Mercurial > hg > camir-aes2014

annotate toolboxes/FullBNT-1.0.7/nethelp3.3/mdn.htm @ 0:e9a9cd732c1e tip

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
first hg version after svn
author wolffd
date Tue, 10 Feb 2015 15:05:51 +0000
parents
children
rev   line source
wolffd@0 1 <html>
wolffd@0 2 <head>
wolffd@0 3 <title>
wolffd@0 4 Netlab Reference Manual mdn
wolffd@0 5 </title>
wolffd@0 6 </head>
wolffd@0 7 <body>
wolffd@0 8 <H1> mdn
wolffd@0 9 </H1>
wolffd@0 10 <h2>
wolffd@0 11 Purpose
wolffd@0 12 </h2>
wolffd@0 13 Creates a Mixture Density Network with specified architecture.
wolffd@0 14
wolffd@0 15 <p><h2>
wolffd@0 16 Synopsis
wolffd@0 17 </h2>
wolffd@0 18 <PRE>
wolffd@0 19 net = mdn(nin, nhidden, ncentres, dimtarget)
wolffd@0 20 net = mdn(nin, nhidden, ncentres, dimtarget, mixtype, ...
wolffd@0 21 prior, beta)
wolffd@0 22 </PRE>
wolffd@0 23
wolffd@0 24
wolffd@0 25 <p><h2>
wolffd@0 26 Description
wolffd@0 27 </h2>
wolffd@0 28 <CODE>net = mdn(nin, nhidden, ncentres, dimtarget)</CODE> takes the number of
wolffd@0 29 inputs,
wolffd@0 30 hidden units for a 2-layer feed-forward
wolffd@0 31 network and the number of centres and target dimension for the
wolffd@0 32 mixture model whose parameters are set from the outputs of the neural network.
wolffd@0 33 The fifth argument <CODE>mixtype</CODE> is used to define the type of mixture
wolffd@0 34 model. (Currently there is only one type supported: a mixture of Gaussians with
wolffd@0 35 a single covariance parameter for each component.) For this model,
wolffd@0 36 the mixture coefficients are computed from a group of softmax outputs,
wolffd@0 37 the centres are equal to a group of linear outputs, and the variances are
wolffd@0 38 obtained by applying the exponential function to a third group of outputs.
wolffd@0 39
wolffd@0 40 <p>The network is initialised by a call to <CODE>mlp</CODE>, and the arguments
wolffd@0 41 <CODE>prior</CODE>, and <CODE>beta</CODE> have the same role as for that function.
wolffd@0 42 Weight initialisation uses the Matlab function <CODE>randn</CODE>
wolffd@0 43 and so the seed for the random weight initialization can be
wolffd@0 44 set using <CODE>randn('state', s)</CODE> where <CODE>s</CODE> is the seed value.
wolffd@0 45 A specialised data structure (rather than <CODE>gmm</CODE>)
wolffd@0 46 is used for the mixture model outputs to improve
wolffd@0 47 the efficiency of error and gradient calculations in network training.
wolffd@0 48 The fields are described in <CODE>mdnfwd</CODE> where they are set up.
wolffd@0 49
wolffd@0 50 <p>The fields in <CODE>net</CODE> are
wolffd@0 51 <PRE>
wolffd@0 52
wolffd@0 53 type = 'mdn'
wolffd@0 54 nin = number of input variables
wolffd@0 55 nout = dimension of target space (not number of network outputs)
wolffd@0 56 nwts = total number of weights and biases
wolffd@0 57 mdnmixes = data structure for mixture model output
wolffd@0 58 mlp = data structure for MLP network
wolffd@0 59 </PRE>
wolffd@0 60
wolffd@0 61
wolffd@0 62 <p><h2>
wolffd@0 63 Example
wolffd@0 64 </h2>
wolffd@0 65 <PRE>
wolffd@0 66
wolffd@0 67 net = mdn(2, 4, 3, 1, 'spherical');
wolffd@0 68 </PRE>
wolffd@0 69
wolffd@0 70 This creates a Mixture Density Network with 2 inputs and 4 hidden units.
wolffd@0 71 The mixture model has 3 components and the target space has dimension 1.
wolffd@0 72
wolffd@0 73 <p><h2>
wolffd@0 74 See Also
wolffd@0 75 </h2>
wolffd@0 76 <CODE><a href="mdnfwd.htm">mdnfwd</a></CODE>, <CODE><a href="mdnerr.htm">mdnerr</a></CODE>, <CODE><a href="mdn2gmm.htm">mdn2gmm</a></CODE>, <CODE><a href="mdngrad.htm">mdngrad</a></CODE>, <CODE><a href="mdnpak.htm">mdnpak</a></CODE>, <CODE><a href="mdnunpak.htm">mdnunpak</a></CODE>, <CODE><a href="mlp.htm">mlp</a></CODE><hr>
wolffd@0 77 <b>Pages:</b>
wolffd@0 78 <a href="index.htm">Index</a>
wolffd@0 79 <hr>
wolffd@0 80 <p>Copyright (c) Ian T Nabney (1996-9)
wolffd@0 81 <p>David J Evans (1998)
wolffd@0 82
wolffd@0 83 </body>
wolffd@0 84 </html>