m**c
发帖数: 88 | 1 一个分布的Cumulative distribution function已知，但是比较复杂，直接微分求其
pdf行不
通， 有什么好的方法来逼近这个pdf啊？
谢谢 | l*********s
发帖数: 5409 | 2 numeric solution not feasible? | t******g
发帖数: 2253 | | g******n
发帖数: 339 | 4 I assume you meant to say the differential of CDF, or the CDF itself can not
be written in a closed form. In this case, as long as you have a numerical
CDF function F(x), you can easily define a numerical PDF function f(x)~[F(x+
delta/2)-F(x-delta/2)]/delta, where delta is small.
The choice of delta is a more complicated issue, if the CDF is not a smooth
function. However, I remember that R has a package called numDeriv, which
may say you some time...
【在 m**c 的大作中提到】
: 一个分布的Cumulative distribution function已知，但是比较复杂，直接微分求其
: pdf行不
: 通， 有什么好的方法来逼近这个pdf啊？
: 谢谢
| m**c
发帖数: 88 | |
|
