BEAT - tutorial/mlp

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.

Unnamed group

Endpoint Name	Data Format	Nature
class_id	system/uint64/1	Input
image	system/array_2d_floats/1	Input
model	tutorial/mlp/1	Output

Parameters 3

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

Name	Type	Default
number-of-hidden-units	uint32	10
number-of-iterations	uint32	50
seed	uint32	0

import bob
import numpy

class Algorithm:

def __init__(self):
      self.data = []
      self.data_id = []
      self.number_of_hidden_units = 10
      self.number_of_iterations = 50
      self.seed = 0

def setup(self, parameters):
      self.number_of_hidden_units = int(parameters.get('number-of-hidden-units',
                                                   self.number_of_hidden_units))
      self.number_of_iterations = int(parameters.get('number-of-iterations',
                                                   self.number_of_iterations))
      self.seed = int(parameters.get('seed', 0))
      return True

def process(self, inputs, outputs):
      self.data.append(inputs["image"].data.value.flatten())
      y = -1*numpy.ones((1, 10), dtype=float)
      y[0, inputs["class_id"].data.value] = 1.
      self.data_id.append(y)

if not(inputs.hasMoreData()):
        self.data = numpy.vstack(self.data)
        self.data_id = numpy.vstack(self.data_id)
        
         # Prepares MLP machine
        if self.number_of_hidden_units == 0:
          machine = bob.machine.MLP((28*28, 10))
        else:
          machine = bob.machine.MLP((28*28, self.number_of_hidden_units, 10))
        rng = bob.core.random.mt19937(self.seed)
        machine.randomize(rng)

# Prepares MLP R-Prop trainer
        batch_size = self.data.shape[0]
        trainer = bob.trainer.MLPRPropTrainer(batch_size, bob.trainer.SquareError(bob.machine.HyperbolicTangentActivation()))
        trainer.initialize(machine)
        # Launch training
        for it in range(self.number_of_iterations):
          trainer.train(machine, self.data, self.data_id)
 
        #import ipdb; ipdb.set_trace() 
        weights = []
        for v in machine.weights: weights.append({'value': v})
        biases = []
        for v in machine.biases: biases.append({'value': v})
        outputs["model"].write({'input_subtract': machine.input_subtract,
                                'input_divide': machine.input_divide,
                                'weights': weights,
                                'biases': 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 implements a training procedure for a multi-layer perceptron (MLP), a neural network architecture that has some well-defined characteristics such as a feed-forward structure. This algorithm assumes 0 or 1 hidden layer, and the training procedure is based on R-prop [Ri93].

This implementation relies on the Bob library.

The inputs are:

image: a two-dimensional array of floats (64 bits), which is flattened for

the training procedure
class_id: an identifier for the class of the image, such that supervised

training is possible

The output model is the MLP in a Bob-related format.

[Ri93]

Riedmiller, H. Braun. A Direct Adaptive Method for Faster Backpropagation Learning: The RPROP Algorithm. IEEE International Conference on Neural Networks 1993, 586-591

Experiments

Attestation:

Privacy:

Status:

			Updated	Name	Databases/Protocols	Analyzers
				smarcel/tutorial/digit/2/mnist-mlp-nhu10-niter100-seed2001	mnist/1@idiap	tutorial/multiclass_postperf/2

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 tutorial mlp_training 2

Endpoint Groups 1

Unnamed group

Parameters 3

Experiments

algorithms

tutorial

mlp_training

2