diff --git a/function.py b/function.py index bba6f81..b1f689a 100644 --- a/function.py +++ b/function.py @@ -1,132 +1,125 @@ -# function.py -import numpy as np -data_columns=[] id = "202210423230" name = "尹家立" -import pandas as pd +# 这里写自己的代码实现 import numpy as np -from sklearn.impute import SimpleImputer +from sklearn.svm import SVC +from sklearn.model_selection import train_test_split +from sklearn.metrics import classification_report, accuracy_score +from sklearn.preprocessing import StandardScaler -def load_data(data_path): - NAN_THRES=0.5 - # 读取数据并指定列 - horse_data = pd.read_csv(data_path, header=None, sep=r"\s+") - horse_data = horse_data.loc[:, :22] # 只取到死亡情况（前23列，索引0-22） - horse_data.columns = data_columns # 指定列名 - horse_data = horse_data.replace("?", float('NaN')) # 替换缺失值标记 +# 数据加载函数 +def load_data(filepath): + """ + 从文件中加载数据，若不自行正确实现，则无法通过评测，**不记录成绩**。 - # 删除无关或缺失值过多的特征 - col_del = ["Hospital Number"] - for col in horse_data.columns: - if horse_data[col].isna().sum() > len(horse_data) * NAN_THRES and col not in col_del: - col_del.append(col) - horse_data = horse_data.drop(columns=col_del) + Args: + - directory (str): 数据文件路径。 - # 删除标签缺失的样本 - horse_data = horse_data.dropna(subset=['outcome']).reset_index(drop=True) - - # 分离特征和标签 - x_data = horse_data.drop(columns=['outcome']) - y_data = horse_data[['outcome']] - - # 标签：保留三分类 - y_data['outcome'] = pd.to_numeric(y_data['outcome'], errors='coerce').astype('int32') - - # 分离离散和连续特征 - x_discrete = x_data[list(set(columns_discrete) - set(col_del))] - x_continue = x_data.drop(columns=x_discrete.columns) - - # 离散特征 one-hot 编码 - for col in col_oneHot: - if col in x_discrete.columns: - column_onehot = pd.get_dummies(x_discrete[col], prefix=col) - x_discrete = x_discrete.drop(columns=col) - x_discrete = pd.concat([x_discrete, column_onehot], axis=1) - - # 用众数填补离散特征缺失值 - imp = SimpleImputer(strategy='most_frequent') - x_discrete = pd.DataFrame(imp.fit_transform(x_discrete), columns=x_discrete.columns) - x_discrete = x_discrete.astype('int32') - - # 用均值填补连续特征缺失值并归一化 - for col in x_continue.columns: - x_continue[col] = pd.to_numeric(x_continue[col], errors='coerce').astype("float64") - x_continue[col] = x_continue[col].fillna(x_continue[col].mean()) - x_continue = (x_continue - x_continue.mean()) / (x_continue.max() - x_continue.min()) - - # 合并特征和标签 - horse_data = pd.concat([x_discrete, x_continue, y_data], axis=1) - horse_data = horse_data.sort_index(axis=1) # 排序列名，确保验证集和训练集一致 - - # 最终返回特征和标签 - return horse_data.drop(columns=['outcome']).values, horse_data['outcome'].values + Returns: + - np.array: 特征数据。 + - np.array: 标签数据。 + """ + data = [] + labels = [] + + with open(filepath, 'r') as file: + for line in file: + fields = line.strip().split() + + # 处理缺失值，用0填充 + fields = [0 if f == '?' else float(f) for f in fields] + + # 标签是倒数第6列 + label = int(fields[-6]) + + # 特征为其他列（不包含倒数第6列标签） + features = np.array(fields[: -6] + fields[-5:], dtype=float) + + # 标签映射：将 1->0, 2->1, 3->2 + if label == 1: + label = 0 + elif label == 2: + label = 1 + elif label == 3: + label = 2 + else: + print(f"无效标签，跳过此样本：{label}") + continue # 跳过无效标签 + + # 添加数据和标签 + data.append(features) + labels.append(label) + + # 转换为numpy数组 + data = np.array(data) + labels = np.array(labels) + + return data, labels - - - - -def train_svm(X, y, learning_rate=0.01, iterations=1000): +def train_svm(X, y, other_params=None): """ 支持向量机训练函数。 Args: - X (numpy array): 特征数据。 - y (numpy array): 标签数据。 - - learning_rate (float): 学习率 默认值为0.01。 - - iterations (int): 训练迭代次数 默认值为1000。 + - other_params (str): 其它参数，可选设置。若需设置其它参数，必须设置其默认值，否则将无法通过自动评测，成绩为零。 Returns: - - numpy array: 训练得到的权重。 + - SVC: 训练好的SVM模型。 """ - m, n = X.shape - weights = np.zeros(n) - - for _ in range(iterations): - for i in range(m): - if y[i] * np.dot(X[i], weights) < 1: - weights += learning_rate * (y[i] * X[i] + (-2 * (1 / iterations) * weights)) - else: - weights += learning_rate * (-2 * (1 / iterations) * weights) - return weights + # 使用高斯核（RBF核）训练SVM + model = SVC(kernel='rbf', C=1.0, gamma='scale', random_state=42) + model.fit(X, y) + + return model -def predict_svm(inX, weights): +def predict_svm(inX, model): """ 支持向量机预测函数。 Args: - inX (numpy array): 输入待分类的样本。 - - weights (numpy array): 训练得到的权重。 + - model (SVC): 训练得到的支持向量机模型。 Returns: - int: 预测的分类标签。 """ - return 1 if np.dot(inX, weights) >= 0 else -1 + return model.predict(inX.reshape(1, -1))[0] if __name__ == "__main__": - # 取得列信息 - with open("./data/horse-colic.names", 'r', encoding='utf-8') as f: - data_names = f.read() - data_columns = re.findall(r'(?<=[0-9]:).*', data_names) - data_columns = [col.strip() for col in data_columns[:-3]] - data_columns - - # 离散列 - columns_discrete = ['surgery?', 'Age', 'temperature of extremities', 'peripheral pulse', - 'mucous membranes', 'capillary refill time', 'pain - a subjective judgement of the horse\'s pain level', 'peristalsis', - 'abdominal distension', 'nasogastric tube', 'nasogastric reflux', 'rectal examination - feces', 'abdomen', 'abdominocentesis appearance'] - - # 独热编码列 - col_oneHot = ['surgery?', 'Age'] - # 加载训练集 + # 加载数据 X_train, y_train = load_data("./data/horse-colic.data") - # 加载测试集 X_test, y_test = load_data("./data/horse-colic.test") + + # 划分数据集 + X_train, X_test, y_train, y_test = train_test_split(X_train, y_train, test_size=0.2, random_state=42, stratify=y_train) + + # 数据标准化 + scaler = StandardScaler() + X_train = scaler.fit_transform(X_train) + X_test = scaler.transform(X_test) + # 训练支持向量机模型 - weights = train_svm(X_train, y_train, learning_rate=0.01, iterations=1000) + print("正在训练支持向量机模型...") + svm_model = train_svm(X_train, y_train) + + # 预测测试集 + y_pred = svm_model.predict(X_test) + + # 输出评估结果 + print("\n分类报告:") + print(classification_report(y_test, y_pred, target_names=["存活", "安乐死", "死亡"])) + + # 输出准确率 + accuracy = accuracy_score(y_test, y_pred) + print(f"测试集准确率: {accuracy * 100:.2f}%") + # 测试单个数据 - y_pred = predict_svm(X_test[11], weights) - print(f"预测结果: {y_pred}, 实际结果: {y_test[11]}") + test_sample = X_test[11] # 选择第12个样本（从0开始计数） + y_pred_single = predict_svm(test_sample, svm_model) + print(f"\n单个样本预测结果: {y_pred_single}, 实际结果: {y_test[11]}")