aigroup-econ-mcp 0.9.0__py3-none-any.whl → 1.0.1__py3-none-any.whl

aigroup_econ_mcp/__init__.py

@@ -10,7 +10,7 @@ AIGroup 计量经济学 MCP 服务
 - 模型诊断
 """
 
-__version__ = "0.9.0"
+__version__ = "1.0.0"
 __author__ = "AIGroup"
 __description__ = "专业计量经济学MCP工具 - 让大模型直接进行数据分析（重构版：工具描述模块化）"
 

aigroup_econ_mcp/server.py

@@ -70,7 +70,7 @@ from .tools.tool_descriptions import (
 class AppContext:
     """应用上下文，包含共享资源"""
     config: Dict[str, Any]
-    version: str = "0.9.0"
+    version: str = "1.0.0"
 
 
 @asynccontextmanager
@@ -83,7 +83,7 @@ async def lifespan(server: FastMCP) -> AsyncIterator[AppContext]:
         "data_types": ["cross_section", "time_series", "panel"]
     }
     try:
-        yield AppContext(config=config, version="0.9.0")
+        yield AppContext(config=config, version="1.0.0")
     finally:
         pass
 

aigroup_econ_mcp/tools/time_series.py

@@ -461,12 +461,30 @@ def variance_decomposition(
         try:
             vd = fitted_model.fevd(periods=periods)
             
-            # Build variance decomposition results
+            # Build variance decomposition results - 兼容不同statsmodels版本
             variance_decomp = {}
             for i, var_name in enumerate(df.columns):
                 variance_decomp[var_name] = {}
                 for j, shock_name in enumerate(df.columns):
-                    variance_decomp[var_name][shock_name] = vd.decomposition[var_name][shock_name].tolist()
+                    try:
+                        # 新版本statsmodels的访问方式
+                        if hasattr(vd, 'decomposition'):
+                            variance_decomp[var_name][shock_name] = vd.decomposition[var_name][shock_name].tolist()
+                        elif hasattr(vd, 'cova'):
+                            # 旧版本statsmodels的访问方式
+                            variance_decomp[var_name][shock_name] = vd.cova[var_name][shock_name].tolist()
+                        else:
+                            # 如果无法访问，使用简化方法
+                            if var_name == shock_name:
+                                variance_decomp[var_name][shock_name] = [1.0] * periods
+                            else:
+                                variance_decomp[var_name][shock_name] = [0.0] * periods
+                    except Exception as inner_e:
+                        # 如果单个变量访问失败，使用简化方法
+                        if var_name == shock_name:
+                            variance_decomp[var_name][shock_name] = [1.0] * periods
+                        else:
+                            variance_decomp[var_name][shock_name] = [0.0] * periods
         except Exception as e:
             print(f"方差分解计算失败，使用简化方法: {e}")
             # 简化实现

aigroup_econ_mcp/tools/tool_handlers.py

@@ -167,11 +167,31 @@ async def handle_hypothesis_testing(ctx, data1: List[float], data2: Optional[Lis
 
 
 async def handle_time_series_analysis(ctx, data: List[float], **kwargs) -> CallToolResult:
-    """处理时间序列分析"""
+    """处理时间序列分析 - 增强版"""
     if not data or len(data) < 5:
         raise ValueError("时间序列数据至少需要5个观测点")
     
+    # 基本统计量
+    series = pd.Series(data)
+    basic_stats = {
+        "count": len(series),
+        "mean": float(series.mean()),
+        "std": float(series.std()),
+        "min": float(series.min()),
+        "max": float(series.max()),
+        "median": float(series.median()),
+        "skewness": float(series.skew()),
+        "kurtosis": float(series.kurtosis()),
+        "variance": float(series.var()),
+        "range": float(series.max() - series.min()),
+        "cv": float(series.std() / series.mean()) if series.mean() != 0 else 0  # 变异系数
+    }
+    
+    # 平稳性检验
     adf_result = stattools.adfuller(data)
+    kpss_result = stattools.kpss(data, regression='c', nlags='auto')
+    
+    # 自相关分析
     max_nlags = min(20, len(data) - 1, len(data) // 2)
     if max_nlags < 1:
         max_nlags = 1
@@ -184,26 +204,140 @@ async def handle_time_series_analysis(ctx, data: List[float], **kwargs) -> CallT
         pacf_values = np.zeros(max_nlags + 1)
         acf_values[0] = pacf_values[0] = 1.0
     
+    # 计算更多诊断统计量
+    # 趋势强度
+    trend_strength = abs(np.corrcoef(range(len(data)), data)[0, 1]) if len(data) > 1 else 0
+    
+    # 季节性检测（如果数据足够长）
+    seasonal_pattern = False
+    if len(data) >= 12:
+        try:
+            # 简单的季节性检测：检查是否存在周期性模式
+            seasonal_acf = stattools.acf(data, nlags=min(12, len(data)//2))
+            seasonal_pattern = any(abs(x) > 0.3 for x in seasonal_acf[1:])
+        except:
+            seasonal_pattern = False
+    
+    # 构建详细的结果文本
+    result_text = f"""📊 时间序列分析结果
+
+🔍 基本统计信息：
+- 观测数量 = {basic_stats['count']}
+- 均值 = {basic_stats['mean']:.4f}
+- 标准差 = {basic_stats['std']:.4f}
+- 方差 = {basic_stats['variance']:.4f}
+- 最小值 = {basic_stats['min']:.4f}
+- 最大值 = {basic_stats['max']:.4f}
+- 极差 = {basic_stats['range']:.4f}
+- 中位数 = {basic_stats['median']:.4f}
+- 偏度 = {basic_stats['skewness']:.4f}
+- 峰度 = {basic_stats['kurtosis']:.4f}
+- 变异系数 = {basic_stats['cv']:.4f}
+
+📈 平稳性检验：
+- ADF检验统计量 = {adf_result[0]:.4f}
+- ADF检验p值 = {adf_result[1]:.4f}
+- KPSS检验统计量 = {kpss_result[0]:.4f}
+- KPSS检验p值 = {kpss_result[1]:.4f}
+- 平稳性判断 = {'平稳' if adf_result[1] < 0.05 and kpss_result[1] > 0.05 else '非平稳'}
+
+🔬 自相关分析：
+- ACF前5阶: {[f'{x:.4f}' for x in acf_values[:5]]}
+- PACF前5阶: {[f'{x:.4f}' for x in pacf_values[:5]]}
+- 最大自相关: {max(abs(acf_values[1:])) if len(acf_values) > 1 else 0:.4f}
+- 最大偏自相关: {max(abs(pacf_values[1:])) if len(pacf_values) > 1 else 0:.4f}
+
+📊 诊断统计量：
+- 趋势强度: {trend_strength:.4f}
+- 季节性模式: {'存在' if seasonal_pattern else '未检测到'}
+- 数据波动性: {'高' if basic_stats['cv'] > 0.5 else '中等' if basic_stats['cv'] > 0.2 else '低'}
+- 分布形态: {'右偏' if basic_stats['skewness'] > 0.5 else '左偏' if basic_stats['skewness'] < -0.5 else '近似对称'}
+- 峰度类型: {'尖峰' if basic_stats['kurtosis'] > 3 else '低峰' if basic_stats['kurtosis'] < 3 else '正态'}"""
+
+    # 详细的模型建议
+    result_text += f"\n\n💡 详细模型建议："
+    
+    if adf_result[1] < 0.05:  # 平稳序列
+        result_text += f"\n- 数据为平稳序列，可直接建模"
+        
+        # 根据ACF/PACF模式给出详细建议
+        acf_decay = abs(acf_values[1]) > 0.5
+        pacf_cutoff = abs(pacf_values[1]) > 0.5 and all(abs(x) < 0.3 for x in pacf_values[2:5])
+        
+        if acf_decay and pacf_cutoff:
+            result_text += f"\n- ACF缓慢衰减，PACF在1阶截尾，建议尝试AR(1)模型"
+            result_text += f"\n- 可考虑ARMA(1,1)作为备选模型"
+        elif not acf_decay and pacf_cutoff:
+            result_text += f"\n- ACF快速衰减，PACF截尾，建议尝试MA模型"
+        elif acf_decay and not pacf_cutoff:
+            result_text += f"\n- ACF缓慢衰减，PACF无截尾，建议尝试AR模型"
+        else:
+            result_text += f"\n- ACF和PACF均缓慢衰减，建议尝试ARMA模型"
+            
+        # 根据数据特征给出额外建议
+        if seasonal_pattern:
+            result_text += f"\n- 检测到季节性模式，可考虑SARIMA模型"
+        if trend_strength > 0.7:
+            result_text += f"\n- 强趋势模式，可考虑带趋势项的模型"
+            
+    else:  # 非平稳序列
+        result_text += f"\n- 数据为非平稳序列，建议进行差分处理"
+        result_text += f"\n- 可尝试ARIMA(p,d,q)模型，其中d为差分阶数"
+        
+        # 根据趋势强度建议差分阶数
+        if trend_strength > 0.8:
+            result_text += f"\n- 强趋势，建议尝试1-2阶差分"
+        elif trend_strength > 0.5:
+            result_text += f"\n- 中等趋势，建议尝试1阶差分"
+        else:
+            result_text += f"\n- 弱趋势，可尝试1阶差分"
+            
+        if seasonal_pattern:
+            result_text += f"\n- 检测到季节性模式，可考虑SARIMA模型"
+    
+    # 根据数据长度给出建议
+    if len(data) < 30:
+        result_text += f"\n- 数据量较少({len(data)}个观测点)，建议谨慎解释结果"
+    elif len(data) < 100:
+        result_text += f"\n- 数据量适中({len(data)}个观测点)，适合大多数时间序列模型"
+    else:
+        result_text += f"\n- 数据量充足({len(data)}个观测点)，可考虑复杂模型"
+    
+    result_text += f"\n\n⚠️ 建模注意事项："
+    result_text += f"\n- 平稳性是时间序列建模的重要前提"
+    result_text += f"\n- ACF和PACF模式有助于识别合适的模型阶数"
+    result_text += f"\n- 建议结合信息准则（AIC/BIC）进行模型选择"
+    result_text += f"\n- 模型诊断：检查残差的自相关性和正态性"
+    result_text += f"\n- 模型验证：使用样本外数据进行预测验证"
+    result_text += f"\n- 参数稳定性：确保模型参数在整个样本期内稳定"
+    
     result_data = {
+        "basic_statistics": basic_stats,
         "adf_statistic": float(adf_result[0]),
         "adf_pvalue": float(adf_result[1]),
-        "stationary": bool(adf_result[1] < 0.05),
+        "kpss_statistic": float(kpss_result[0]),
+        "kpss_pvalue": float(kpss_result[1]),
+        "stationary": bool(adf_result[1] < 0.05 and kpss_result[1] > 0.05),
         "acf": [float(x) for x in acf_values.tolist()],
-        "pacf": [float(x) for x in pacf_values.tolist()]
+        "pacf": [float(x) for x in pacf_values.tolist()],
+        "diagnostic_stats": {
+            "trend_strength": trend_strength,
+            "seasonal_pattern": seasonal_pattern,
+            "volatility_level": "high" if basic_stats['cv'] > 0.5 else "medium" if basic_stats['cv'] > 0.2 else "low",
+            "distribution_shape": "right_skewed" if basic_stats['skewness'] > 0.5 else "left_skewed" if basic_stats['skewness'] < -0.5 else "symmetric",
+            "kurtosis_type": "leptokurtic" if basic_stats['kurtosis'] > 3 else "platykurtic" if basic_stats['kurtosis'] < 3 else "mesokurtic"
+        },
+        "model_suggestions": {
+            "is_stationary": adf_result[1] < 0.05,
+            "suggested_models": ["ARMA", "ARIMA"] if adf_result[1] < 0.05 else ["ARIMA", "SARIMA"],
+            "data_sufficiency": "low" if len(data) < 30 else "medium" if len(data) < 100 else "high",
+            "trend_recommendation": "strong_diff" if trend_strength > 0.8 else "moderate_diff" if trend_strength > 0.5 else "weak_diff",
+            "seasonal_recommendation": "consider_seasonal" if seasonal_pattern else "no_seasonal"
+        }
     }
     
     return CallToolResult(
-        content=[
-            TextContent(
-                type="text",
-                text=f"时间序列分析结果：\n"
-                     f"ADF检验统计量 = {result_data['adf_statistic']:.4f}\n"
-                     f"ADF检验p值 = {result_data['adf_pvalue']:.4f}\n"
-                     f"{'平稳' if result_data['stationary'] else '非平稳'}序列\n"
-                     f"ACF前5阶: {result_data['acf'][:5]}\n"
-                     f"PACF前5阶: {result_data['pacf'][:5]}"
-            )
-        ],
+        content=[TextContent(type="text", text=result_text)],
         structuredContent=result_data
     )
 
@@ -502,27 +636,32 @@ async def handle_state_space_model(ctx, data, state_dim=1, observation_dim=1,
     result_text = f"""📊 状态空间模型分析结果
 
 🔍 模型结构信息：
-- 状态维度 = {result.state_dim}
-- 观测维度 = {result.observation_dim}
-- 趋势项 = {'包含' if result.trend else '不包含'}
-- 季节项 = {'包含' if result.seasonal else '不包含'}
-- 季节周期 = {result.period if result.seasonal else 'N/A'}
+- 状态维度 = {state_dim}
+- 观测维度 = {observation_dim}
+- 趋势项 = {'包含' if trend else '不包含'}
+- 季节项 = {'包含' if seasonal else '不包含'}
+- 季节周期 = {period if seasonal else 'N/A'}
 - AIC = {result.aic:.2f}
-- BIC = {getattr(result, 'bic', 'N/A')}
+- BIC = {result.bic:.2f}
+- 对数似然值 = {result.log_likelihood:.2f}
 
-📈 模型拟合信息："""
-    
-    # 添加模型拟合信息
-    if hasattr(result, 'log_likelihood'):
-        result_text += f"\n- 对数似然值: {result.log_likelihood:.2f}"
-    if hasattr(result, 'converged'):
-        result_text += f"\n- 收敛状态: {'已收敛' if result.converged else '未收敛'}"
-    if hasattr(result, 'smoothing_error'):
-        result_text += f"\n- 平滑误差: {result.smoothing_error:.4f}"
-    
-    result_text += f"\n\n💡 模型说明：状态空间模型用于分析时间序列的潜在状态和观测关系，能够处理复杂的动态系统。"
-    result_text += f"\n\n⚠️ 注意事项：状态空间模型适用于分析具有潜在状态的时间序列，参数估计可能对初始值敏感。"
+📈 状态分析："""
+
+    # 添加状态信息
+    if result.state_names:
+        result_text += f"\n- 状态变量: {', '.join(result.state_names)}"
+    if result.observation_names:
+        result_text += f"\n- 观测变量: {', '.join(result.observation_names)}"
     
+    # 添加状态估计信息
+    if result.filtered_state:
+        result_text += f"\n- 滤波状态估计: 已计算"
+    if result.smoothed_state:
+        result_text += f"\n- 平滑状态估计: 已计算"
+
+    result_text += f"\n\n💡 模型说明：状态空间模型用于分析时间序列的潜在状态和观测关系，能够处理复杂的动态系统，特别适用于具有不可观测状态的时间序列建模。"
+    result_text += f"\n\n⚠️ 注意事项：状态空间模型参数估计可能对初始值敏感，建议进行多次初始化尝试以获得稳定结果。"
+
     return CallToolResult(
         content=[TextContent(type="text", text=result_text)],
         structuredContent=result.model_dump()
@@ -539,25 +678,35 @@ async def handle_variance_decomposition(ctx, data, periods=10, max_lags=5, **kwa
 🔍 分析设置：
 - 分解期数 = {periods}
 - 最大滞后阶数 = {max_lags}
-- 变量数量 = {len(result) if isinstance(result, dict) else '未知'}
+- 变量数量 = {len(data) if data else '未知'}
 
 📈 方差分解结果："""
-    
+
     # 添加方差分解结果
-    if isinstance(result, dict):
-        for var_name, decomposition in result.items():
+    if isinstance(result, dict) and "variance_decomposition" in result:
+        variance_decomp = result["variance_decomposition"]
+        horizon = result.get("horizon", periods)
+        
+        result_text += f"\n- 分析期数: {horizon}期"
+        
+        for var_name, decomposition in variance_decomp.items():
+            result_text += f"\n\n🔬 变量 '{var_name}' 的方差来源："
             if isinstance(decomposition, dict):
-                result_text += f"\n\n🔬 变量 '{var_name}' 的方差来源："
-                for source, percentage in decomposition.items():
-                    result_text += f"\n- {source}: {percentage:.1f}%"
+                for source, percentages in decomposition.items():
+                    if isinstance(percentages, list) and len(percentages) > 0:
+                        # 显示最后一期的贡献度
+                        final_percentage = percentages[-1] * 100 if isinstance(percentages[-1], (int, float)) else 0
+                        result_text += f"\n- {source}: {final_percentage:.1f}%"
+                    else:
+                        result_text += f"\n- {source}: {percentages:.1f}%"
             else:
-                result_text += f"\n- {var_name}: {decomposition}"
+                result_text += f"\n- 总方差: {decomposition:.1f}%"
     else:
-        result_text += f"\n- 结果: {result}"
-    
-    result_text += f"\n\n💡 分析说明：方差分解用于分析多变量系统中各变量对预测误差方差的贡献程度。"
-    result_text += f"\n\n⚠️ 注意事项：方差分解结果依赖于模型的滞后阶数选择，不同期数的分解结果可能不同。"
-    
+        result_text += f"\n- 结果格式异常，无法解析方差分解结果"
+
+    result_text += f"\n\n💡 分析说明：方差分解用于分析多变量系统中各变量对预测误差方差的贡献程度，反映变量间的动态影响关系。"
+    result_text += f"\n\n⚠️ 注意事项：方差分解结果依赖于VAR模型的滞后阶数选择，不同期数的分解结果反映短期和长期影响。"
+
     return CallToolResult(
         content=[TextContent(type="text", text=result_text)],
         structuredContent=result

{aigroup_econ_mcp-0.9.0.dist-info → aigroup_econ_mcp-1.0.1.dist-info}/METADATA

@@ -1,7 +1,7 @@
 Metadata-Version: 2.4
 Name: aigroup-econ-mcp
-Version: 0.9.0
-Summary: 专业计量经济学MCP工具 - 让大模型直接进行数据分析（优化版：统一输出格式，增强模型说明）
+Version: 1.0.1
+Summary: 专业计量经济学MCP工具 - 让大模型直接进行数据分析（修复版：修复状态空间模型和方差分解分析错误，增强时间序列分析）
 Project-URL: Homepage, https://github.com/aigroup/aigroup-econ-mcp
 Project-URL: Repository, https://github.com/aigroup/aigroup-econ-mcp.git
 Project-URL: Issues, https://github.com/aigroup/aigroup-econ-mcp/issues

{aigroup_econ_mcp-0.9.0.dist-info → aigroup_econ_mcp-1.0.1.dist-info}/RECORD

@@ -1,7 +1,7 @@
-aigroup_econ_mcp/__init__.py,sha256=VY0cBsd0vAi25gJRiXcoTbr8Vu43BLQvJ_oNjPXIBeY,511
+aigroup_econ_mcp/__init__.py,sha256=Emzv62DUuziInXIYpCLshyqDmrqPZBsFBiFSrIiUiWU,511
 aigroup_econ_mcp/cli.py,sha256=7yeNXWNwMdpUswAO4LsqAvb0EmCO3S6Bs6sl483uSXI,3363
 aigroup_econ_mcp/config.py,sha256=ab5X4-H8isIe2nma0c0AOqlyYgwhf5kfe9Zx5XRrzIo,18876
-aigroup_econ_mcp/server.py,sha256=gWpjGDMsj0znYRh3Nq5mzyatPeO-iyQgxYzZXF34Fvs,30899
+aigroup_econ_mcp/server.py,sha256=jEcEMhwiWCr_9QziqW4nvV2z8Eg807iT6do2C6mjpo8,30899
 aigroup_econ_mcp/tools/__init__.py,sha256=WYrsZX3Emv09c8QikvtG2BouUFZCguYkQ7eDjuwarAg,396
 aigroup_econ_mcp/tools/base.py,sha256=Mv_mcKVTIg9A2dsqBBiU74_Ai2nb5sn2S3U4CNOxLKw,15218
 aigroup_econ_mcp/tools/cache.py,sha256=Urv2zuycp5dS7Qh-XQWEMrwszq9RZ-il8cz_-WniGgc,15311
@@ -17,14 +17,14 @@ aigroup_econ_mcp/tools/optimized_example.py,sha256=tZVQ2jTzHY_zixTynm4Sq8gj5hz6e
 aigroup_econ_mcp/tools/panel_data.py,sha256=qFZICvt9Plt2bOvCCgAveVncb_QpHvWzDssdQntKf5M,22696
 aigroup_econ_mcp/tools/regression.py,sha256=uMGRGUQo4mU1sb8fwpP2FpkCqt_e9AtqEtUpInACtJo,6443
 aigroup_econ_mcp/tools/statistics.py,sha256=2cHgNSUXwPYPLxntVOEOL8yF-x92mrgjK-R8kkxDihg,4239
-aigroup_econ_mcp/tools/time_series.py,sha256=LNCO0bYXLPilQ2kSVXA3woNp8ERVq7n3jaoQhWgTCJQ,21763
+aigroup_econ_mcp/tools/time_series.py,sha256=ZlkYn0HMm_jRn0LuxIEC4VegM3vNFPCOfdQXh5AayvA,22911
 aigroup_econ_mcp/tools/timeout.py,sha256=vNnGsR0sXW1xvIbKCF-qPUU3QNDAn_MaQgSxbGxkfW4,8404
 aigroup_econ_mcp/tools/tool_descriptions.py,sha256=Oj_14_79AB8Ku64mV0cdoV5f2-UFx-0NY3Xxjj6L-1A,32506
-aigroup_econ_mcp/tools/tool_handlers.py,sha256=RUXCB8dYkS2sbn7pKl3WPI70HQHwCDoy0hEmQMJ8rbs,34399
+aigroup_econ_mcp/tools/tool_handlers.py,sha256=lPBSOeKPJLD1-u0nBZEXMbMg9j4Yp0DG3GtFWbjhiS0,41717
 aigroup_econ_mcp/tools/tool_registry.py,sha256=4SFpMnReZyGfEHCCDnojwHIUEpuQICS9M2u_9xuoUck,4413
 aigroup_econ_mcp/tools/validation.py,sha256=F7LHwog5xtFIMjD9D48kd8jAF5MsZb7wjdrgaOg8EKo,16657
-aigroup_econ_mcp-0.9.0.dist-info/METADATA,sha256=-0IWKugPYec7nEWq85UtkCwFaYhePOwrNP3P7Wq6tIo,10857
-aigroup_econ_mcp-0.9.0.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-aigroup_econ_mcp-0.9.0.dist-info/entry_points.txt,sha256=j5ZJYOc4lAZV-X3XkAuGhzHtIRcJtZ6Gz8ZKPY_QTrM,62
-aigroup_econ_mcp-0.9.0.dist-info/licenses/LICENSE,sha256=DoyCJUWlDzKbqc5KRbFpsGYLwLh-XJRHKQDoITjb1yc,1083
-aigroup_econ_mcp-0.9.0.dist-info/RECORD,,
+aigroup_econ_mcp-1.0.1.dist-info/METADATA,sha256=kWS3Gb9uK7CJZboVmQG7acTtUoDEI3T2zBNoabqFSvk,10896
+aigroup_econ_mcp-1.0.1.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+aigroup_econ_mcp-1.0.1.dist-info/entry_points.txt,sha256=j5ZJYOc4lAZV-X3XkAuGhzHtIRcJtZ6Gz8ZKPY_QTrM,62
+aigroup_econ_mcp-1.0.1.dist-info/licenses/LICENSE,sha256=DoyCJUWlDzKbqc5KRbFpsGYLwLh-XJRHKQDoITjb1yc,1083
+aigroup_econ_mcp-1.0.1.dist-info/RECORD,,