Examples of RBM

Examples of com.github.neuralnetworks.architecture.types.RBM

37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
     * @see com.github.neuralnetworks.training.rbm.CDTrainerBase#updateWeights(com.github.neuralnetworks.architecture.Matrix, com.github.neuralnetworks.architecture.Matrix, com.github.neuralnetworks.architecture.Matrix, com.github.neuralnetworks.architecture.Matrix)
     * before each update the kernel update parameters are refreshed
     */
    @Override
    protected void updateWeights(Matrix posPhaseVisible, Matrix posPhaseHidden, Matrix negPhaseVisible, Matrix negPhaseHidden) {
  RBM rbm = getNeuralNetwork();

  int mbs = posPhaseHidden.getColumns();

  if (weightUpdatesKernel == null || weightUpdatesKernel.getMiniBatchSize() != mbs) {
      weightUpdatesKernel = new CDWeightUpdatesKernel(posPhaseVisible, posPhaseHidden, negPhaseVisible, negPhaseHidden, rbm.getMainConnections().getConnectionGraph(), getLearningRate(), getMomentum(), getl1weightDecay(), getl2weightDecay());
  }
  Environment.getInstance().getExecutionStrategy().execute(weightUpdatesKernel, rbm.getMainConnections().getConnectionGraph().getRows());

  // update visible bias
  if (rbm.getVisibleBiasConnections() != null) {
      if (visibleBiasUpdatesKernel == null || visibleBiasUpdatesKernel.getMiniBatchSize() != mbs) {
    visibleBiasUpdatesKernel = new CDBiasUpdatesKernel(rbm.getVisibleBiasConnections().getConnectionGraph().getElements(), posPhaseVisible, negPhaseVisible, getLearningRate(), getMomentum());
      }

      Environment.getInstance().getExecutionStrategy().execute(visibleBiasUpdatesKernel, rbm.getVisibleBiasConnections().getConnectionGraph().getElements().length);
  }

  // update hidden bias
  if (rbm.getHiddenBiasConnections() != null) {
      if (hiddenBiasUpdatesKernel == null || hiddenBiasUpdatesKernel.getMiniBatchSize() != mbs) {
    hiddenBiasUpdatesKernel = new CDBiasUpdatesKernel(rbm.getHiddenBiasConnections().getConnectionGraph().getElements(), posPhaseHidden, negPhaseHidden, getLearningRate(), getMomentum());
      }

      Environment.getInstance().getExecutionStrategy().execute(hiddenBiasUpdatesKernel, rbm.getHiddenBiasConnections().getConnectionGraph().getElements().length);
  }
    }
View Full Code Here

Examples of com.github.neuralnetworks.architecture.types.RBM

157
158
159
160
161
162
163
164
165
166
167
  Environment.getInstance().setExecutionMode(EXECUTION_MODE.CPU);
  Environment.getInstance().setUseDataSharedMemory(true);
  Environment.getInstance().setUseWeightsSharedMemory(true);

  // RBM with 4 visible and 3 hidden units
  RBM rbm = NNFactory.rbm(4, 3, true);

  // training and testing input providers
  IrisInputProvider trainInputProvider = new IrisInputProvider(new IrisTargetMultiNeuronOutputConverter(), false);
  trainInputProvider.addInputModifier(new ScalingInputFunction(trainInputProvider));
  IrisInputProvider testInputProvider = new IrisInputProvider(new IrisTargetMultiNeuronOutputConverter(), false);
View Full Code Here

Examples of com.github.neuralnetworks.architecture.types.RBM

71
72
73
74
75
76
77
78
79
80
81
  assertEquals(0, bpt.getOutputError().getTotalNetworkError(), 0.1);
    }

    @Test
    public void testRBM() {
  RBM rbm = NNFactory.rbm(784, 10, false);
  MnistInputProvider trainInputProvider = new MnistInputProvider("train-images.idx3-ubyte", "train-labels.idx1-ubyte", 1, 1, new MnistTargetMultiNeuronOutputConverter());
  trainInputProvider.addInputModifier(new ScalingInputFunction(255));
  MnistInputProvider testInputProvider = new MnistInputProvider("t10k-images.idx3-ubyte", "t10k-labels.idx1-ubyte", 1000, 1, new MnistTargetMultiNeuronOutputConverter());
  testInputProvider.addInputModifier(new ScalingInputFunction(255));
View Full Code Here

Examples of com.github.neuralnetworks.architecture.types.RBM

86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
    @Test
    public void testDBNCalculation() {
  DBN dbn = NNFactory.dbn(new int [] {3, 2, 2}, true);
  dbn.setLayerCalculator(NNFactory.lcWeightedSum(dbn, null));

  RBM firstRBM = dbn.getFirstNeuralNetwork();
  Util.fillArray(firstRBM.getMainConnections().getConnectionGraph().getElements(), 0.2f);
  Util.fillArray(firstRBM.getVisibleBiasConnections().getConnectionGraph().getElements(), 0.1f);
  Util.fillArray(firstRBM.getHiddenBiasConnections().getConnectionGraph().getElements(), 0.3f);

  RBM secondRBM = dbn.getLastNeuralNetwork();
  Util.fillArray(secondRBM.getMainConnections().getConnectionGraph().getElements(), 0.4f);
  Util.fillArray(secondRBM.getVisibleBiasConnections().getConnectionGraph().getElements(), 0.8f);
  Util.fillArray(secondRBM.getHiddenBiasConnections().getConnectionGraph().getElements(), 0.5f);

  Set<Layer> calculatedLayers = new HashSet<>();
  calculatedLayers.add(dbn.getInputLayer());

  ValuesProvider results = new ValuesProvider();
View Full Code Here

Examples of com.github.neuralnetworks.architecture.types.RBM

141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
    @Test
    public void testDNNLayerTrainer() {
  DBN dbn = NNFactory.dbn(new int [] {3, 2, 2}, true);
  dbn.setLayerCalculator(NNFactory.lcSigmoid(dbn, null));

  RBM firstRBM = dbn.getFirstNeuralNetwork();

  Matrix cg1 = firstRBM.getMainConnections().getConnectionGraph();
  cg1.set(0.2f, 0, 0);
  cg1.set(0.4f, 0, 1);
  cg1.set(-0.5f, 0, 2);
  cg1.set(-0.3f, 1, 0);
  cg1.set(0.1f, 1, 1);
  cg1.set(0.2f, 1, 2);

  Matrix cgb1 = firstRBM.getVisibleBiasConnections().getConnectionGraph();
  cgb1.set(0f, 0, 0);
  cgb1.set(0f, 1, 0);
  cgb1.set(0f, 2, 0);

  Matrix cgb2 = firstRBM.getHiddenBiasConnections().getConnectionGraph();
  cgb2.set(-0.4f, 0, 0);
  cgb2.set(0.2f, 1, 0);

  SimpleInputProvider inputProvider = new SimpleInputProvider(new float[][] { { 1, 0, 1 } }, null, 1, 1);

  AparapiCDTrainer firstTrainer = TrainerFactory.cdSigmoidTrainer(firstRBM, null, null, null, null, 1f, 0f, 0f, 0f, 1, true);
  firstTrainer.setLayerCalculator(NNFactory.rbmSigmoidSigmoid(firstRBM));

  RBM secondRBM = dbn.getLastNeuralNetwork();

  AparapiCDTrainer secondTrainer = TrainerFactory.cdSigmoidTrainer(secondRBM, null, null, null, null, 1f, 0f, 0f, 0f, 1, true);
  secondTrainer.setLayerCalculator(NNFactory.rbmSigmoidSigmoid(secondRBM));

  Map<NeuralNetwork, OneStepTrainer<?>> layerTrainers = new HashMap<>();
View Full Code Here

Examples of com.github.neuralnetworks.architecture.types.RBM

198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
    public void testDNNLayerTrainer2() {
  DBN dbn = NNFactory.dbn(new int [] {3, 3, 2}, true);
  dbn.setLayerCalculator(NNFactory.lcSigmoid(dbn, null));

  LayerCalculatorImpl lc = (LayerCalculatorImpl) dbn.getLayerCalculator();
  RBM firstRBM = dbn.getFirstNeuralNetwork();
  lc.addConnectionCalculator(firstRBM.getHiddenLayer(), new AparapiWeightedSumConnectionCalculator());

  Matrix m1 = firstRBM.getMainConnections().getConnectionGraph();
  m1.set(1, 0, 0);
  m1.set(0, 0, 1);
  m1.set(0, 0, 2);
  m1.set(0, 1, 0);
  m1.set(1, 1, 1);
  m1.set(0, 1, 2);
  m1.set(0, 2, 0);
  m1.set(0, 2, 1);
  m1.set(1, 2, 2);

  RBM secondRBM = dbn.getLastNeuralNetwork();
 
  Matrix cg1 = secondRBM.getMainConnections().getConnectionGraph();
  cg1.set(0.2f, 0, 0);
  cg1.set(0.4f, 0, 1);
  cg1.set(-0.5f, 0, 2);
  cg1.set(-0.3f, 1, 0);
  cg1.set(0.1f, 1, 1);
  cg1.set(0.2f, 1, 2);
 
  Matrix cgb1 = secondRBM.getVisibleBiasConnections().getConnectionGraph();
  cgb1.set(0f, 0, 0);
  cgb1.set(0f, 1, 0);
  cgb1.set(0f, 2, 0);
 
  Matrix cgb2 = secondRBM.getHiddenBiasConnections().getConnectionGraph();
  cgb2.set(-0.4f, 0, 0);
  cgb2.set(0.2f, 1, 0);
 
  SimpleInputProvider inputProvider = new SimpleInputProvider(new float[][] { { 1, 0, 1 } }, null, 1, 1);
View Full Code Here

Examples of com.github.neuralnetworks.architecture.types.RBM

31
32
33
34
35
36
37
38
39
40
41
  return input;
    }

    @Override
    protected void learnInput(int batch) {
  RBM nn = getNeuralNetwork();

  getLayerCalculator().gibbsSampling(nn, getGibbsSamplingCount(), batch == 0 ? true : !getIsPersistent());

  // update weights
  updateWeights();
View Full Code Here

Examples of com.github.neuralnetworks.architecture.types.RBM

94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
    public void testDBNCalculation() {
  Environment.getInstance().setUseWeightsSharedMemory(true);
  DBN dbn = NNFactory.dbn(new int [] {3, 2, 2}, true);
  dbn.setLayerCalculator(NNFactory.lcWeightedSum(dbn, null));

  RBM firstRBM = dbn.getFirstNeuralNetwork();

  Tensor t = firstRBM.getMainConnections().getWeights();
  float[] e1 = t.getElements();
  t.forEach(i -> e1[i] = 0.2f);
 
  t = firstRBM.getVisibleBiasConnections().getWeights();
  float[] e2 = t.getElements();
  t.forEach(i -> e2[i] = 0.1f);

  t = firstRBM.getHiddenBiasConnections().getWeights();
  float[] e3 = t.getElements();
  t.forEach(i -> e3[i] = 0.3f);

  RBM secondRBM = dbn.getLastNeuralNetwork();

  t = secondRBM.getMainConnections().getWeights();
  float[] e4 = t.getElements();
  t.forEach(i -> e4[i] = 0.4f);

  t = secondRBM.getVisibleBiasConnections().getWeights();
  float[] e5 = t.getElements();
  t.forEach(i -> e5[i] = 0.8f);

  t = secondRBM.getHiddenBiasConnections().getWeights();
  float[] e6 = t.getElements();
  t.forEach(i -> e6[i] = 0.5f);

  Set<Layer> calculatedLayers = new HashSet<>();
  calculatedLayers.add(dbn.getInputLayer());
View Full Code Here

Examples of com.github.neuralnetworks.architecture.types.RBM

196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
    public void testDNNLayerTrainer() {
  Environment.getInstance().setUseWeightsSharedMemory(true);
  DBN dbn = NNFactory.dbn(new int [] {3, 2, 2}, true);
  dbn.setLayerCalculator(NNFactory.lcSigmoid(dbn, null));

  RBM firstRBM = dbn.getFirstNeuralNetwork();

  Matrix cg1 = firstRBM.getMainConnections().getWeights();
  cg1.set(0.2f, 0, 0);
  cg1.set(0.4f, 0, 1);
  cg1.set(-0.5f, 0, 2);
  cg1.set(-0.3f, 1, 0);
  cg1.set(0.1f, 1, 1);
  cg1.set(0.2f, 1, 2);

  Matrix cgb1 = firstRBM.getVisibleBiasConnections().getWeights();
  cgb1.set(0f, 0, 0);
  cgb1.set(0f, 1, 0);
  cgb1.set(0f, 2, 0);

  Matrix cgb2 = firstRBM.getHiddenBiasConnections().getWeights();
  cgb2.set(-0.4f, 0, 0);
  cgb2.set(0.2f, 1, 0);

  SimpleInputProvider inputProvider = new SimpleInputProvider(new float[][] { { 1, 0, 1 } }, null);

  AparapiCDTrainer firstTrainer = TrainerFactory.cdSigmoidTrainer(firstRBM, null, null, null, null, 1f, 0f, 0f, 0f, 1, 1, 1, true);

  RBM secondRBM = dbn.getLastNeuralNetwork();

  AparapiCDTrainer secondTrainer = TrainerFactory.cdSigmoidTrainer(secondRBM, null, null, null, null, 1f, 0f, 0f, 0f, 1, 1, 1, true);

  Map<NeuralNetwork, OneStepTrainer<?>> layerTrainers = new HashMap<>();
  layerTrainers.put(firstRBM, firstTrainer);
View Full Code Here

Examples of com.github.neuralnetworks.architecture.types.RBM

251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
    public void testDNNLayerTrainer2() {
  Environment.getInstance().setUseWeightsSharedMemory(true);
  DBN dbn = NNFactory.dbn(new int [] {3, 3, 2}, true);
  dbn.setLayerCalculator(NNFactory.lcSigmoid(dbn, null));

  RBM firstRBM = dbn.getFirstNeuralNetwork();

  LayerCalculatorImpl lc = (LayerCalculatorImpl) dbn.getLayerCalculator();
  lc.addConnectionCalculator(firstRBM.getHiddenLayer(), new AparapiWeightedSumConnectionCalculator());

  Matrix m1 = firstRBM.getMainConnections().getWeights();
  m1.set(1, 0, 0);
  m1.set(0, 0, 1);
  m1.set(0, 0, 2);
  m1.set(0, 1, 0);
  m1.set(1, 1, 1);
  m1.set(0, 1, 2);
  m1.set(0, 2, 0);
  m1.set(0, 2, 1);
  m1.set(1, 2, 2);

  RBM secondRBM = dbn.getLastNeuralNetwork();
 
  Matrix cg1 = secondRBM.getMainConnections().getWeights();
  cg1.set(0.2f, 0, 0);
  cg1.set(0.4f, 0, 1);
  cg1.set(-0.5f, 0, 2);
  cg1.set(-0.3f, 1, 0);
  cg1.set(0.1f, 1, 1);
  cg1.set(0.2f, 1, 2);
 
  Matrix cgb1 = secondRBM.getVisibleBiasConnections().getWeights();
  cgb1.set(0f, 0, 0);
  cgb1.set(0f, 1, 0);
  cgb1.set(0f, 2, 0);
 
  Matrix cgb2 = secondRBM.getHiddenBiasConnections().getWeights();
  cgb2.set(-0.4f, 0, 0);
  cgb2.set(0.2f, 1, 0);

  SimpleInputProvider inputProvider = new SimpleInputProvider(new float[][] { { 1, 0, 1 } }, null);
View Full Code Here
TOP
Copyright © 2018 www.massapi.com. All rights reserved.
All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.