python - Pysal:如何使用 pysal 回归模型进行交叉验证？

Question

我正在使用 pysal 进行空间回归。这就是我正在做的事情。我首先划分了训练集和验证集

import pysal as ps
m_train = ps.model.spreg.GM_Lag(ytrain, xtrain, w=w_train, spat_diag=True)

如何使用 m_train 的系数在验证集上测试模型？

print(m_train.summary)

REGRESSION
----------
SUMMARY OF OUTPUT: SPATIAL TWO STAGE LEAST SQUARES
--------------------------------------------------
Data set            :     unknown
Weights matrix      :     unknown
Dependent Variable  :     dep_var                Number of Observations:         138
Mean dependent var  :      0.2200                Number of Variables   :          12
S.D. dependent var  :      2.2411                Degrees of Freedom    :         126
Pseudo R-squared    :      0.6609
Spatial Pseudo R-squared: omitted due to rho outside the boundary (-1, 1).
------------------------------------------------------------------------------------
            Variable     Coefficient       Std.Error     z-Statistic     Probability
------------------------------------------------------------------------------------
            CONSTANT      -0.1456487       0.1166442      -1.2486579       0.2117902
               var_1       0.0326638       0.1420966       0.2298707       0.8181922
               var_2      -0.0252494       0.1520906      -0.1660155       0.8681447
               var_3      -0.0899973       0.1598643      -0.5629610       0.5734615
               var_4       0.0720296       0.1980007       0.3637845       0.7160190
               var_5      -0.1850660       0.1932144      -0.9578273       0.3381499
               var_6       0.3282914       0.1482416       2.2145705       0.0267896
               var_7       0.0769414       0.1665854       0.4618738       0.6441718
               var_8       0.0285464       0.1570794       0.1817320       0.8557930
               var_9       0.0427946       0.1613301       0.2652607       0.7908086
              var_10      -0.1375104       0.1636461      -0.8402910       0.4007452
           W_dep_var       1.0880815       0.0960350      11.3300537       0.0000000
------------------------------------------------------------------------------------
Instrumented: W_dep_var
Instruments: W_var_1, W_var_10, W_var_2, W_var_3, W_var_4, W_var_5, W_var_6,
             W_var_7, W_var_8, W_var_9
Warning: *** WARNING: Estimate for spatial lag coefficient is outside the boundary (-1, 1). ***

DIAGNOSTICS FOR SPATIAL DEPENDENCE
TEST                           MI/DF       VALUE           PROB
Anselin-Kelejian Test             1           9.856          0.0017
================================ END OF REPORT =====================================

Answer 1

我也遇到了同样的问题，并尝试使用以下方法解决。我不完全确定这是否是正确/正确的方法。我也很想知道是否有更好的方法来解决这个问题。

    x = xtest.values
    betas = m_train.betas    # beta coefficients of the spatial lag model
    betasX = betas[:-1]
    betaRho = betas[-1]

    wtArr, ids = wt.full()        # convert pysal weight matrix into an array

    In = np.zeros((wtArr.shape[0], wtArr.shape[0]), float)
    np.fill_diagonal(In, 1)       

    rho = betasRho[0]
    inter = np.subtract(In, rho*wtArr)
    BRho = np.linalg.inv(inter)

    xNew = np.c_[np.ones(wtArr.shape[0]), x]
    xNew = np.asmatrix(xNew)
    predY = BRho.dot(xNew).dot(betasX)

获得预测的 y 后，您可以使用其他方法测试其性能。