| MATLAB File Help: dml.hmm | View code for dml.hmm | Default Topics |
hmm Hidden Markov model with continuous observations.
DESCRIPTION
input X is of size repetitions x features x timepoints
note:
- nhidden specifies the number of hidden states; this is used only when Y
is absent or NaN
EXAMPLE
rand('seed',3); randn('seed',3);
nsamples = 1000; ncov = 10; ncycles = 10;
Y = repmat([ones(100,1); 2*ones(100,1)],[5 1]);
k = dml.hmm('verbose',true);
X = repmat(Y,[1 ncov]); X(X(:)==1) = randn(1,sum(X(:)==1)); X(X(:)==2) = 0.1+3*randn(1,sum(X(:)==2));
X = reshape(X',[1 size(X,2) size(X,1)]);
Y = reshape(Y',[1 size(Y,2) size(Y,1)]);
k = k.train(X,Y); % everything assumed observed
Z = k.test(X);
plot(squeeze(Y),'k-');
hold on;
plot(squeeze(Z),'ro');
REFERENCES
Pattern Recognition and Machine Learning, Bishop
BNT toolbox
DEVELOPER
Marcel van Gerven (m.vangerven@donders.ru.nl)| Superclasses | dml.method |
| Sealed | false |
| Construct on load | false |
| hmm | Hidden Markov model with continuous observations. |
| LL | likelihood |
| Sigma | conditional covariance |
| cov_prior | prior on the covariance matrix (diagonal term) |
| cov_type | covariance type (full, diag, spherical) |
| indims | dimensions of the input data (excluding the trial dim and time dim in time series data) |
| maxiter | max number of em iterations |
| mixmat | mixing matrix |
| mu | conditional means |
| nhidden | number of hidden states; state assumed observed if hidden |
| nmixture | number of gaussian mixture components |
| prior | class prior |
| restart | when false, starts at the previously learned parameters; needed for online learning and grid search |
| tied_cov | tie the covariances |
| transmat | transition matrix |
| verbose | whether or not to generate diagnostic output |
| likelihood | return the log likelihood of an observed sequence | |
| model | this method does not return a model | |
| test | Viterbi decoding; outputs most likely path per sequence | |
| train | data represented as repetitions x features x timepoints |