我有一个数据帧,系数来自glm(下面是 betas
) . 数据框包含协变量标签,协变量形式和估计值 . 形式是线性(Li),平方/二次(Sq)和log(Ps) .
betas <- structure(list(CovGen = c("A", "B", "C", "D", "E", "F", "G",
"G", "H"), Form = c("Li", "Li", "Li", "Li", "Li", "Li", "Li",
"Sq", "Ps"), Estimate = c(0.0294573176934061, 0.0100315121169383,
-0.0155864186367343, -0.00871344935814372, 0.0362538988332902,
-0.0263072916746069, 0.0865742118052235, 0.0614689145750204,
0.00229745713752781)), .Names = c("CovGen", "Form", "Estimate"
), row.names = c(NA, 9L), class = "data.frame")
betas
CovGen Form Estimate
1 A Li 0.029457318
2 B Li 0.010031512
3 C Li -0.015586419
4 D Li -0.008713449
5 E Li 0.036253899
6 F Li -0.026307292
7 G Li 0.086574212
8 G Sq 0.061468915
9 H Ps 0.002297457
我尝试应用系数估计值来手动预测新数据帧的值( dat
包含在此处使用 dput
) .
dat <- structure(list(B = c(-1.47218074669544, -1.46929972689195, -1.46641870708846,
-1.46353768728497, -1.46065666748148, -1.45777564767799), C = c(-1.09847692593512,
-1.09375316152745, -1.08902939711978, -1.08430563271211, -1.07958186830444,
-1.07485810389677), D = c(-1.0109875688763, -1.00407851818141,
-0.997169467486518, -0.990260416791627, -0.983351366096736, -0.976442315401845
), E = c(-3.19632050296668, -3.19041566990116, -3.18451083683563,
-3.17860600377011, -3.17270117070458, -3.16679633763906), F = c(-2.81211918021003,
-2.80673925496675, -2.80135932972346, -2.79597940448018, -2.7905994792369,
-2.78521955399362), G = c(-2.32916817000267, -2.32368219245727,
-2.31819621491187, -2.31271023736647, -2.30722425982107, -2.30173828227567
), H = c(0.442067970883549, 0.417909464459238, 0.393750958034926,
0.369592451610615, 0.345433945186303, 0.321275438761992)), .Names = c("B",
"C", "D", "E", "F", "G", "H"), row.names = c(NA, 6L), class = "data.frame") "C", "D", "E", "F", "G", "H"), row.names = c(NA, 6L), class = "data.frame")
> dat
B C D E F G H
1 -1.472181 -1.098477 -1.0109876 -3.196321 -2.812119 -2.329168 0.4420680
2 -1.469300 -1.093753 -1.0040785 -3.190416 -2.806739 -2.323682 0.4179095
3 -1.466419 -1.089029 -0.9971695 -3.184511 -2.801359 -2.318196 0.3937510
4 -1.463538 -1.084306 -0.9902604 -3.178606 -2.795979 -2.312710 0.3695925
5 -1.460657 -1.079582 -0.9833514 -3.172701 -2.790599 -2.307224 0.3454339
6 -1.457776 -1.074858 -0.9764423 -3.166796 -2.785220 -2.301738 0.3212754
我试图将 dat
df中的新数据乘以相应的beta并考虑函数形式 . 更具体地说,在这里包含的例子中,我想将G beta的Sq形式应用于 dat$G^2
,将Ps H beta应用于 log(dat$H)
. 所有其他测试版和值可以简单地直接相乘而不考虑函数形式 . 请注意,A beta未应用于 dat
df中的新值 .
我可能需要通过奖励 ifelse
声明,但我想知道是否有其他想法和/或建议 .
我在一个更大的循环中工作,并且每个协变量都没有一致的形式 .
期望的结果是矩阵或df,其中列包含每个β-形式组合的预测值 . 例如,将存在包含除G之外的所有测试值的预测值的单个列,其将具有G和G ^ 2的预测值 .
提前致谢 .
2 回答
您可以尝试这样的解决方案
但这不适用于您当前的数据,因为没有
dat$A
而您无法记录负数 .我试着构建公式,然后使用
model.matrix
和矩阵乘法,如下所示:正如MrFlick所说,这不适用于您当前的样本数据 .