Package com.github.neuralnetworks.architecture.types

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

33
34
35
36
37
38
39
40
41
42
43
    public void testContrastiveDivergence() {
  Environment.getInstance().setExecutionMode(EXECUTION_MODE.SEQ);

  // RBM with 6 visible, 2 hidden units and a bias
  Environment.getInstance().setUseWeightsSharedMemory(true);
  RBM rbm = NNFactory.rbm(6, 2, true);

  // We'll use a simple dataset of symptoms of a flu illness. There are 6
  // input features and the first three are symptoms of the illness - for
  // example 1 0 0 0 0 0 means that a patient has high temperature, 0 1
  // 0 0 0 0 - coughing, 1 1 0 0 0 0 - coughing and high temperature
View Full Code Here
74
75
76
77
78
79
80
81
82
83
84
    public void testPersistentContrastiveDivergence() {
  Environment.getInstance().setExecutionMode(EXECUTION_MODE.SEQ);

  // RBM with 6 visible, 2 hidden units and bias
  Environment.getInstance().setUseWeightsSharedMemory(true);
  RBM rbm = NNFactory.rbm(6, 2, true);
 
  // We'll use a simple dataset of symptoms of a flu illness. There are 6
  // input features and the first three are symptoms of the illness - for
  // example 1 0 0 0 0 0 means that a patient has high temperature, 0 1
  // 0 0 0 0 - coughing, 1 1 0 0 0 0 - coughing and high temperature
View Full Code Here
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
    }

    @Test
    public void testRBMLayerCalculator1() {
  Environment.getInstance().setUseWeightsSharedMemory(true);
  RBM rbm = NNFactory.rbm(2, 2, false);
  rbm.setLayerCalculator(NNFactory.lcSigmoid(rbm, null));

  Matrix cg1 = rbm.getMainConnections().getWeights();
  cg1.set(0.1f, 0, 0);
  cg1.set(0.8f, 0, 1);
  cg1.set(0.4f, 1, 0);
  cg1.set(0.6f, 1, 1);


  ValuesProvider vp = TensorFactory.tensorProvider(rbm, 1, true);
  Matrix visible = vp.get(rbm.getVisibleLayer());
  visible.set(0.35f, 0, 0);
  visible.set(0.9f, 1, 0);

  Set<Layer> calculated = new HashSet<Layer>();
  calculated.add(rbm.getVisibleLayer());
  rbm.getLayerCalculator().calculate(rbm, rbm.getHiddenLayer(), calculated, vp);

  Matrix hidden = vp.get(rbm.getHiddenLayer());
  assertEquals(0.68, hidden.get(0, 0), 0.01);
  assertEquals(0.6637, hidden.get(1, 0), 0.01);
    }
View Full Code Here
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
    }
   
    @Test
    public void testRBMLayerCalculator2() {
  Environment.getInstance().setUseWeightsSharedMemory(true);
  RBM rbm = NNFactory.rbm(2, 2, false);
  rbm.setLayerCalculator(NNFactory.lcSigmoid(rbm, null));
 
  Matrix cg1 = rbm.getMainConnections().getWeights();
  cg1.set(0.1f, 0, 0);
  cg1.set(0.8f, 1, 0);
  cg1.set(0.4f, 0, 1);
  cg1.set(0.6f, 1, 1);
 
  ValuesProvider vp = TensorFactory.tensorProvider(rbm, 1, true);
  Matrix hidden = vp.get(rbm.getHiddenLayer());
  hidden.set(0.35f, 0, 0);
  hidden.set(0.9f, 1, 0);
 
  Set<Layer> calculated = new HashSet<Layer>();
  calculated.add(rbm.getHiddenLayer());
  rbm.getLayerCalculator().calculate(rbm, rbm.getVisibleLayer(), calculated, vp);
 
  Matrix visible = vp.get(rbm.getVisibleLayer());
  assertEquals(0.68, visible.get(0, 0), 0.01);
  assertEquals(0.6637, visible.get(1, 0), 0.01);
    }
View Full Code Here
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
    }

    @Test
    public void testRBMLayerCalculator3() {
  Environment.getInstance().setUseWeightsSharedMemory(true);
  RBM rbm = NNFactory.rbm(3, 2, true);
  rbm.setLayerCalculator(NNFactory.lcSigmoid(rbm, null));

  Matrix cg1 = rbm.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 = rbm.getHiddenBiasConnections().getWeights();
  cgb1.set(-0.4f, 0, 0);
  cgb1.set(0.2f, 1, 0);

  ValuesProvider vp = TensorFactory.tensorProvider(rbm, 1, true);
  Matrix visible = vp.get(rbm.getVisibleLayer());
  visible.set(1f, 0, 0);
  visible.set(0f, 1, 0);
  visible.set(1f, 2, 0);

  Set<Layer> calculated = new HashSet<Layer>();
  calculated.add(rbm.getVisibleLayer());
  rbm.getLayerCalculator().calculate(rbm, rbm.getHiddenLayer(), calculated, vp);

  Matrix hidden = vp.get(rbm.getHiddenLayer());
  assertEquals(0.332, hidden.get(0, 0), 0.001);
  assertEquals(0.525, hidden.get(1, 0), 0.001);
    }
View Full Code Here
195
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
    }

    @Test
    public void testRBMLayerCalculator4() {
  Environment.getInstance().setUseWeightsSharedMemory(true);
  RBM rbm = NNFactory.rbm(2, 3, true);
  rbm.setLayerCalculator(NNFactory.lcSigmoid(rbm, null));

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

  Matrix cgb1 = rbm.getVisibleBiasConnections().getWeights();
  cgb1.set(-0.4f, 0, 0);
  cgb1.set(0.2f, 1, 0);

  ValuesProvider vp = TensorFactory.tensorProvider(rbm, 1, true);
  Matrix hidden = vp.get(rbm.getHiddenLayer());
  hidden.set(1f, 0, 0);
  hidden.set(0f, 1, 0);
  hidden.set(1f, 2, 0);

  Set<Layer> calculated = new HashSet<Layer>();
  calculated.add(rbm.getHiddenLayer());
  rbm.getLayerCalculator().calculate(rbm, rbm.getVisibleLayer(), calculated, vp);

  Matrix visible = vp.get(rbm.getVisibleLayer());
  assertEquals(0.332, visible.get(0, 0), 0.001);
  assertEquals(0.525, visible.get(1, 0), 0.001);
    }
View Full Code Here
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
    @Test
    public void testOneStepContrastiveDivergence() {
  //Environment.getInstance().setExecutionMode(EXECUTION_MODE.SEQ);

  Environment.getInstance().setUseWeightsSharedMemory(true);
  RBM rbm = NNFactory.rbm(3, 2, true);

  Matrix cg1 = rbm.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 = rbm.getVisibleBiasConnections().getWeights();
  cgb1.set(0f, 0, 0);
  cgb1.set(0f, 1, 0);
  cgb1.set(0f, 2, 0);

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

  AparapiCDTrainer t = TrainerFactory.cdSigmoidTrainer(rbm, new SimpleInputProvider(new float[][] { { 1, 0, 1 } }, null), null, null, null, 1f, 0f, 0f, 0f, 1, 1, 1, true);
View Full Code Here
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
    @Test
    public void testTwoStepContrastiveDivergence() {
  //Environment.getInstance().setExecutionMode(EXECUTION_MODE.SEQ);
  Environment.getInstance().setUseWeightsSharedMemory(true);
  Environment.getInstance().setUseDataSharedMemory(true);
  RBM rbm = NNFactory.rbm(3, 2, true);

  Matrix cg1 = rbm.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 = rbm.getVisibleBiasConnections().getWeights();
  cgb1.set(0f, 0, 0);
  cgb1.set(0f, 1, 0);
  cgb1.set(0f, 2, 0);

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

  AparapiCDTrainer t = TrainerFactory.cdSigmoidTrainer(rbm, new SimpleInputProvider(new float[][] { { 1, 0, 1 }, { 1, 1, 0 } }, null), null, null, null, 1f, 0f, 0f, 0f, 1, 1, 1, false);
View Full Code Here
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
    public void handleEvent(TrainingEvent event) {
  // transfer of learned weights from lower to the higher RBM
  if (event instanceof LayerTrainingFinished) {
      LayerTrainingFinished e = (LayerTrainingFinished) event;
      CDTrainerBase t = (CDTrainerBase) e.currentTrainer;
      RBM current = t.getNeuralNetwork();
      List<? extends NeuralNetwork> list = getNeuralNetwork().getNeuralNetworks();

      if (list.indexOf(current) < list.size() - 1) {
    RBM next = (RBM) list.get(list.indexOf(current) + 1);
    if (current.getMainConnections().getWeights().getSize() == next.getMainConnections().getWeights().getSize()) {
        TensorFactory.copy(current.getMainConnections().getWeights(), next.getMainConnections().getWeights());
    }

    if (current.getVisibleBiasConnections() != null && next.getVisibleBiasConnections() != null && current.getVisibleBiasConnections().getWeights().getSize() == next.getVisibleBiasConnections().getWeights().getSize()) {
        TensorFactory.copy(current.getVisibleBiasConnections().getWeights(), next.getVisibleBiasConnections().getWeights());
    }
   
    if (current.getHiddenBiasConnections() != null && next.getHiddenBiasConnections() != null && current.getHiddenBiasConnections().getWeights().getSize() == next.getHiddenBiasConnections().getWeights().getSize()) {
        TensorFactory.copy(current.getHiddenBiasConnections().getWeights(), next.getHiddenBiasConnections().getWeights());
    }
      }
  }
    }
View Full Code Here
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
69
     * @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() {
  RBM rbm = getNeuralNetwork();

  RBMLayerCalculator lc = getLayerCalculator();
  int mbs = lc.getPositivePhaseVisible().getDimensions()[lc.getPositivePhaseVisible().getDimensions().length - 1];

  if (weightUpdatesKernel == null || weightUpdatesKernel.getMiniBatchSize() != mbs) {
      weightUpdatesKernel = new CDWeightUpdatesKernel(lc.getPositivePhaseVisible(), lc.getPositivePhaseHidden(), lc.getNegativePhaseVisible(), lc.getNegativePhaseHidden(), rbm.getMainConnections().getWeights(), getLearningRate(), getMomentum(), getl1weightDecay(), getl2weightDecay());
  }
  Environment.getInstance().getExecutionStrategy().execute(weightUpdatesKernel, rbm.getMainConnections().getWeights().getRows());

  // update visible bias
  if (rbm.getVisibleBiasConnections() != null) {
      if (visibleBiasUpdatesKernel == null || visibleBiasUpdatesKernel.getMiniBatchSize() != mbs) {
    visibleBiasUpdatesKernel = new CDBiasUpdatesKernel(rbm.getVisibleBiasConnections().getWeights(), lc.getPositivePhaseVisible(), lc.getNegativePhaseVisible(), getLearningRate(), getMomentum());
      }

      Environment.getInstance().getExecutionStrategy().execute(visibleBiasUpdatesKernel, rbm.getVisibleBiasConnections().getWeights().getSize());
  }

  // update hidden bias
  if (rbm.getHiddenBiasConnections() != null) {
      if (hiddenBiasUpdatesKernel == null || hiddenBiasUpdatesKernel.getMiniBatchSize() != mbs) {
    hiddenBiasUpdatesKernel = new CDBiasUpdatesKernel(rbm.getHiddenBiasConnections().getWeights(), lc.getPositivePhaseHidden(), lc.getNegativePhaseHidden(), getLearningRate(), getMomentum());
      }

      Environment.getInstance().getExecutionStrategy().execute(hiddenBiasUpdatesKernel, rbm.getHiddenBiasConnections().getWeights().getSize());
  }
    }
View Full Code Here
TOP

Related Classes of com.github.neuralnetworks.architecture.types.RBM

  • com.github.neuralnetworks.architecture.FullyConnected
  • com.github.neuralnetworks.architecture.Layer
  • com.github.neuralnetworks.samples.test.IrisTest
  • com.github.neuralnetworks.samples.test.MnistTest
  • com.github.neuralnetworks.test.DNNTest
  • com.github.neuralnetworks.test.RBMTest
  • com.github.neuralnetworks.training.rbm.AparapiCDTrainer
  • com.github.neuralnetworks.training.rbm.CDTrainerBase
  • com.github.neuralnetworks.training.rbm.DBNTrainer
    • Copyright © 2018 www.massapicom. 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.