Would you kindly do another vedio series on the Hierarchical version of HMM ? And when shall we prefer to use the Hierarchical version ? It would be great if you provide an implementation as well in Python , R , Mathlab
Thank you for the great video! I would like to point out that it is not obvious in 9:30 to get from \alpha(x_t) * \beta(x_t) = p(x_t, Y). My thought is that \alpha(x_t) * \beta(x_t) = p(x_t, y_1~y_T) * p(y_{t+1} ~ y_T | x_t) = p(y_1~y_t | x_t) * p(x_t) * p(y_{t+1} ~ y_T | x_t) *=* p(y_1 ~ y_T | x_t) * p(x_t) = p(x_t, y_1 ~ y_T). The '*=*' place is derived from the Markov assumption, which can be explained as "given the current state x_t, the furture state x_{t+1} so as the outcome y_{t+1} is independent of previous states {x_1 ~ x_{t-1}}, so as the previous outcomes {y_1 ~ y_{t-1}}", therefore we can merge the probability as shown. (wondering if my thought is correct...)
Completed a project thanks to this video. You're the best man!!!
Such an amazing video. Very clear to understand! Thanks much for the effort.
Thanks for the great explanation! Finally understood the implementation of HMM`s
Would you kindly do another vedio series on the Hierarchical version of HMM ? And when shall we prefer to use the Hierarchical version ? It would be great if you provide an implementation as well in Python , R , Mathlab
thanks
At 9:48 he says p(y1,y2,y3,x3) x p(y4,y5,y6|x3) = p(x3, Y) where Y = {y1,y2,...,y6} anyone figured out how?
figured out ... y1,y2,y3 independent of y4,y5,y6 given x3; that is : p(a,b,c) = p(b,c| a) x p(a) = p(b|a) p(c|a) p(a) = p(a,b) p(c|a)
How was the expression for p(x2,y1,y2) derived at 11:48? Shouldn't p(x2,y1,y2) = p(x2|y2,y1)p(y2|y1)p(y1)?
p(x2,y1,y2)=sigma_x1(p(x1,x2,y1,y2))=sigma_x1(p(y2|x2,x1,y1)p(x2|x1,y1)p(x1,y1)=sigma_x1(p(y2|x2)p(x2|x1)p(x1,y1))
@@eliesfeir4511 Thank you! This is much clearer.
Thank you for the great video! I would like to point out that it is not obvious in 9:30 to get from \alpha(x_t) * \beta(x_t) = p(x_t, Y). My thought is that \alpha(x_t) * \beta(x_t) = p(x_t, y_1~y_T) * p(y_{t+1} ~ y_T | x_t) = p(y_1~y_t | x_t) * p(x_t) * p(y_{t+1} ~ y_T | x_t) *=* p(y_1 ~ y_T | x_t) * p(x_t) = p(x_t, y_1 ~ y_T). The '*=*' place is derived from the Markov assumption, which can be explained as "given the current state x_t, the furture state x_{t+1} so as the outcome y_{t+1} is independent of previous states {x_1 ~ x_{t-1}}, so as the previous outcomes {y_1 ~ y_{t-1}}", therefore we can merge the probability as shown. (wondering if my thought is correct...)
Your math checks out to me, but I am new to this as well.
6:52
thanks
thanks
thanks
thanks