Difference between revisions of "MnM Modules"

Revision as of 17:01, 3 May 2009 (view source) Norbert (talk | contribs) ← Older edit		Revision as of 17:01, 3 May 2009 (view source) Norbert (talk | contribs) Newer edit →
Line 28:		Line 28:
	* mnm.xmul		* mnm.xmul
−
−	{{Module |
−	| name=ftm.mess
−	| brief=
−	| descr=
−	| arguments=none
−	| attributes=#triggerall <bool: flag> - all inlets trigger evaluation and output<br>#loadbang <bool: flag> - message box outputs at loadbang<br>#untuple <bool: flag> - single tuples are output as lists<br>
−	| messages=postdoc - post external doc to console<br>
−	| inlets=1 - input list<br>
−	| outlets=1 - message outlet<br>
−	}}
	{{Module |		{{Module |

---

Revision as of 17:01, 3 May 2009

• mnm.biqoefs
• mnm.biquad
• mnm.delta
• mnm.diag
• mnm.dtw
• mnm.gmmem
• mnm.hist
• mnm.knn
• mnm.lu
• mnm.mahalanobis
• mnm.mean
• mnm.meanstd
• mnm.median
• mnm.minmax
• mnm.moments
• mnm.nmd
• mnm.nmf
• mnm.obsprob
• mnm.onepole
• mnm.qr
• mnm.stats
• mnm.submat
• mnm.sum
• mnm.svd
• mnm.transpose
• mnm.viterbi
• mnm.xdist2
• mnm.xmul

---

mnm.biqoefs		Biquad coefficients
		Calculates biquad coefficients for various filter types.
		arguments: none attributes: mode - filter type (default is lowpass) f0 - cutoff or centre frequency (default is half Nyquist frequency) unit - unit for f0: ratio to Nyquist frequency (default) or in Hertz sr - sample rate for f0 in Hertz (default is 44100.) q - quality/resonance (default is 1.) qnorm - if qnorm = 1, divide q by 1./sqrt(2.) so as to get a monotonic filter response with q = 1. (instead of 1./sqrt(2.), which is the default with qnorm = 0) gain - linear gain (default is 1.) coefsas - output coefficients as an fmat or as a list (default is fmat) out - biquad coefficients messages: postdoc - post external doc to console bang - bang to output coefficients out - biquad coefficients inlets: none outlets: 1 - biquad coefficients

---

mnm.biquad		Biquad filtering
		Calculates biquad filtering over vectors (rows or columns) or stream of values (of any dimension).
		arguments: set the inputs initial size and numbers attributes: mode <'df1' | 'df2t'> - set biquad structure (default 'df1') dim - set the dimension on which to operate: col, row, auto (default) or stream (element by element). out - filtered values outstate - output state messages: postdoc - post external doc to console insize - change the inputs size and numbers coefs - set the biquad coefficients getstate - get the biquad state clear - reset the internal state out - filtered values outstate - output state inlets: 1 - input values outlets: 1 - filtered values 2 - output state

---

mnm.delta		Inter-frame regression.
		Calculates derivative of incoming matrices or vectors.
		arguments: 1 - initialize the input size 2 - initialize the filter size attributes: insize - set the input size filtersize - set the filter size inadddel - add a delay to the delayed input norm - normalization mode 1 (default) or 0 outdelayed - output delayed inputs (in phase with deltas) out - output deltas outstate - internal values messages: postdoc - post external doc to console clear - clear the memory of inputs getstate - get the internal weights vector getnorm - get the normalization factor getring - get the input ring buffer getdelay - get the filter delay outdelayed - output delayed inputs (in phase with deltas) out - output deltas outstate - internal values inlets: 1 - multiply matrix with given right or left operand outlets: 1 - output delayed inputs (in phase with deltas) 2 - output deltas 3 - internal values

---

mnm.diag		Matrix diagonal
		Returns a copy of the diagonal of the incoming matrix in a vector.The length of the result is the minimum of the dimensions of the input.
		arguments: none attributes: out - output diagonal vector messages: postdoc - post external doc to console out - output diagonal vector inlets: 1 - input matrix outlets: 1 - output diagonal vector

---

mnm.dtw		Dynamic time warping.
		Calculates DTW on incoming matrix or vector.
		arguments: 1 - fmat to be used as right operand attributes: outa - reference to an external fmat to store A outb - reference to an external fmat to store B messages: postdoc - post external doc to console inlets: 1 - left hand side fmat operand 2 - right hand side fmat operand outlets: 1 - s1 2 - s2

---

mnm.gmmem		Expectation maximization for Gaussian mixture models
		GMM EM has to be documented.
		arguments: 1 - number of centers to use attributes: outcenters - reference to external fmat to store centers outcov - reference to external fmat to store covariance outpriors - reference to external fmat to store priors mode - (diagonal|full|spherical) covariance computation types criteria - criteria ncenters - number of centers messages: postdoc - post external doc to console inlets: 1 - fmat outlets: 1 - fmat centers 2 - fmat covariance 3 - fmat priors

---

mnm.hist		Histogram
		Calculates histogram of incoming matrix elements. The input matrix, list or vector element's occurences are counted in the given number of bins in between the min and max value
		arguments: 1 - number of bins attributes: out - output histogram vector bpf - output two-column fmat with bin indices and histogram values norm <symbol: off|max|sum> -- normalise histogram so that max or sum is 1 messages: postdoc - post external doc to console out - output histogram vector set_n - number of bins inlets: 1 - intput matrix or list outlets: 1 - output histogram vector 2 - output min data value 3 - output max data value

---

mnm.knn		K nearest neighbour search
		Find the k nearest neighbours and their distances to the query point in multi-dimensional data using an efficient multidimensional search tree with logarithmic time complexity.
		arguments: 1 - max number k of nearest neighbours to search 2 - max radius of nearest neighbours to search (0 for unlimited) attributes: setdata - matrix(N, D) of data setsigma - matrix(1, D) of sigma = 1/weights dmode - decomposition mode: orthogonal, hyperplane, pca mmode - pivot calculation mode: mean, middle, median sort - sort output by distance height - given tree height if positive, subtract from maxheight if negative outdist <fmat|fvec|list|jitter: distance(n, 1)> - distances to n <= k nearest neighbours outind <fmat|fvec|list|jitter: indices(n, 1)> - data row indices of n <= k nearest neighbours messages: postdoc - post external doc to console outdist <fmat|fvec|list|jitter: distance(n, 1)> - distances to n <= k nearest neighbours outind <fmat|fvec|list|jitter: indices(n, 1)> - data row indices of n <= k nearest neighbours setk - max number k of nearest neighbours to search setradius - max radius of nearest neighbours to search (0 for unlimited) getmeanvectors <fmat: out> - copy mean vectors for the M tree nodes to copy fmat(M, D) out getsplitplanes <fmat: out> - copy vectors perpendicular to the hyperplanes splitting the M tree nodes to fmat(M, D) out getprofile <dict: out> - copy profiling info to given dict and clear print [<symbol: 'info'|'raw'|'data'|'compact'|'nodes'|'profile'>] - print tree info of varying detail (default: nodes), print and clear profiling info if keyword 'profile' is given inlets: 1 <fmat|fvec|list|jitter: x(1, D)> - query vector to search k-nearest neighbours of 2 <fmat|fvec|list|jitter: data(N, D)> 3 <fmat|fvec|list|jitter: sigma(1, D)> outlets: 1 <fmat|fvec|list|jitter: distance(n, 1)> - distances to n <= k nearest neighbours 2 <fmat|fvec|list|jitter: indices(n, 1)> - data row indices of n <= k nearest neighbours

---

mnm.lu		Lower-upper decomposition
		Calculates LU decomposition on incoming matrix.
		arguments: none attributes: outl - L outu - U outpivot - pivot outx - X outdet - determinant messages: postdoc - post external doc to console determinant - computes determinant of decomposed fmat solve - solves system with incoming fmat and decomposed fmat outl - L outu - U outpivot - pivot outx - X outdet - determinant inlets: 1 - matrix to decompose outlets: 1 - L 2 - U 3 - pivot 4 - X 5 - determinant

---

mnm.mahalanobis		Mahalanobis distance.
		Calculates mahalanobis distance on incoming matrices or vectors.
		arguments: <matrix|vector|list: mean> <matrix|vector|list: covariance> - init mean and covariance attributes: out - mahalanobis distance messages: postdoc - post external doc to console set_mu - matrix of mean values set_sigma - matrix of covariance out - mahalanobis distance inlets: 1 <matrix|vector|list: query vector> 2 <matrix: mu> 3 <matrix: sigma> outlets: 1 - mahalanobis distance

---

mnm.mean		Mean filtering
		Calculates mean filtering over vectors (rows or columns) or stream of values (of any dimension).
		arguments: set the inputs initial size and numbers attributes: filtersize - set the maximum filter size (default is 0 for using the input size) dim - set the dimension on which to operate: col, row, auto (default) or stream (element by element). outtype - set the output type: fmat, float or auto (default, matches the input type). out - filtered values outstate - output state messages: postdoc - post external doc to console insize - change the inputs size and numbers getstate - get the mean state clear - reset the internal state out - filtered values outstate - output state inlets: 1 - input values outlets: 1 - filtered values 2 - output state

---

mnm.meanstd		Mean and standard deviation
		Calculates the arithmetic mean and standard deviation of each column or row (depending on 'mode' argument) as one row or column vector
		arguments: 1 <1|2|'row'|'col': mode switch> compute over rows or columns attributes: mode <1|2|'row'|'col': mode switch> compute over rows or columns [rows] scalar <bool: switch> output a simple float value (instead of 1 x 1 matrix) for scalar results [on] outmean - mean output vector or value outstd - standard deviation output vector or value messages: postdoc - post external doc to console outmean - mean output vector or value outstd - standard deviation output vector or value inlets: 1 - input matrix outlets: 1 - mean output vector or value 2 - standard deviation output vector or value

---

mnm.median		Median filtering
		Calculates median filtering over vectors (rows or columns) or stream of values (of any dimension).
		arguments: set the inputs initial size and numbers attributes: filtersize - set the maximum filter size (default is 0 for using the input size) dim - set the dimension on which to operate: col, row, auto (default) or stream (element by element). outtype - set the output type: fmat, float or auto (default, matches the input type). out - filtered values outstate - output state messages: postdoc - post external doc to console insize - change the inputs size and numbers getstate - get the median state clear - reset the internal state out - filtered values outstate - output state inlets: 1 - input values outlets: 1 - filtered values 2 - output state

---

mnm.minmax		Minimum and maximum
		Calculates the min, index of min, max, index of max of each column or row (depending on argument) as one row or column vector
		arguments: 1 - (1|2|row|col) sum over rows or columns attributes: mode - (1|2|row|col) sum over rows or columns scalar <bool: switch> output a simple float value (instead of 1 x 1 matrix) for scalar results [on] outmin - min outargmin - argmin outmax - max outargmax - argmax messages: postdoc - post external doc to console outmin - min outargmin - argmin outmax - max outargmax - argmax inlets: 1 - incoming matrix, vector, or list outlets: 1 - min 2 - argmin 3 - max 4 - argmax

---

mnm.moments		Statistical moments
		Calculates moments from first to specified order.
		arguments: 1 <num: order> - moments maximum order [1] attributes: std <'0'|'1': switch> - compute the standards moments for orders > 2 [1] sumasfloat <'0'|'1': switch> - enable/disable float sum output [0] out - moments outsum - input sums mode - (1|2|row|col) calculate over rows (same as 1) or columns (default, same as 2) for multicolumn inputs. messages: postdoc - post external doc to console order <num: order> - set moments maximum order out - moments outsum - input sums inlets: 1 - input vector outlets: 1 - moments 2 - input sums

---

mnm.nmd		Non-zero matrix decomposition
		Calculates NMD on incoming matrix.
		arguments: none attributes: outh - out H criteria - criteria sH - sH itermax - itermax messages: postdoc - post external doc to console inlets: 1 - fmat 2 - fmat 3 - outlets: 1 - fmat

---

mnm.nmf		Non-zero matrix factorization
		Calculates NMF on incoming matrix.
		arguments: 1 - number of components attributes: outw - reference to external fmat to store W outh - reference to external fmat to store H criteria - (float) stopping criteria rdim - number of components itermax - maximum number of iterations messages: postdoc - post external doc to console inlets: 1 - fmat to be decomposed outlets: 1 - W 2 - H

---

mnm.obsprob		Obsprob
		Obsprob has to be documented.
		arguments: none attributes: none messages: postdoc - post external doc to console inlets: 1 - ref. to observation frame : fmat [D (=feature space dim) , 1] 2 - ref. to states matrix outlets: 1 - log(B) : fmat [K (=nb of states) , 1] 2 - test

---

mnm.onepole		Onepole filtering
		Calculates onepole filtering (low-pass or high-pass) over vectors (rows or columns) or stream of values (of any dimension).
		arguments: set the inputs initial size and numbers attributes: f0 - set the onepole f0, normalised by the Nyquist frequency (default is 1.) dim - set the dimension on which to operate: col, row, auto (default) or stream (element by element). mode - filter type (feault is lowpass). out - filtered values outstate - output state messages: postdoc - post external doc to console getstate - get the onepole state clear - reset the internal state out - filtered values outstate - output state insize - change the inputs size and numbers inlets: 1 - input values outlets: 1 - filtered values 2 - output state

---

mnm.qr		Orthogonal-right decomposition
		Calculates QR decomposition on incoming matrix.
		arguments: none attributes: outq - Q outr - R outx - X messages: postdoc - post external doc to console solve - solve system with QR decomposition outq - Q outr - R outx - X inlets: 1 - fmat to be decomposed outlets: 1 - Q 2 - R 3 - X

---

mnm.stats		Stats
		Stats has to be documented.
		arguments: output stats attributes: norm - switch normalize messages: postdoc - post external doc to console bang - output stats clear - clear histogram set - set histogram vector (fmat) inlets: 1 - data input outlets: 1 - average 2 - standard deviation 3 - count

---

mnm.submat		Sub-matrix
		Copies sub-matrix outof incoming matrix.
		arguments: none attributes: out - sub-matrix messages: postdoc - post external doc to console begin - start of submatrix coordinates end - end of submatrix coordinates out - sub-matrix inlets: 1 - fmat outlets: 1 - sub-matrix

---

mnm.sum		Matrix sum
		Calculates sum over entire matrix, matrix rows or matrix columns.
		arguments: 1 - (1|2|row|col) sum over rows or columns attributes: out - sum of fmat mode - 'row'|'col'|1|2 -- perform sum over rows or columns type <symbol: 'float'|'fmat'> -- always output a matrix even for scalar results messages: postdoc - post external doc to console out - sum of fmat inlets: 1 - incoming matrix to be summed outlets: 1 - sum of fmat

---

mnm.svd		Singular value decomposition
		Calculates SVD decomposition on incoming matrix.
		arguments: 1 - number of singular values attributes: mode - (auto|manual) automatically eliminate negligible singular values outu - output matrix for U outs - output matrix for S outvt - output matrix for V' messages: postdoc - post external doc to console outu - output matrix for U outs - output matrix for S outvt - output matrix for V' inlets: 1 - matrix to be decomposed by SVD outlets: 1 - output matrix for U 2 - output matrix for S 3 - output matrix for V'

---

mnm.transpose		Transpose
		Calculates transposed matrix.
		arguments: none attributes: out - transposed matrix messages: postdoc - post external doc to console out - transposed matrix inlets: 1 - matrix to be transposed outlets: 1 - transposed matrix

---

mnm.viterbi		Viterbi algorithm
		Calculates Viterbi path on incoming matrices or vector.
		arguments: none attributes: line - on|off line verbose - verbose or not (0|1) latency - maximum latency (nb of frames) get - debug : get intern values. messages: postdoc - post external doc to console reinit - input message reinit bang - bang to decode locpaths - get locpaths matrix inlets: 1 - bang to decode, or reinit message to reset PSIs and DELTAs 2 - observation matrix logB[T,K] 3 - state prior distribution Pi[1,K] 4 - state transition matrix A[K,K] outlets: 1 - decoded best path 2 - debug

---

mnm.xdist2		Square of Euclidean distance
		Calculates the square of the Euclidean distance between a vector and each line of a matrix.
		arguments: matrix to be used as right operand attributes: swap - (yes|no) swaps operands out - result squared distance matrix messages: postdoc - post external doc to console out - result squared distance matrix inlets: 1 <matrix: left operand> 2 <matrix: right operand> outlets: 1 - result squared distance matrix

---

mnm.xmul		Matrix multiplication
		Calculates matrix multiplication as in out = left x right. The left and right operands of the matrix multiplication are given by the respective inlets unless the swap option is enabled.The dimensions of the resulting output matrix are corresponding to the minimum dimensions of the two operators.
		arguments: set right or (with swap enabled) left matrix multiplication operand attributes: out - output matrix object swap <bool: switch> swaps operands messages: postdoc - post external doc to console out - output matrix object inlets: 1 - multiply matrix with given right or left operand 2 - set right or (with swap enabled) left matrix multiplication operand outlets: 1 - output matrix object