Esempi in Python per RecurrentNeuralNetwork

Esempio n. 1

File: recurrentTest.py Progetto: openube/pyNeuralNetwork

from nn.components.layers.RecurrentLayer import *
from nn.components.layers.SigmoidLayer import *
from nn.components.layers.InputLayer import *
from nn.components.connections.FullConnection import *
from nn.components.connections.Bias import *
from nn.components.objectives.MSEObjective import *
from nn.models.RecurrentNeuralNetwork import *
from trainers.SGD import SGDTrainer

import numpy as np

# Build a network

net = RecurrentNeuralNetwork()

inp = InputLayer(1)
rec = RecurrentLayer(5)
out = SigmoidLayer(1)

rec.zeroInitialHistoryBatch(2)

i2r = FullConnection(inp.output, rec.input)
r2o = FullConnection(rec.output, out.input)
rbias = Bias(rec.input)
obias = Bias(out.input)

tgt = InputLayer(1)
obj = MSEObjective(out.output, tgt.output)


i2r.randomize()

Esempio n. 2

from nn.components.layers.RecurrentLayer import *
from nn.components.layers.SigmoidLayer import *
from nn.components.layers.InputLayer import *
from nn.components.connections.FullConnection import *
from nn.components.connections.Bias import *
from nn.components.objectives.MSEObjective import *
from nn.models.RecurrentNeuralNetwork import *
from trainers.SGD import SGDTrainer

import numpy as np

# Build a network

net = RecurrentNeuralNetwork()

inp = InputLayer(1)
rec = RecurrentLayer(5)
out = SigmoidLayer(1)

rec.zeroInitialHistoryBatch(2)

i2r = FullConnection(inp.output, rec.input)
r2o = FullConnection(rec.output, out.input)
rbias = Bias(rec.input)
obias = Bias(out.input)

tgt = InputLayer(1)
obj = MSEObjective(out.output, tgt.output)

i2r.randomize()
r2o.randomize()

Esempio n. 3

File: gaitRNNmodel.py Progetto: reedanders/pyNeuralNetwork

def buildNetwork(rnn_size):
   """
   Build an RNN with softmax output
   """

   net = RecurrentNeuralNetwork()

   input_layer = InputLayer(27)       # Input is 27 unit vector
   recurrent_layer = RecurrentLayer(rnn_size)
   output_layer = SoftmaxLayer(34)

   input_to_recurrent = FullConnection(input_layer.output, recurrent_layer.input)
   recurrent_to_output = FullConnection(recurrent_layer.output, output_layer.input)
   recurrent_bias = Bias(recurrent_layer.input)
   output_bias = Bias(output_layer.input) 

   target_layer = InputLayer(34)      # 34 classes!
   objective_layer = CrossEntropyObjective(output_layer.output, target_layer.output)

   net.addLayer(input_layer)
   net.addLayer(target_layer)
   net.addConnection(input_to_recurrent)
   net.addConnection(recurrent_bias)
   net.addRecurrentLayer(recurrent_layer)
   net.addConnection(recurrent_to_output)
   net.addConnection(output_bias)
   net.addLayer(output_layer)
   net.addObjective(objective_layer)

   net.setInputLayer(input_layer)
   net.setTargetLayer(target_layer)
   net.setOutputLayer(output_layer)
   net.setObjective(objective_layer)

   net.randomize()

   return net