具有lm的线性模型:如何获得预测值总和的预测方差 [英] linear model with `lm`: how to get prediction variance of sum of predicted values
问题描述
我正在对具有多个预测变量的线性模型的预测值求和,如下面的示例所示,并希望计算该总和的组合方差,标准误差和可能的置信区间.
I'm summing the predicted values from a linear model with multiple predictors, as in the example below, and want to calculate the combined variance, standard error and possibly confidence intervals for this sum.
lm.tree <- lm(Volume ~ poly(Girth,2), data = trees)
假设我有一组Girths
:
newdat <- list(Girth = c(10,12,14,16)
我要为其预测总数Volume
的
:
for which I want to predict the total Volume
:
pr <- predict(lm.tree, newdat, se.fit = TRUE)
total <- sum(pr$fit)
# [1] 111.512
如何获取total
的方差?
Similar questions are here (for GAMs), but I'm not sure how to proceed with the vcov(lm.trees)
. I'd be grateful for a reference for the method.
推荐答案
您需要获取完整的方差-协方差矩阵,然后将其所有元素求和.这是一个小证明:
此处的证明使用的是另一个定理,您可以从 Covariance-wikipedia :
The proof here is using another theorem, which you can find from Covariance-wikipedia:
具体来说,我们采用的线性变换是全1的列矩阵.生成的二次形式如下,其中所有x_i
和x_j
是1.
Specifically, the linear transform we take is a column matrix of all 1's. The resulting quadratic form is computed as following, with all x_i
and x_j
being 1.
## your model
lm.tree <- lm(Volume ~ poly(Girth, 2), data = trees)
## newdata (a data frame)
newdat <- data.frame(Girth = c(10, 12, 14, 16))
重新实现predict.lm
以计算方差-协方差矩阵
有关predict.lm
的工作原理,请参见 predict.lm()如何计算置信区间和预测区间?.以下小型函数lm_predict
模仿了它的功能,除了
Re-implement predict.lm
to compute variance-covariance matrix
See How does predict.lm() compute confidence interval and prediction interval? for how predict.lm
works. The following small function lm_predict
mimics what it does, except that
- 它不会构造置信度或预测间隔(但构造非常简单,正如该问题解答A所述);
- 如果
diag = FALSE
,它可以计算完整的预测值方差-协方差矩阵; - 它返回方差(对于预测值和残差),而不是标准误差;
- 它不能做
type = "terms"
;它只能预测响应变量.
- it does not construct confidence or prediction interval (but construction is very straightforward as explained in that Q & A);
- it can compute complete variance-covariance matrix of predicted values if
diag = FALSE
; - it returns variance (for both predicted values and residuals), not standard error;
- it can not do
type = "terms"
; it only predict response variable.
lm_predict <- function (lmObject, newdata, diag = TRUE) {
## input checking
if (!inherits(lmObject, "lm")) stop("'lmObject' is not a valid 'lm' object!")
## extract "terms" object from the fitted model, but delete response variable
tm <- delete.response(terms(lmObject))
## linear predictor matrix
Xp <- model.matrix(tm, newdata)
## predicted values by direct matrix-vector multiplication
pred <- c(Xp %*% coef(lmObject))
## efficiently form the complete variance-covariance matrix
QR <- lmObject$qr ## qr object of fitted model
piv <- QR$pivot ## pivoting index
r <- QR$rank ## model rank / numeric rank
if (is.unsorted(piv)) {
## pivoting has been done
B <- forwardsolve(t(QR$qr), t(Xp[, piv]), r)
} else {
## no pivoting is done
B <- forwardsolve(t(QR$qr), t(Xp), r)
}
## residual variance
sig2 <- c(crossprod(residuals(lmObject))) / df.residual(lmObject)
if (diag) {
## return point-wise prediction variance
VCOV <- colSums(B ^ 2) * sig2
} else {
## return full variance-covariance matrix of predicted values
VCOV <- crossprod(B) * sig2
}
list(fit = pred, var.fit = VCOV, df = lmObject$df.residual, residual.var = sig2)
}
我们可以将其输出与predict.lm
的输出进行比较:
We can compare its output with that of predict.lm
:
predict.lm(lm.tree, newdat, se.fit = TRUE)
#$fit
# 1 2 3 4
#15.31863 22.33400 31.38568 42.47365
#
#$se.fit
# 1 2 3 4
#0.9435197 0.7327569 0.8550646 0.8852284
#
#$df
#[1] 28
#
#$residual.scale
#[1] 3.334785
lm_predict(lm.tree, newdat)
#$fit
#[1] 15.31863 22.33400 31.38568 42.47365
#
#$var.fit ## the square of `se.fit`
#[1] 0.8902294 0.5369327 0.7311355 0.7836294
#
#$df
#[1] 28
#
#$residual.var ## the square of `residual.scale`
#[1] 11.12079
尤其是:
oo <- lm_predict(lm.tree, newdat, FALSE)
oo
#$fit
#[1] 15.31863 22.33400 31.38568 42.47365
#
#$var.fit
# [,1] [,2] [,3] [,4]
#[1,] 0.89022938 0.3846809 0.04967582 -0.1147858
#[2,] 0.38468089 0.5369327 0.52828797 0.3587467
#[3,] 0.04967582 0.5282880 0.73113553 0.6582185
#[4,] -0.11478583 0.3587467 0.65821848 0.7836294
#
#$df
#[1] 28
#
#$residual.var
#[1] 11.12079
请注意,方差-协方差矩阵不是天真的方法:Xp %*% vcov(lmObject) % t(Xp)
,这很慢.
Note that the variance-covariance matrix is not computed in a naive way: Xp %*% vcov(lmObject) % t(Xp)
, which is slow.
在您的情况下,聚合操作是oo$fit
中所有值的总和.此聚合的均值和方差为
In your case, the aggregation operation is the sum of all values in oo$fit
. The mean and variance of this aggregation are
sum_mean <- sum(oo$fit) ## mean of the sum
# 111.512
sum_variance <- sum(oo$var.fit) ## variance of the sum
# 6.671575
通过使用t分布和模型中的剩余自由度,您可以进一步为该聚合值构造置信区间(CI).
You can further construct confidence interval (CI) for this aggregated value, by using t-distribution and the residual degree of freedom in the model.
alpha <- 0.95
Qt <- c(-1, 1) * qt((1 - alpha) / 2, lm.tree$df.residual, lower.tail = FALSE)
#[1] -2.048407 2.048407
## %95 CI
sum_mean + Qt * sqrt(sum_variance)
#[1] 106.2210 116.8029
构建预测间隔(PI)需要进一步考虑残差.
Constructing prediction interval (PI) needs further account for residual variance.
## adjusted variance-covariance matrix
VCOV_adj <- with(oo, var.fit + diag(residual.var, nrow(var.fit)))
## adjusted variance for the aggregation
sum_variance_adj <- sum(VCOV_adj) ## adjusted variance of the sum
## 95% PI
sum_mean + Qt * sqrt(sum_variance_adj)
#[1] 96.86122 126.16268
汇总(通常)
一般的聚合操作可以是oo$fit
的线性组合:
w[1] * fit[1] + w[2] * fit[2] + w[3] * fit[3] + ...
例如,求和运算的所有权重均为1;平均操作的所有权重为0.25(如果有4个数据).这是一个函数,它具有一个权重向量,一个显着性水平以及lm_predict
返回的值,以产生聚合统计信息.
For example, the sum operation has all weights being 1; the mean operation has all weights being 0.25 (in case of 4 data). Here is function that takes a weight vector, a significance level and what is returned by lm_predict
to produce statistics of an aggregation.
agg_pred <- function (w, predObject, alpha = 0.95) {
## input checing
if (length(w) != length(predObject$fit)) stop("'w' has wrong length!")
if (!is.matrix(predObject$var.fit)) stop("'predObject' has no variance-covariance matrix!")
## mean of the aggregation
agg_mean <- c(crossprod(predObject$fit, w))
## variance of the aggregation
agg_variance <- c(crossprod(w, predObject$var.fit %*% w))
## adjusted variance-covariance matrix
VCOV_adj <- with(predObject, var.fit + diag(residual.var, nrow(var.fit)))
## adjusted variance of the aggregation
agg_variance_adj <- c(crossprod(w, VCOV_adj %*% w))
## t-distribution quantiles
Qt <- c(-1, 1) * qt((1 - alpha) / 2, predObject$df, lower.tail = FALSE)
## names of CI and PI
NAME <- c("lower", "upper")
## CI
CI <- setNames(agg_mean + Qt * sqrt(agg_variance), NAME)
## PI
PI <- setNames(agg_mean + Qt * sqrt(agg_variance_adj), NAME)
## return
list(mean = agg_mean, var = agg_variance, CI = CI, PI = PI)
}
对先前的求和运算进行快速测试:
A quick test on the previous sum operation:
agg_pred(rep(1, length(oo$fit)), oo)
#$mean
#[1] 111.512
#
#$var
#[1] 6.671575
#
#$CI
# lower upper
#106.2210 116.8029
#
#$PI
# lower upper
# 96.86122 126.16268
对平均操作进行快速测试:
And a quick test for average operation:
agg_pred(rep(1, length(oo$fit)) / length(oo$fit), oo)
#$mean
#[1] 27.87799
#
#$var
#[1] 0.4169734
#
#$CI
# lower upper
#26.55526 29.20072
#
#$PI
# lower upper
#24.21531 31.54067
备注
此答案已得到改进,可为使用lm()进行线性回归:提供汇总预测的预测间隔值.
Remark
This answer is improved to provide easy-to-use functions for Linear regression with `lm()`: prediction interval for aggregated predicted values.
太好了!太感谢了!我忘了提到一件事:在我的实际应用中,我需要汇总约300,000个预测,这将创建一个完整的方差-协方差矩阵,其大小约为700GB.您是否知道是否有一种计算上更有效的方法可以直接求出方差-协方差矩阵的和?
This is great! Thank you so much! There is one thing I forgot to mention: in my actual application I need to sum ~300,000 predictions which would create a full variance-covariance matrix which is about ~700GB in size. Do you have any idea if there is a computationally more efficient way to directly get to the sum of the variance-covariance matrix?
感谢使用`lm()`进行线性回归:汇总预测值的预测间隔有用的评论.是的,这是可能的,并且(显着)在计算上也更便宜.此刻,lm_predict
这样形成方差-协方差:
Thanks to the OP of Linear regression with `lm()`: prediction interval for aggregated predicted values for this very helpful comment. Yes, it is possible and it is also (significantly) computationally cheaper. At the moment, lm_predict
form the variance-covariance as such:
agg_pred
将预测方差(用于构造CI)计算为二次形式:w'(B'B)w
,将预测方差(用于构造PI)计算为另一个二次形式w'(B'B + D)w
,其中D
是对角矩阵剩余方差.显然,如果我们将这两个功能融合在一起,就会有更好的计算策略:
agg_pred
computes the prediction variance (for constructing CI) as a quadratic form: w'(B'B)w
, and the prediction variance (for construction PI) as another quadratic form w'(B'B + D)w
, where D
is a diagonal matrix of residual variance. Obviously if we fuse those two functions, we have a better computational strategy:
避免计算B
和B'B
;我们已将所有矩阵-矩阵乘法替换为矩阵-向量乘法. B
和B'B
没有存储空间;仅用于u
,它只是一个向量.这是融合的实现.
Computation of B
and B'B
is avoided; we have replaced all matrix-matrix multiplication to matrix-vector multiplication. There is no memory storage for B
and B'B
; only for u
which is just a vector. Here is the fused implementation.
## this function requires neither `lm_predict` nor `agg_pred`
fast_agg_pred <- function (w, lmObject, newdata, alpha = 0.95) {
## input checking
if (!inherits(lmObject, "lm")) stop("'lmObject' is not a valid 'lm' object!")
if (!is.data.frame(newdata)) newdata <- as.data.frame(newdata)
if (length(w) != nrow(newdata)) stop("length(w) does not match nrow(newdata)")
## extract "terms" object from the fitted model, but delete response variable
tm <- delete.response(terms(lmObject))
## linear predictor matrix
Xp <- model.matrix(tm, newdata)
## predicted values by direct matrix-vector multiplication
pred <- c(Xp %*% coef(lmObject))
## mean of the aggregation
agg_mean <- c(crossprod(pred, w))
## residual variance
sig2 <- c(crossprod(residuals(lmObject))) / df.residual(lmObject)
## efficiently compute variance of the aggregation without matrix-matrix computations
QR <- lmObject$qr ## qr object of fitted model
piv <- QR$pivot ## pivoting index
r <- QR$rank ## model rank / numeric rank
u <- forwardsolve(t(QR$qr), c(crossprod(Xp, w))[piv], r)
agg_variance <- c(crossprod(u)) * sig2
## adjusted variance of the aggregation
agg_variance_adj <- agg_variance + c(crossprod(w)) * sig2
## t-distribution quantiles
Qt <- c(-1, 1) * qt((1 - alpha) / 2, lmObject$df.residual, lower.tail = FALSE)
## names of CI and PI
NAME <- c("lower", "upper")
## CI
CI <- setNames(agg_mean + Qt * sqrt(agg_variance), NAME)
## PI
PI <- setNames(agg_mean + Qt * sqrt(agg_variance_adj), NAME)
## return
list(mean = agg_mean, var = agg_variance, CI = CI, PI = PI)
}
让我们快速测试一下.
## sum opeartion
fast_agg_pred(rep(1, nrow(newdat)), lm.tree, newdat)
#$mean
#[1] 111.512
#
#$var
#[1] 6.671575
#
#$CI
# lower upper
#106.2210 116.8029
#
#$PI
# lower upper
# 96.86122 126.16268
## average operation
fast_agg_pred(rep(1, nrow(newdat)) / nrow(newdat), lm.tree, newdat)
#$mean
#[1] 27.87799
#
#$var
#[1] 0.4169734
#
#$CI
# lower upper
#26.55526 29.20072
#
#$PI
# lower upper
#24.21531 31.54067
是的,答案是正确的!
这篇关于具有lm的线性模型:如何获得预测值总和的预测方差的文章就介绍到这了,希望我们推荐的答案对大家有所帮助,也希望大家多多支持IT屋!