# @coda/ml

# gmmTrain

 gmmTrain(options: Object, source: Stream): Stream

Train a Gaussian Mixture Model for recognition from a stream of recording events.

WARNING

The XMM Training operators (gmmTrain, hmmTrain, hhmmTrain) use a web worker for training. Make sure to copy the file @coda/ml/dist/xmm.worker.js to the root of your application.

Parameter	Type	Default	Description
options	Object	{}	Training parameters
options.gaussians	Number	3	Number of gaussian components
options.regularizationAbs	Number	0.01	Absolute regularization
options.regularizationRel	Number	0.1	Relative regularization
options.covarianceMode	String	'full'	Type of covariance matrix
source	Stream		Source stream (recording events)

Returns Stream Stream of model parameters

Example

// Generate a smooth random signal
a = periodic(20)
.rand({ size: 2 })
.biquad({ f0: 2 })
.plot({ legend: 'Data Stream' });

// Setup a data recorder
b = a.recorder({ name: 'data' });

// Dynamically train when changes occur in the recorder
model = b.gmmTrain({ gaussians: 3 });

// Perform real-time recognition
c = a.gmmPredict({ model })
  .plot({ fill: 'bottom', stacked: true, legend: 'GMM-based recognition' });

# gmmPredict

 gmmPredict(options: Object, source: Stream): Stream

Real-time recognition using GMMsThe operator applies to the same data stream as the one used for training, and the model
parameters should be passed as a stream (model argument) generated by the gmmTrain
operator.

TODO

More details here...

Parameter	Type	Default	Description
options	Object	{}	Recognizer options
options.model	Stream	null	Stream of model parameters obtained through <a

  href="/ api / gmmTrain"
  target=""
>
  gmmTrain
</a>.|

|options.likelihoodWindow|string|1|Size of the likelihood smoothing window| |options.output|string|'smoothedLogLikelihoods'|Type of output data.| |source|Stream||input data stream| Returns Stream

Example

// Generate a smooth random signal
a = periodic(20)
.rand({ size: 2 })
.biquad({ f0: 2 })
.plot({ legend: 'Data Stream' });

// Setup a data recorder
b = a.recorder({ name: 'data' });

// Dynamically train when changes occur in the recorder
model = b.gmmTrain({ gaussians: 3 });

// Perform real-time recognition
c = a.gmmPredict({ model })
  .plot({ fill: 'bottom', stacked: true, legend: 'GMM-based recognition' });

# hmmTrain

 hmmTrain(options: Object, source: Stream): Stream

Train a Hidden Markov Model for recognition from a stream of recording events

WARNING

The XMM Training operators (gmmTrain, hmmTrain, hhmmTrain) use a web worker for training. Make sure to copy the file @coda/ml/dist/xmm.worker.js to the root of your application.

Parameter	Type	Default	Description
options	Object	{}	Training parameters
options.states	Number	5	Number of hidden states
options.gaussians	Number	1	Number of gaussian components per state
options.regularizationAbs	Number	0.01	Absolute regularization
options.regularizationRel	Number	0.1	Relative regularization
options.covarianceMode	String	'full'	Type of covariance matrix
source	Stream		Source stream (recording events)

Returns Stream Stream of model parameters

Example

// Generate a smooth random signal
a = periodic(20)
.rand({ size: 2 })
.biquad({ f0: 2 })
.plot({ legend: 'Data Stream' });

// Setup a data recorder
b = a.recorder({ name: 'data' });

// Dynamically train when changes occur in the recorder
model = b.hmmTrain({ states: 5 });

// Perform real-time recognition
c = a.hmmPredict({ model })
  .plot({ fill: 'bottom', stacked: true, legend: 'HMM-based recognition' });

# hmmPredict

 hmmPredict(options: Object, source: Stream): Stream

Real-time recognition using HMMsThe operator applies to the same data stream as the one used for training, and the model
parameters should be passed as a stream (model argument) generated by the hmmTrain
operator.

TODO

More details here...

Parameter	Type	Default	Description
options	Object	{}	Recognizer options
options.model	Stream	null	Stream of model parameters obtained through <a

  href="/ api / hmmTrain"
  target=""
>
  hmmTrain
</a>.|

Example

// Generate a smooth random signal
a = periodic(20)
.rand({ size: 2 })
.biquad({ f0: 2 })
.plot({ legend: 'Data Stream' });

// Setup a data recorder
b = a.recorder({ name: 'data' });

// Dynamically train when changes occur in the recorder
model = b.hmmTrain({ states: 5 });

// Perform real-time recognition
c = a.hmmPredict({ model })
  .plot({ fill: 'bottom', stacked: true, legend: 'HMM-based recognition' });

# hhmmTrain

 hhmmTrain(options: Object, source: Stream): Stream

Train a Hierarchical Hidden Markov Model for recognition from a stream of recording events.

WARNING

The XMM Training operators (gmmTrain, hmmTrain, hhmmTrain) use a web worker for training. Make sure to copy the file @coda/ml/dist/xmm.worker.js to the root of your application.

Parameter	Type	Default	Description
options	Object	{}	Training parameters
options.states	Number	5	Number of hidden states
options.gaussians	Number	1	Number of gaussian components per state
options.regularizationAbs	Number	0.01	Absolute regularization
options.regularizationRel	Number	0.1	Relative regularization
options.covarianceMode	String	'full'	Type of covariance matrix
source	Stream		Source stream (recording events)

Returns Stream Stream of model parameters

Example

// Generate a smooth random signal
a = periodic(20)
.rand({ size: 2 })
.biquad({ f0: 2 })
.plot({ legend: 'Data Stream' });

// Setup a data recorder
b = a.recorder({ name: 'data' });

// Dynamically train when changes occur in the recorder
model = b.hhmmTrain({ states: 5 });

// Perform real-time recognition
c = a.hhmmPredict({ model })
  .plot({ fill: 'bottom', stacked: true, legend: 'HMM-based recognition' });

# hhmmPredict

 hhmmPredict(options: Object, source: Stream): Stream

Real-time recognition using Hierarchical HMMsThe operator applies to the same data stream as the one used for training, and the model
parameters should be passed as a stream (model argument) generated by the hhmmTrain
operator.

TODO

More details here...

Parameter	Type	Default	Description
options	Object	{}	Recognizer options
options.model	Stream	null	Stream of model parameters obtained through <a

  href="/ api / hhmmTrain"
  target=""
>
  hhmmTrain
</a>.|

Example

// Generate a smooth random signal
a = periodic(20)
.rand({ size: 2 })
.biquad({ f0: 2 })
.plot({ legend: 'Data Stream' });

// Setup a data recorder
b = a.recorder({ name: 'data' });

// Dynamically train when changes occur in the recorder
model = b.hhmmTrain({ states: 5 });

// Perform real-time recognition
c = a.hhmmPredict({ model })
  .plot({ fill: 'bottom', stacked: true, legend: 'HMM-based recognition' });

# pcaTrain

 pcaTrain(source: Stream): Stream

Estimate Principal Component Analysis from a set of recordings

Parameter	Type	Default	Description
source	Stream		Recorder source

Returns Stream

# pcaPredict

 pcaPredict(pcaParamStream: Stream, source: Stream): Stream

Compute real-time PCA projection from a data stream.

Parameter	Type	Default	Description
pcaParamStream	Stream		PCA Parameter stream (from pcaTrain)
source	Stream		input data stream

Returns Stream

# scaleTrain

 scaleTrain(source: Stream): Stream

Return the extremum values of a recorder stream

Parameter	Type	Default	Description
source	Stream		Recorder source

Returns Stream Scalar stream (WHAT ?????? {"type":"linkReference","identifier":"min, max","referenceType":"shortcut","children":[{"type":"text","value":"min, max","position":{"start":{"line":1,"column":17,"offset":16},"end":{"line":1,"column":25,"offset":24},"indent":[]}}],"position":{"start":{"line":1,"column":16,"offset":15},"end":{"line":1,"column":26,"offset":25},"indent":[]}})

# scalePredict

 scalePredict(minmaxstream: Stream, source: Stream): Stream

Compute real-time scaling from a data stream.

Parameter	Type	Default	Description
minmaxstream	Stream		stream of extremum values (from scaleTrain)
source	Stream		input data stream

Returns Stream

← @coda/midi @coda/sensors →