This algorithm is a legacy one. The API has changed since its implementation. New versions and forks will need to be updated.

This algorithm is splittable

Endpoint Groups 1

Algorithms have at least one input and one output. All algorithm endpoints are organized in groups. Groups are used by the platform to indicate which inputs and outputs are synchronized together. The first group is automatically synchronized with the channel defined by the block in which the algorithm is deployed.

Group: main

Endpoint Name	Data Format	Nature
features	system/array_2d_floats/1	Input
client_id	system/uint64/1	Input
subspace	chichan/a-collection-of-linear_machines/1	Output

Parameters 2

Parameters allow users to change the configuration of an algorithm when scheduling an experiment

Name	Description	Type	Default	Range/Choices
percentage-of-pca-energy		float32	0.949999988079071
number-of-lda-components		uint32	30

import bob
import numpy

class Algorithm:

def __init__(self):
        self.percentage_of_pca_energy = 0.98
        self.number_of_lda_components = 30
        self.data = {}

def setup(self, parameters):
        self.percentage_of_pca_energy = parameters.get('percentage-of-pca-energy',
                                                   self.percentage_of_pca_energy)
        self.number_of_lda_components = parameters.get('number-of-lda-components',
                                                   self.number_of_lda_components)
        return True

def _project_data_for_lda(self, machine,i):
        tdata = []
        for client_id, client_files in self.data.iteritems():
            # at least two files per client are required!
            if len(client_files) < 2:
                # "Skipping client since the number of client files is only %d" %len(client_files)
                continue
            tdata.append(numpy.vstack([machine(feature) for feature in client_files[i]]))
        return tdata

def _perform_pca(self, pca_machine, training_set):
        # "Perform PCA on data"
        data = []
        for client_features in training_set:
            data.append(numpy.vstack([machine(feature) for feature in client_features]))
        return data

def process(self, inputs, outputs):
        features = inputs['features'].data.value

c_id = inputs["client_id"].data.value
        machines=[]
        if c_id in self.data.keys():
            for idx in xrange(features.shape[0]) :
                self.data[c_id][idx].append(features[idx])
        else: 
            feature_dict={}
            for idx in xrange(features.shape[0]) :
                feature_dict[idx]=[features[idx]]           
            self.data[c_id] = feature_dict
        if not(inputs.hasMoreData()):            
#            numpy.lib.asarray_chkfinite(self.data[self.data.keys()[0]][self.data[self.data.keys()[0]].keys()[0]][0])
#            assert(len(self.data[self.data.keys()[0]][self.data[self.data.keys()[0]].keys()[0]][0])==0,'data[c_id][0]==0')
#            assert(len(self.data[self.data.keys()[0]][self.data[self.data.keys()[0]].keys()[0]][0])>=59,'data[c_id][0]>=59')
#            assert(len(self.data[self.data.keys()[0]][self.data[self.data.keys()[0]].keys()[0]][0])<=59,'data[c_id][0]>=59')
            lda_trainer = bob.trainer.FisherLDATrainer()
            trainer = bob.trainer.PCATrainer()    
            for i in self.data[self.data.keys()[0]].keys():
                # PCA
                data_pca = numpy.vstack([self.data[j_id][i] for j_id in self.data.keys()])
#                if data_pca.shape[1]==59:
#                    print cc

pca_machine, eigen_values = trainer.train(data_pca)
                del data_pca # Reduce memory usage
                #remove zeros
                non_zeros_idx=len(numpy.where(eigen_values>0)[0])
                n_eigen_values=eigen_values.copy()
                n_eigen_values.resize(non_zeros_idx)
                assert(eigen_values == n_eigen_values,'eigen_values == n_eigen_values')
                cum_eigen_values=numpy.cumsum(n_eigen_values/sum(n_eigen_values))
                number_of_pca_components=numpy.where(cum_eigen_values >self.percentage_of_pca_energy)[0]

if number_of_pca_components[0] <2 :
                    raise ValueError('number_of_pca_components_%d[0]:%d'%(i,number_of_pca_components[0]))                        
                pca_machine.resize(pca_machine.shape[0], (number_of_pca_components[0]+1))
                # outputs data
                numpy.lib.asarray_chkfinite(pca_machine.weights)
               
                # LDA
                data_lda = self._project_data_for_lda(pca_machine,i)

lda_machine, lda_variances = lda_trainer.train(data_lda)
                del data_lda # Reduce memory usage
                if lda_machine.shape[1]>lda_machine.shape[0] :
                    lda_machine.resize(lda_machine.shape[0], lda_machine.shape[0])                
                
                if lda_machine.shape[1]>int(self.number_of_lda_components) :
                    lda_machine.resize(lda_machine.shape[0], int(self.number_of_lda_components))
                    
                if numpy.isnan(lda_machine.weights).any() or numpy.isinf(lda_machine.weights).any():
                    raise ValueError([lda_variances,number_of_LDA_components,number_of_pca_components[0]+1,lda_machine.shape, len(number_of_LDA_components)])                      
                
                   
                pcalda_subspace = numpy.dot(pca_machine.weights,lda_machine.weights)
                if numpy.isnan(pcalda_subspace).any() or numpy.isinf(pcalda_subspace).any():
                    raise ValueError(pcalda_subspace.shape)

machines.append({
                        'input_subtract': pca_machine.input_subtract,
                        'input_divide':   pca_machine.input_divide,
                        'weights':        pcalda_subspace,
                        'biases':         lda_machine.biases
                    })
        # outputs data
            outputs["subspace"].write({"machine_array": machines})            
            print machines[0]
            print machines[-1]

return True

The code for this algorithm in Python
The ruler at 80 columns indicate suggested POSIX line breaks (for readability).
The editor will automatically enlarge to accomodate the entirety of your input
Use keyboard shortcuts for search/replace and faster editing. For example, use Ctrl-F (PC) or Cmd-F (Mac) to search through this box

This algorithm performs principal component analysis (PCA) [PCA] on a given dataset using the singular value decomposition (SVD) [SVD], followed by linear discriminant analysis (LDA) [LDA].

This implementation relies on the `Bob <http://www.idiap.ch/software/bob>`_ library.

The inputs are:

image: A training set of floating point vectors as a two-dimensional array of floats (64 bits), the number of rows corresponding to the number of training samples, and the number of columns to the dimensionality of the training samples.
client_id: Client (class/subject) identifier as an unsigned 64 bits integer.

The outputs are subspace_pca and subspace_lda for the PCA and LDA transformation, respectively.

[SVD]

http://en.wikipedia.org/wiki/Singular_value_decomposition

[PCA]

http://en.wikipedia.org/wiki/Principal_component_analysis

[LDA]

http://en.wikipedia.org/wiki/Linear_discriminant_analysis

Docutils System Messages

System Message: ERROR/3 (<string>, line 5); backlink

Unknown target name: "bob <http://www.idiap.ch/software/bob>".

Experiments

Attestation:

Privacy:

Status:

Name	Databases/Protocols	Analyzers
smarcel/chichan/full_pre_mlbphs_projection/2/mobio-f_TT_MLBPH_PCA98_LDA300_postperf-iso	mobio/2@female	tutorial/eerhter_postperf_iso/1
smarcel/chichan/full_pre_mlbphs_projection/2/mobio-m_TT_MLBPH_PCA98_LDA300_postperf-iso	mobio/2@male	tutorial/eerhter_postperf_iso/1
chichan/chichan/full_pre_mlbphs_projection/2/Prep_MLBPH_XM2VTS_LDA	xm2vts/1@lp1,xm2vts/1@darkened-lp1	tutorial/eerhter_postperf/1
chichan/chichan/full_pre_mlbphs_projection/2/Prep_MLBPH_XM2VTS_no_uniform_p98LDA	xm2vts/1@lp1,xm2vts/1@darkened-lp1	tutorial/eerhter_postperf/1

Scientific Python 2.7 (0.0.4) 8

This table shows the number of times this algorithm has been successfully run using the given environment. Note this does not provide sufficient information to evaluate if the algorithm will run when submitted to different conditions.

algorithms chichan comp_combine-lda-and-pca-weight-together 110

Endpoint Groups 1

Group: main

Parameters 2

Docutils System Messages

Experiments

algorithms

chichan

comp_combine-lda-and-pca-weight-together

110