1234523767123

# train.py import numpy as np import pandas as pd import torch import torch.nn as nn from torch.utils.data import Dataset, DataLoader, random_split from sklearn.preprocessing import OneHotEncoder from model import FCNet import os import pickle import sklearn from packaging import version # 设置随机种子以确保结果可复现 torch.manual_seed(42) np.random.seed(42) # 自定义数据集 class AminoAcidDataset(Dataset): def __init__(self, single_csv, double_csv, triple_csv): # 读取 CSV 文件，指定没有表头并手动命名列 single = pd.read_csv(single_csv, header=None, names=['AminoAcid', 'Value']) double = pd.read_csv(double_csv, header=None, names=['AminoAcidPair', 'Value1', 'Value2']) triple = pd.read_csv(triple_csv, header=None, names=['AminoAcidTriple', 'Value1', 'Value2']) # 创建氨基酸到数值的映射字典 self.single_dict = single.set_index('AminoAcid')['Value'].to_dict() # 获取所有独热编码的氨基酸，加上一个填充字符 '-' self.amino_acids = sorted(self.single_dict.keys()) + ['-'] # 根据 scikit-learn 版本选择参数 if version.parse(sklearn.__version__) >= version.parse("1.2"): self.encoder = OneHotEncoder(sparse_output=False) # 新版本使用 sparse_output else: self.encoder = OneHotEncoder(sparse=False) # 旧版本使用 sparse self.encoder.fit(np.array(self.amino_acids).reshape(-1, 1)) # 准备数据 self.X = [] self.y = [] # 处理单个氨基酸，补齐为三个位置 for index, row in single.iterrows(): aa = row['AminoAcid'] if aa not in self.single_dict: print(f"警告：氨基酸 '{aa}' 不在单个氨基酸字典中，已跳过。") continue # 填充为三个位置，例如 'A--' aa_triplet = aa + '--' try: one_hot1 = self.encoder.transform([[aa_triplet[0]]])[0] one_hot2 = self.encoder.transform([[aa_triplet[1]]])[0] one_hot3 = self.encoder.transform([[aa_triplet[2]]])[0] except Exception as e: print(f"错误：无法对氨基酸 '{aa_triplet}' 进行编码，已跳过。错误信息：{e}") continue value1_aa1 = self.single_dict.get(aa_triplet[0], 0.0) value1_aa2 = self.single_dict.get(aa_triplet[1], 0.0) value1_aa3 = self.single_dict.get(aa_triplet[2], 0.0) feature_part1 = np.concatenate([one_hot1, [value1_aa1]]) feature_part2 = np.concatenate([one_hot2, [value1_aa2]]) feature_part3 = np.concatenate([one_hot3, [value1_aa3]]) feature = np.stack([feature_part1, feature_part2, feature_part3]) # (3, N + 1) self.X.append(feature) self.y.append([float(row['Value']), 0.0]) # 对于单个氨基酸，第二个值设为0.0 # 处理双氨基酸组合，补齐为三个位置 for index, row in double.iterrows(): aa_pair = row['AminoAcidPair'] if not isinstance(aa_pair, str): print(f"警告：第 {index} 行的氨基酸对不是字符串，值为 '{aa_pair}'，已跳过。") continue if len(aa_pair) != 2: print(f"警告：第 {index} 行的氨基酸对 '{aa_pair}' 不包含两个氨基酸，已跳过。") continue aa1, aa2 = aa_pair[0], aa_pair[1] if aa1 not in self.single_dict or aa2 not in self.single_dict: print(f"警告：氨基酸 '{aa1}' 或 '{aa2}' 不在单个氨基酸字典中，已跳过。") continue # 填充为三个位置，例如 'AC-' aa_triplet = aa1 + aa2 + '-' try: one_hot1 = self.encoder.transform([[aa_triplet[0]]])[0] one_hot2 = self.encoder.transform([[aa_triplet[1]]])[0] one_hot3 = self.encoder.transform([[aa_triplet[2]]])[0] except Exception as e: print(f"错误：无法对氨基酸 '{aa_triplet}' 进行编码，已跳过。错误信息：{e}") continue value1_aa1 = self.single_dict.get(aa_triplet[0], 0.0) value1_aa2 = self.single_dict.get(aa_triplet[1], 0.0) value1_aa3 = self.single_dict.get(aa_triplet[2], 0.0) feature_part1 = np.concatenate([one_hot1, [value1_aa1]]) feature_part2 = np.concatenate([one_hot2, [value1_aa2]]) feature_part3 = np.concatenate([one_hot3, [value1_aa3]]) feature = np.stack([feature_part1, feature_part2, feature_part3]) # (3, N + 1) self.X.append(feature) self.y.append([float(row['Value1']), float(row['Value2'])]) # 处理三个氨基酸组合 for index, row in triple.iterrows(): aa_triple = row['AminoAcidTriple'] if not isinstance(aa_triple, str): print(f"警告：第 {index} 行的氨基酸三联不是字符串，值为 '{aa_triple}'，已跳过。") continue if len(aa_triple) != 3: print(f"警告：第 {index} 行的氨基酸三联 '{aa_triple}' 不包含三个氨基酸，已跳过。") continue aa1, aa2, aa3 = aa_triple[0], aa_triple[1], aa_triple[2] if aa1 not in self.single_dict or aa2 not in self.single_dict or aa3 not in self.single_dict: print(f"警告：氨基酸 '{aa1}', '{aa2}' 或 '{aa3}' 不在单个氨基酸字典中，已跳过。") continue try: one_hot1 = self.encoder.transform([[aa1]])[0] one_hot2 = self.encoder.transform([[aa2]])[0] one_hot3 = self.encoder.transform([[aa3]])[0] except Exception as e: print(f"错误：无法对氨基酸 '{aa1}{aa2}{aa3}' 进行编码，已跳过。错误信息：{e}") continue value1_aa1 = self.single_dict.get(aa1, 0.0) value1_aa2 = self.single_dict.get(aa2, 0.0) value1_aa3 = self.single_dict.get(aa3, 0.0) feature_part1 = np.concatenate([one_hot1, [value1_aa1]]) feature_part2 = np.concatenate([one_hot2, [value1_aa2]]) feature_part3 = np.concatenate([one_hot3, [value1_aa3]]) feature = np.stack([feature_part1, feature_part2, feature_part3]) # (3, N + 1) self.X.append(feature) self.y.append([float(row['Value1']), float(row['Value2'])]) self.X = np.array(self.X, dtype=np.float32) # 形状：(样本数, 3, N + 1) self.y = np.array(self.y, dtype=np.float32) # 形状：(样本数, 2) def __len__(self): return len(self.X) def __getitem__(self, idx): return self.X[idx], self.y[idx] def train(): # 数据文件路径 single_csv = '../data/single.csv' double_csv = '../data/double.csv' triple_csv = '../data/triple.csv' # 检查文件是否存在 if not os.path.exists(single_csv): print(f"错误：单个氨基酸数据文件 '{single_csv}' 不存在。") return if not os.path.exists(double_csv): print(f"错误：组合氨基酸数据文件 '{double_csv}' 不存在。") return if not os.path.exists(triple_csv): print(f"错误：三联氨基酸数据文件 '{triple_csv}' 不存在。") return # 创建数据集 dataset = AminoAcidDataset(single_csv, double_csv, triple_csv) print(f"总样本数量：{len(dataset)}") if len(dataset) == 0: print("错误：数据集中没有有效的样本。请检查数据文件的内容和格式。") return # 数据集划分：80% 训练，20% 验证 train_size = int(0.8 * len(dataset)) val_size = len(dataset) - train_size train_dataset, val_dataset = rando