82 lines
2.2 KiB
Python
82 lines
2.2 KiB
Python
"""
|
||
药店销售预测系统 - 评估指标计算函数
|
||
"""
|
||
|
||
import numpy as np
|
||
from sklearn.metrics import mean_squared_error, mean_absolute_error, r2_score
|
||
|
||
def evaluate_model(y_true, y_pred):
|
||
"""
|
||
计算模型评估指标
|
||
|
||
参数:
|
||
y_true: 真实值
|
||
y_pred: 预测值
|
||
|
||
返回:
|
||
包含各种评估指标的字典
|
||
"""
|
||
# 确保输入是一维数组
|
||
y_true = np.array(y_true).flatten()
|
||
y_pred = np.array(y_pred).flatten()
|
||
|
||
# 计算均方误差 (MSE)
|
||
mse = mean_squared_error(y_true, y_pred)
|
||
|
||
# 计算均方根误差 (RMSE)
|
||
rmse = np.sqrt(mse)
|
||
|
||
# 计算平均绝对误差 (MAE)
|
||
mae = mean_absolute_error(y_true, y_pred)
|
||
|
||
# 计算决定系数 (R^2)
|
||
r2 = r2_score(y_true, y_pred)
|
||
|
||
# 计算平均绝对百分比误差 (MAPE)
|
||
# 避免除以零
|
||
mask = y_true != 0
|
||
if np.any(mask):
|
||
mape = np.mean(np.abs((y_true[mask] - y_pred[mask]) / y_true[mask])) * 100
|
||
else:
|
||
# 如果所有真实值都为0，无法计算MAPE，返回0
|
||
mape = 0.0
|
||
|
||
return {
|
||
'mse': mse,
|
||
'rmse': rmse,
|
||
'mae': mae,
|
||
'r2': r2,
|
||
'mape': mape
|
||
}
|
||
|
||
def compare_models(results_dict):
|
||
"""
|
||
比较多个模型的性能
|
||
|
||
参数:
|
||
results_dict: 字典，键为模型名称，值为评估指标字典
|
||
|
||
返回:
|
||
比较结果的字典
|
||
"""
|
||
comparison = {}
|
||
|
||
# 获取所有指标名称
|
||
metrics = list(next(iter(results_dict.values())).keys())
|
||
|
||
# 对每个指标，找出最佳模型
|
||
for metric in metrics:
|
||
if metric in ['mse', 'rmse', 'mae', 'mape']: # 这些指标越小越好
|
||
best_model = min(results_dict.items(), key=lambda x: x[1][metric])[0]
|
||
best_value = min(model[metric] for model in results_dict.values())
|
||
else: # r2 越大越好
|
||
best_model = max(results_dict.items(), key=lambda x: x[1][metric])[0]
|
||
best_value = max(model[metric] for model in results_dict.values())
|
||
|
||
comparison[metric] = {
|
||
'best_model': best_model,
|
||
'best_value': best_value,
|
||
'all_values': {model_name: results[metric] for model_name, results in results_dict.items()}
|
||
}
|
||
|
||
return comparison |