Gaussian mixture model - does an improper uniform prior give a proper posterior?
Clash Royale CLAN TAG#URR8PPP
.everyoneloves__top-leaderboard:empty,.everyoneloves__mid-leaderboard:empty margin-bottom:0;
up vote
2
down vote
favorite
We draw $n$ i.i.d. points $x_1 , x_2 , ..., x_n$ from the following Gaussian mixture:
$$p(x|mu_1,mu_2) = frac12 textN (x|mu_1,1) + frac12 textN (x|mu_2,1).$$
The prior is the improper prior $p(mu_1, mu_2) propto 1$. Does the posterior have a finite integral (i.e., is it a proper distribution)?
prior posterior gaussian-mixture conjugate-prior
New contributor
add a comment |Â
up vote
2
down vote
favorite
We draw $n$ i.i.d. points $x_1 , x_2 , ..., x_n$ from the following Gaussian mixture:
$$p(x|mu_1,mu_2) = frac12 textN (x|mu_1,1) + frac12 textN (x|mu_2,1).$$
The prior is the improper prior $p(mu_1, mu_2) propto 1$. Does the posterior have a finite integral (i.e., is it a proper distribution)?
prior posterior gaussian-mixture conjugate-prior
New contributor
add a comment |Â
up vote
2
down vote
favorite
up vote
2
down vote
favorite
We draw $n$ i.i.d. points $x_1 , x_2 , ..., x_n$ from the following Gaussian mixture:
$$p(x|mu_1,mu_2) = frac12 textN (x|mu_1,1) + frac12 textN (x|mu_2,1).$$
The prior is the improper prior $p(mu_1, mu_2) propto 1$. Does the posterior have a finite integral (i.e., is it a proper distribution)?
prior posterior gaussian-mixture conjugate-prior
New contributor
We draw $n$ i.i.d. points $x_1 , x_2 , ..., x_n$ from the following Gaussian mixture:
$$p(x|mu_1,mu_2) = frac12 textN (x|mu_1,1) + frac12 textN (x|mu_2,1).$$
The prior is the improper prior $p(mu_1, mu_2) propto 1$. Does the posterior have a finite integral (i.e., is it a proper distribution)?
prior posterior gaussian-mixture conjugate-prior
prior posterior gaussian-mixture conjugate-prior
New contributor
New contributor
edited 2 hours ago
Ben
17.5k22286
17.5k22286
New contributor
asked 3 hours ago
aastha agrrawal
111
111
New contributor
New contributor
add a comment |Â
add a comment |Â
1 Answer
1
active
oldest
votes
up vote
2
down vote
From your specified model you have likelihood function:
$$L_mathbfx(mu_1,mu_2) = (8 pi)^-n/2 prod_i=1^n Bigg[ exp Big( -frac12 (x_i-mu_1)^2 Big) + exp Big( -frac12 (x_i-mu_2)^2 Big) Bigg].$$
To facilitate the analysis, let $mathscrB equiv 1,2 ^n$ be the set of all $n$-length binary vectors of values of the two means. For any vector $mathbfb in mathscrB$ let $mathcalS(mathbfb) equiv i=1,...,n $ be the set of index values that use the first mean. With the improper uniform prior $pi (mu_1, mu_2) propto 1$ you get the posterior:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&propto L_mathbfx(mu_1,mu_2) \[6pt]
&propto prod_i=1^n Bigg[ exp Big( -frac12 (x_i-mu_1)^2 Big) + exp Big( -frac12 (x_i-mu_2)^2 Big) Bigg] \[6pt]
&= sum_mathbfb in mathscrB exp Big( -frac12 sum_i=1^n (x_i-mu_b_i)^2 Big) \[6pt]
&= sum_mathbfb in mathscrB exp Big( -frac12 sum_i in mathcalS(mathbfb) (x_i-mu_1)^2 Big) exp Big( -frac12 sum_i notin mathcalS(mathbfb) (x_i-mu_2)^2 Big). \[6pt]
endaligned endequation$$
Now, letting $barx_1^(k) (mathbfb) equiv sum_i in mathcalS(mathbfb) x_i^k$ and $barx_2^(k) (mathbfb) equiv sum_i notin mathcalS(mathbfb) x_i^k$ we have:
$$beginequation beginaligned
exp Big( -frac12 sum_i in mathcalS(mathbfb) (x_i-mu_1)^2 Big)
&= exp Big( -frac12 Big( mu_1^2 + 2 mu_1 sum_i in mathcalS(mathbfb) x_i + sum_i in mathcalS(mathbfb) x_i^2 Big) Big) \[6pt]
&= exp Big( -frac12 ( mu_1^2 + 2 mu_1 barx_1^(1) (mathbfb) + barx_1^(2) (mathbfb) ) Big) \[10pt]
&= exp Big( -frac12 ( barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 ) Big)
exp Big( -frac12 (mu_1 - barx_1^(1) (mathbfb))^2 Big) \[10pt]
&propto exp Big( -frac12 ( barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 ) Big) cdot textN(mu_1 | barx_1^(1) (mathbfb), 1), \[10pt]
exp Big( -frac12 sum_i notin mathcalS(mathbfb) (x_i-mu_2)^2 Big) &propto exp Big( -frac12 ( barx_2^(2) (mathbfb) - barx_2^(1) (mathbfb)^2 ) Big) cdot textN(mu_2 | barx_2^(1) (mathbfb), 1). \[10pt]
endaligned endequation$$
So, letting $H(mathbfb) equiv barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 + barx_2^(2) (mathbfb) - barx_2^(1) (mathbfb)^2$ we have:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&propto sum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big) cdot textN(mu_1 | barx_1^(1) (mathbfb), 1) cdot textN(mu_2 | barx_2^(1) (mathbfb), 1). \[6pt]
endaligned endequation$$
This kernel has a finite integral, and yields the proper posterior density:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&= fracsum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big) cdot textN(mu_1 sum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big). \[6pt]
endaligned endequation$$
add a comment |Â
1 Answer
1
active
oldest
votes
1 Answer
1
active
oldest
votes
active
oldest
votes
active
oldest
votes
up vote
2
down vote
From your specified model you have likelihood function:
$$L_mathbfx(mu_1,mu_2) = (8 pi)^-n/2 prod_i=1^n Bigg[ exp Big( -frac12 (x_i-mu_1)^2 Big) + exp Big( -frac12 (x_i-mu_2)^2 Big) Bigg].$$
To facilitate the analysis, let $mathscrB equiv 1,2 ^n$ be the set of all $n$-length binary vectors of values of the two means. For any vector $mathbfb in mathscrB$ let $mathcalS(mathbfb) equiv i=1,...,n $ be the set of index values that use the first mean. With the improper uniform prior $pi (mu_1, mu_2) propto 1$ you get the posterior:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&propto L_mathbfx(mu_1,mu_2) \[6pt]
&propto prod_i=1^n Bigg[ exp Big( -frac12 (x_i-mu_1)^2 Big) + exp Big( -frac12 (x_i-mu_2)^2 Big) Bigg] \[6pt]
&= sum_mathbfb in mathscrB exp Big( -frac12 sum_i=1^n (x_i-mu_b_i)^2 Big) \[6pt]
&= sum_mathbfb in mathscrB exp Big( -frac12 sum_i in mathcalS(mathbfb) (x_i-mu_1)^2 Big) exp Big( -frac12 sum_i notin mathcalS(mathbfb) (x_i-mu_2)^2 Big). \[6pt]
endaligned endequation$$
Now, letting $barx_1^(k) (mathbfb) equiv sum_i in mathcalS(mathbfb) x_i^k$ and $barx_2^(k) (mathbfb) equiv sum_i notin mathcalS(mathbfb) x_i^k$ we have:
$$beginequation beginaligned
exp Big( -frac12 sum_i in mathcalS(mathbfb) (x_i-mu_1)^2 Big)
&= exp Big( -frac12 Big( mu_1^2 + 2 mu_1 sum_i in mathcalS(mathbfb) x_i + sum_i in mathcalS(mathbfb) x_i^2 Big) Big) \[6pt]
&= exp Big( -frac12 ( mu_1^2 + 2 mu_1 barx_1^(1) (mathbfb) + barx_1^(2) (mathbfb) ) Big) \[10pt]
&= exp Big( -frac12 ( barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 ) Big)
exp Big( -frac12 (mu_1 - barx_1^(1) (mathbfb))^2 Big) \[10pt]
&propto exp Big( -frac12 ( barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 ) Big) cdot textN(mu_1 | barx_1^(1) (mathbfb), 1), \[10pt]
exp Big( -frac12 sum_i notin mathcalS(mathbfb) (x_i-mu_2)^2 Big) &propto exp Big( -frac12 ( barx_2^(2) (mathbfb) - barx_2^(1) (mathbfb)^2 ) Big) cdot textN(mu_2 | barx_2^(1) (mathbfb), 1). \[10pt]
endaligned endequation$$
So, letting $H(mathbfb) equiv barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 + barx_2^(2) (mathbfb) - barx_2^(1) (mathbfb)^2$ we have:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&propto sum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big) cdot textN(mu_1 | barx_1^(1) (mathbfb), 1) cdot textN(mu_2 | barx_2^(1) (mathbfb), 1). \[6pt]
endaligned endequation$$
This kernel has a finite integral, and yields the proper posterior density:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&= fracsum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big) cdot textN(mu_1 sum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big). \[6pt]
endaligned endequation$$
add a comment |Â
up vote
2
down vote
From your specified model you have likelihood function:
$$L_mathbfx(mu_1,mu_2) = (8 pi)^-n/2 prod_i=1^n Bigg[ exp Big( -frac12 (x_i-mu_1)^2 Big) + exp Big( -frac12 (x_i-mu_2)^2 Big) Bigg].$$
To facilitate the analysis, let $mathscrB equiv 1,2 ^n$ be the set of all $n$-length binary vectors of values of the two means. For any vector $mathbfb in mathscrB$ let $mathcalS(mathbfb) equiv i=1,...,n $ be the set of index values that use the first mean. With the improper uniform prior $pi (mu_1, mu_2) propto 1$ you get the posterior:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&propto L_mathbfx(mu_1,mu_2) \[6pt]
&propto prod_i=1^n Bigg[ exp Big( -frac12 (x_i-mu_1)^2 Big) + exp Big( -frac12 (x_i-mu_2)^2 Big) Bigg] \[6pt]
&= sum_mathbfb in mathscrB exp Big( -frac12 sum_i=1^n (x_i-mu_b_i)^2 Big) \[6pt]
&= sum_mathbfb in mathscrB exp Big( -frac12 sum_i in mathcalS(mathbfb) (x_i-mu_1)^2 Big) exp Big( -frac12 sum_i notin mathcalS(mathbfb) (x_i-mu_2)^2 Big). \[6pt]
endaligned endequation$$
Now, letting $barx_1^(k) (mathbfb) equiv sum_i in mathcalS(mathbfb) x_i^k$ and $barx_2^(k) (mathbfb) equiv sum_i notin mathcalS(mathbfb) x_i^k$ we have:
$$beginequation beginaligned
exp Big( -frac12 sum_i in mathcalS(mathbfb) (x_i-mu_1)^2 Big)
&= exp Big( -frac12 Big( mu_1^2 + 2 mu_1 sum_i in mathcalS(mathbfb) x_i + sum_i in mathcalS(mathbfb) x_i^2 Big) Big) \[6pt]
&= exp Big( -frac12 ( mu_1^2 + 2 mu_1 barx_1^(1) (mathbfb) + barx_1^(2) (mathbfb) ) Big) \[10pt]
&= exp Big( -frac12 ( barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 ) Big)
exp Big( -frac12 (mu_1 - barx_1^(1) (mathbfb))^2 Big) \[10pt]
&propto exp Big( -frac12 ( barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 ) Big) cdot textN(mu_1 | barx_1^(1) (mathbfb), 1), \[10pt]
exp Big( -frac12 sum_i notin mathcalS(mathbfb) (x_i-mu_2)^2 Big) &propto exp Big( -frac12 ( barx_2^(2) (mathbfb) - barx_2^(1) (mathbfb)^2 ) Big) cdot textN(mu_2 | barx_2^(1) (mathbfb), 1). \[10pt]
endaligned endequation$$
So, letting $H(mathbfb) equiv barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 + barx_2^(2) (mathbfb) - barx_2^(1) (mathbfb)^2$ we have:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&propto sum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big) cdot textN(mu_1 | barx_1^(1) (mathbfb), 1) cdot textN(mu_2 | barx_2^(1) (mathbfb), 1). \[6pt]
endaligned endequation$$
This kernel has a finite integral, and yields the proper posterior density:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&= fracsum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big) cdot textN(mu_1 sum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big). \[6pt]
endaligned endequation$$
add a comment |Â
up vote
2
down vote
up vote
2
down vote
From your specified model you have likelihood function:
$$L_mathbfx(mu_1,mu_2) = (8 pi)^-n/2 prod_i=1^n Bigg[ exp Big( -frac12 (x_i-mu_1)^2 Big) + exp Big( -frac12 (x_i-mu_2)^2 Big) Bigg].$$
To facilitate the analysis, let $mathscrB equiv 1,2 ^n$ be the set of all $n$-length binary vectors of values of the two means. For any vector $mathbfb in mathscrB$ let $mathcalS(mathbfb) equiv i=1,...,n $ be the set of index values that use the first mean. With the improper uniform prior $pi (mu_1, mu_2) propto 1$ you get the posterior:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&propto L_mathbfx(mu_1,mu_2) \[6pt]
&propto prod_i=1^n Bigg[ exp Big( -frac12 (x_i-mu_1)^2 Big) + exp Big( -frac12 (x_i-mu_2)^2 Big) Bigg] \[6pt]
&= sum_mathbfb in mathscrB exp Big( -frac12 sum_i=1^n (x_i-mu_b_i)^2 Big) \[6pt]
&= sum_mathbfb in mathscrB exp Big( -frac12 sum_i in mathcalS(mathbfb) (x_i-mu_1)^2 Big) exp Big( -frac12 sum_i notin mathcalS(mathbfb) (x_i-mu_2)^2 Big). \[6pt]
endaligned endequation$$
Now, letting $barx_1^(k) (mathbfb) equiv sum_i in mathcalS(mathbfb) x_i^k$ and $barx_2^(k) (mathbfb) equiv sum_i notin mathcalS(mathbfb) x_i^k$ we have:
$$beginequation beginaligned
exp Big( -frac12 sum_i in mathcalS(mathbfb) (x_i-mu_1)^2 Big)
&= exp Big( -frac12 Big( mu_1^2 + 2 mu_1 sum_i in mathcalS(mathbfb) x_i + sum_i in mathcalS(mathbfb) x_i^2 Big) Big) \[6pt]
&= exp Big( -frac12 ( mu_1^2 + 2 mu_1 barx_1^(1) (mathbfb) + barx_1^(2) (mathbfb) ) Big) \[10pt]
&= exp Big( -frac12 ( barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 ) Big)
exp Big( -frac12 (mu_1 - barx_1^(1) (mathbfb))^2 Big) \[10pt]
&propto exp Big( -frac12 ( barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 ) Big) cdot textN(mu_1 | barx_1^(1) (mathbfb), 1), \[10pt]
exp Big( -frac12 sum_i notin mathcalS(mathbfb) (x_i-mu_2)^2 Big) &propto exp Big( -frac12 ( barx_2^(2) (mathbfb) - barx_2^(1) (mathbfb)^2 ) Big) cdot textN(mu_2 | barx_2^(1) (mathbfb), 1). \[10pt]
endaligned endequation$$
So, letting $H(mathbfb) equiv barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 + barx_2^(2) (mathbfb) - barx_2^(1) (mathbfb)^2$ we have:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&propto sum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big) cdot textN(mu_1 | barx_1^(1) (mathbfb), 1) cdot textN(mu_2 | barx_2^(1) (mathbfb), 1). \[6pt]
endaligned endequation$$
This kernel has a finite integral, and yields the proper posterior density:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&= fracsum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big) cdot textN(mu_1 sum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big). \[6pt]
endaligned endequation$$
From your specified model you have likelihood function:
$$L_mathbfx(mu_1,mu_2) = (8 pi)^-n/2 prod_i=1^n Bigg[ exp Big( -frac12 (x_i-mu_1)^2 Big) + exp Big( -frac12 (x_i-mu_2)^2 Big) Bigg].$$
To facilitate the analysis, let $mathscrB equiv 1,2 ^n$ be the set of all $n$-length binary vectors of values of the two means. For any vector $mathbfb in mathscrB$ let $mathcalS(mathbfb) equiv i=1,...,n $ be the set of index values that use the first mean. With the improper uniform prior $pi (mu_1, mu_2) propto 1$ you get the posterior:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&propto L_mathbfx(mu_1,mu_2) \[6pt]
&propto prod_i=1^n Bigg[ exp Big( -frac12 (x_i-mu_1)^2 Big) + exp Big( -frac12 (x_i-mu_2)^2 Big) Bigg] \[6pt]
&= sum_mathbfb in mathscrB exp Big( -frac12 sum_i=1^n (x_i-mu_b_i)^2 Big) \[6pt]
&= sum_mathbfb in mathscrB exp Big( -frac12 sum_i in mathcalS(mathbfb) (x_i-mu_1)^2 Big) exp Big( -frac12 sum_i notin mathcalS(mathbfb) (x_i-mu_2)^2 Big). \[6pt]
endaligned endequation$$
Now, letting $barx_1^(k) (mathbfb) equiv sum_i in mathcalS(mathbfb) x_i^k$ and $barx_2^(k) (mathbfb) equiv sum_i notin mathcalS(mathbfb) x_i^k$ we have:
$$beginequation beginaligned
exp Big( -frac12 sum_i in mathcalS(mathbfb) (x_i-mu_1)^2 Big)
&= exp Big( -frac12 Big( mu_1^2 + 2 mu_1 sum_i in mathcalS(mathbfb) x_i + sum_i in mathcalS(mathbfb) x_i^2 Big) Big) \[6pt]
&= exp Big( -frac12 ( mu_1^2 + 2 mu_1 barx_1^(1) (mathbfb) + barx_1^(2) (mathbfb) ) Big) \[10pt]
&= exp Big( -frac12 ( barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 ) Big)
exp Big( -frac12 (mu_1 - barx_1^(1) (mathbfb))^2 Big) \[10pt]
&propto exp Big( -frac12 ( barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 ) Big) cdot textN(mu_1 | barx_1^(1) (mathbfb), 1), \[10pt]
exp Big( -frac12 sum_i notin mathcalS(mathbfb) (x_i-mu_2)^2 Big) &propto exp Big( -frac12 ( barx_2^(2) (mathbfb) - barx_2^(1) (mathbfb)^2 ) Big) cdot textN(mu_2 | barx_2^(1) (mathbfb), 1). \[10pt]
endaligned endequation$$
So, letting $H(mathbfb) equiv barx_1^(2) (mathbfb) - barx_1^(1) (mathbfb)^2 + barx_2^(2) (mathbfb) - barx_2^(1) (mathbfb)^2$ we have:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&propto sum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big) cdot textN(mu_1 | barx_1^(1) (mathbfb), 1) cdot textN(mu_2 | barx_2^(1) (mathbfb), 1). \[6pt]
endaligned endequation$$
This kernel has a finite integral, and yields the proper posterior density:
$$beginequation beginaligned
pi (mu_1, mu_2 | mathbfx)
&= fracsum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big) cdot textN(mu_1 sum_mathbfb in mathscrB exp Big( -frac12 H(mathbfb) Big). \[6pt]
endaligned endequation$$
answered 2 hours ago
Ben
17.5k22286
17.5k22286
add a comment |Â
add a comment |Â
aastha agrrawal is a new contributor. Be nice, and check out our Code of Conduct.
aastha agrrawal is a new contributor. Be nice, and check out our Code of Conduct.
aastha agrrawal is a new contributor. Be nice, and check out our Code of Conduct.
aastha agrrawal is a new contributor. Be nice, and check out our Code of Conduct.
Sign up or log in
StackExchange.ready(function ()
StackExchange.helpers.onClickDraftSave('#login-link');
);
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
StackExchange.ready(
function ()
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fstats.stackexchange.com%2fquestions%2f375699%2fgaussian-mixture-model-does-an-improper-uniform-prior-give-a-proper-posterior%23new-answer', 'question_page');
);
Post as a guest
Sign up or log in
StackExchange.ready(function ()
StackExchange.helpers.onClickDraftSave('#login-link');
);
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Sign up or log in
StackExchange.ready(function ()
StackExchange.helpers.onClickDraftSave('#login-link');
);
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Sign up or log in
StackExchange.ready(function ()
StackExchange.helpers.onClickDraftSave('#login-link');
);
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Sign up using Google
Sign up using Facebook
Sign up using Email and Password