F********E
发帖数: 1025 | 1 What does "bandwidth" mean in non-parameter fitting? Does a smaller "
bandwidth" imply a more "squeezed" shape as scale parameter mean in normal
distribution?
Any link to a easy-to-read tutorial link will also be appreciated! |
h***i
发帖数: 3844 | 2 翻翻fan jianqing 的lcal polynomial 就可以了
bandwidth是调节goodness of fit 和overfitting的。
【在 F********E 的大作中提到】
: What does "bandwidth" mean in non-parameter fitting? Does a smaller "
: bandwidth" imply a more "squeezed" shape as scale parameter mean in normal
: distribution?
: Any link to a easy-to-read tutorial link will also be appreciated!
|
F********E
发帖数: 1025 | 3 Didn't get the book yet. Just wondering, does the value of the automatically
optimized bandwidth (by Matlab) imply something related to the shape?
【在 h***i 的大作中提到】
: 翻翻fan jianqing 的lcal polynomial 就可以了
: bandwidth是调节goodness of fit 和overfitting的。
|
D******n
发帖数: 2836 | 4 smoothing parameter, the larger the smooother , the smalller the more rugged
【在 F********E 的大作中提到】
: What does "bandwidth" mean in non-parameter fitting? Does a smaller "
: bandwidth" imply a more "squeezed" shape as scale parameter mean in normal
: distribution?
: Any link to a easy-to-read tutorial link will also be appreciated!
|
s*****n
发帖数: 2174 | 5 bandwidth 一般是local fitting的参数, 相当于在local多大的范围内进行regression
.
从bias-variance trade off的角度, bandwidth大, 则bias大, 但是variance小.
bandwidth小, 则bias小, variance大. Optimal bandwidth一般是通过某些预先设定的
loss function来找到的, 比如cross-validation.
从smoothing的角度, bandwidth相当于smoothing的强度, bandwidth越大, 结果越
smooth. |
F********E
发帖数: 1025 | 6 Got it. Sort of disappointed since I need some numerical value to quantify
the extent of the "squeezing" of the distribution shape. And I am wondering
why nonparametric fitting doesn't give a similar value like the scale
parameter (at least in matlab).
Actually, I have a pdf very similar to an exponential shape, and apparently
the normal distribution is expected to fail. So I tried the exponential
fitting in Matlab using dfittool. Surprisingly, it ended up with an exp
shape with the peak at the
【在 D******n 的大作中提到】
: smoothing parameter, the larger the smooother , the smalller the more rugged
|
h***i
发帖数: 3844 | 7 你还是用R吧,matlab用过,但是没有R熟悉.
wondering
apparently
to
【在 F********E 的大作中提到】
: Got it. Sort of disappointed since I need some numerical value to quantify
: the extent of the "squeezing" of the distribution shape. And I am wondering
: why nonparametric fitting doesn't give a similar value like the scale
: parameter (at least in matlab).
: Actually, I have a pdf very similar to an exponential shape, and apparently
: the normal distribution is expected to fail. So I tried the exponential
: fitting in Matlab using dfittool. Surprisingly, it ended up with an exp
: shape with the peak at the
|
