BEAT - sbhatta/iqm_train

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

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
featureSet	system/array_2d_floats/1	Input
class	system/text/1	Input
subspace	tutorial/linear_machine/1	Output

import numpy
import bob
import bob.learn.linear

#same functionality as version 2, but implemented using bob-v2.

class Algorithm:

def __init__(self):
        self.data = [[], []]

def process(self, inputs, outputs):
        
        # accumulates the input data in different
        # containers for hit or miss
        flat_image = inputs["featureSet"].data.value
        if inputs["class"].data.text == 'real':
            self.data[1].append(flat_image) #positive/real class
        else:
            self.data[0].append(flat_image) #negative/attack class
        
        if not(inputs.hasMoreData()):
            # trains the LDA projection
            self.data[0] = numpy.vstack(self.data[0])
            self.data[1] = numpy.vstack(self.data[1])
            
            #first, normalize the input features
            trSet = numpy.vstack((self.data[1], self.data[0]))
            trMean = trSet.mean(axis=0)
            trStd = trSet.std(axis=0)
            posTS = self.data[1] - trMean
            posTS /= trStd
            negTS = self.data[0] - trMean
            negTS /=trStd
            del trSet

trainer = bob.learn.linear.FisherLDATrainer(use_pinv=True)
            machine, eigen_values = trainer.train([negTS, posTS])
            machine.resize(machine.shape[0], 1)
            # save the normalization parameters
            machine.input_subtract = trMean
            machine.input_divide = trStd
            
            #using training-data scores to 'set the sign' of the classifier
            flipSign = 1
            scores_train_pos = machine(posTS)[:,0]
            scores_train_neg = machine(negTS)[:,0]
            medPosScore = numpy.median(scores_train_pos)
            medNegScore = numpy.median(scores_train_neg)
            if medPosScore < medNegScore : flipSign = -1
            machine.weights = flipSign*machine.weights

# outputs data
            outputs["subspace"].write({
                'input_subtract': machine.input_subtract,
                'input_divide':   machine.input_divide,
                'weights':        machine.weights,
                'biases':         machine.biases,
            })

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 will run Linear Discriminant Analysis [LDA] for a binary classification problem using images as inputs.

Inputs:: featureSet: a 2d float array of size Nxd, where N is the number of patterns and d is the length of each pattern. class: a text label. The label can take one of two values: 'real' or 'attack'

[LDA]

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

Experiments

Attestation:

Privacy:

Status:

			Updated	Name	Databases/Protocols	Analyzers
				sbhatta/sbhatta/iqm-face-antispoofing-test/2/replay2-antispoofing-iqm-lda	replay/2@grandtest	sbhatta/iqm_spoof_eer_analyzer/9

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 sbhatta iqm_train_lda 8

Endpoint Groups 1

Group: main

Experiments

algorithms

sbhatta

iqm_train_lda

8