m****s 发帖数: 402 | 1 Kalman Filter,
Neural Network,
Bayessian Network,
Hidden Markov Model,
感觉都很相似, 比较容易混,
牛人来说说它们区别在哪呢? | f**********r 发帖数: 2137 | | w********d 发帖数: 275 | 3
these are totally different concept
each model can have supervised and unsupervised case,
depends on the training data and differ is parameter estimation.
kalman filter and hmm are two kind of bayesian network,
which is called direct graphical model.
(that is ,bayesian network is direted graphical model)
kalman filter and hmm are two kind of
direct graphical model.
Also there is another kind of undirect graphical model, such as
markov random fields. | D*******a 发帖数: 3688 | 4 估计就是个general interview question | f*********y 发帖数: 43 | 5 诚心问两个问题:
(1)能否给个intro,各个approach在哪种场合有优势,或者明显优势?
in particular, 预测股票市场(time series)那个适合?
(2)另外,Kalman filter 里面,基本上是等时间差data预测,i.e.,
x_hat~(n+1) = A x_hat(n) + B u(n+1)
//很多场合,u=0,therefore, x_hat~(n+1) = A x_hat(n);
P = ...
K_gain = ...
x_hat(n+1) = ...
根据上述公式,已知了1,2,... n,可以预测n+1
请问,我可否在前n-1 measurement point时候只预测下一步,
在第n个measurement point的时候突然直接预测第n+k步?
附:http://en.wikipedia.org/wiki/Kalman_filter
【在 w********d 的大作中提到】 : : these are totally different concept : each model can have supervised and unsupervised case, : depends on the training data and differ is parameter estimation. : kalman filter and hmm are two kind of bayesian network, : which is called direct graphical model. : (that is ,bayesian network is direted graphical model) : kalman filter and hmm are two kind of : direct graphical model. : Also there is another kind of undirect graphical model, such as
| f******k 发帖数: 297 | 6 what does kalman filter have anything to do with bayesian framework? it
neither has any prior settings nor calculate any posteriors...
【在 w********d 的大作中提到】 : : these are totally different concept : each model can have supervised and unsupervised case, : depends on the training data and differ is parameter estimation. : kalman filter and hmm are two kind of bayesian network, : which is called direct graphical model. : (that is ,bayesian network is direted graphical model) : kalman filter and hmm are two kind of : direct graphical model. : Also there is another kind of undirect graphical model, such as
| w********d 发帖数: 275 | 7 hmm and kalman filter can be thought as the simplest
dynamic bayesian network
【在 f******k 的大作中提到】 : what does kalman filter have anything to do with bayesian framework? it : neither has any prior settings nor calculate any posteriors...
|
|