用AEO算法来优化神经网络超参数") best_params = aeo_algorithm_original(self.param_ranges, num_generations, competition_factor, lambda params: self.objective_function(params, self.x_train, self.y_train)) # 解析最优超参数并训练神经网络 logger.info("解析最优超参数:") best_learning_rate, best_batch_size, best_num_hidden_layers, best_activation_idx = best_params[:4] best_activation = self.param_ranges['activation'][best_activation_idx] # best_hidden_dims = best_params[4:best_num_hidden_layers + 4].astype(int) best_hidden_dims = best_params[4] logger.info("最优参数为：\t\nbest_learning_rate\tbest_batch_size\tbest_num_hidden_layers\tbest_activation\t" "best_hidden_dims".format( best_learning_rate, best_batch_size, best_num_hidden_layers, best_activation, best_hidden_dims )) best_model = ANNModel(self.input_dim, best_hidden_dims, self.output_dim, best_activation).to(device) criterion = torch.nn.MSELoss() best_optimizer = optim.SGD(best_model.parameters(), lr=best_learning_rate) for epoch in range(1000): best_optimizer.zero_grad() inputs = torch.tensor(self.x_train, dtype=torch.float32, device=device) targets = torch.tensor(self.y_train, dtype=torch.float32, device=device) outputs = best_model(inputs) loss = criterion(outputs, targets) loss.backward() best_optimizer.step() # 打印日志输出 logger.info(f'Epoch [{epoch + 1}/1000], Loss: {loss.item()}') print(f'Epoch [{epoch + 1}/1000], Loss: {loss.item()}') # 使用训练好的模型进行预测 logger.info("使用训练好的模型进行预测:") test_inputs = torch.tensor(self.x_test, dtype=torch.float32, device=device) # predictions = best_model(test_inputs).cpu().numpy() # y_test = torch.tensor(self.y_test, dtype=torch.float32, device=device) predictions = best_model(test_inputs).detach().numpy() print("最优学习率：", best_learning_rate) print("最优批大小：", best_batch_size) print("最优隐藏层数：", best_num_hidden_layers) print("最优激活函数：", best_activation) print("最优隐藏层维度：", best_hid