当您使用 训练时method="lasso"
,来自 elasticnet 的 enet 被称为:
lasso$finalModel$call
elasticnet::enet(x = as.matrix(x), y = y, lambda = 0)
小插图写道:
LARS-EN 算法以与最小二乘拟合相同的计算成本同时为收缩参数的所有值计算完整的弹性净解
在 下lasso$finalModel$beta.pure
,您拥有所有 16 组系数的系数,对应于 下的 16 个 L1 范数值lasso$finalModel$L1norm
:
length(lasso$finalModel$L1norm)
[1] 16
dim(lasso$finalModel$beta.pure)
[1] 16 13
您也可以使用 predict 来查看它:
predict(lasso$finalModel,type="coef")
$s
[1] 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
$fraction
[1] 0.00000000 0.06666667 0.13333333 0.20000000 0.26666667 0.33333333
[7] 0.40000000 0.46666667 0.53333333 0.60000000 0.66666667 0.73333333
[13] 0.80000000 0.86666667 0.93333333 1.00000000
$mode
[1] "step"
$coefficients
crim zn indus chas nox rm age
0 0.00000000 0.0000000 0.00000000 0.0000000 0.0000000 0.000000 0.00000000
1 0.00000000 0.0000000 0.00000000 0.0000000 0.0000000 0.000000 0.00000000
2 0.00000000 0.0000000 0.00000000 0.0000000 0.0000000 1.677765 0.00000000
3 0.00000000 0.0000000 0.00000000 0.0000000 0.0000000 2.571071 0.00000000
4 0.00000000 0.0000000 0.00000000 0.0000000 0.0000000 2.716138 0.00000000
5 0.00000000 0.0000000 0.00000000 0.2586083 0.0000000 2.885615 0.00000000
6 -0.05232643 0.0000000 0.00000000 0.3543411 0.0000000 2.953605 0.00000000
7 -0.13286554 0.0000000 0.00000000 0.4095229 0.0000000 2.984026 0.00000000
8 -0.21665925 0.0000000 0.00000000 0.5196189 -0.5933941 3.003512 0.00000000
9 -0.32168140 0.3326103 0.00000000 0.6044308 -1.0246080 2.973693 0.00000000
10 -0.33568474 0.3771889 -0.02165730 0.6165190 -1.0728128 2.967696 0.00000000
11 -0.42820289 0.4522827 -0.09212253 0.6407298 -1.2474934 2.932427 0.00000000
12 -0.62605363 0.7005114 0.00000000 0.6574277 -1.5655601 2.832726 0.00000000
13 -0.88747102 1.0150162 0.00000000 0.6856705 -1.9476465 2.694820 0.00000000
14 -0.91679342 1.0613165 0.09956489 0.6837833 -2.0217269 2.684401 0.00000000
15 -0.92906457 1.0826390 0.14103943 0.6824144 -2.0587536 2.676877 0.01948534
插入符号调整的超参数是最大 L1 范数的分数,因此在您提供的结果中,它将是 1,即 max :
lasso
The lasso
506 samples
13 predictor
Pre-processing: centered (13), scaled (13)
Resampling: Cross-Validated (10 fold)
Summary of sample sizes: 51, 51, 51, 50, 51, 50, ...
Resampling results across tuning parameters:
fraction RMSE Rsquared MAE
0.001 9.182599 0.5075081 6.646013
0.010 9.022117 0.5075081 6.520153
0.100 7.597607 0.5572499 5.402851
1.000 6.158513 0.6033310 4.140362
RMSE was used to select the optimal model using the smallest value.
The final value used for the model was fraction = 1.
要获得最佳分数的系数:
predict(lasso$finalModel,type="coef",s=16)
$s
[1] 16
$fraction
[1] 1
$mode
[1] "step"
$coefficients
crim zn indus chas nox rm
-0.92906457 1.08263896 0.14103943 0.68241438 -2.05875361 2.67687661
age dis rad tax ptratio black
0.01948534 -3.10711605 2.66485220 -2.07883689 -2.06264585 0.85010886
lstat
-3.74733185