Package com.github.neuralnetworks.tensor

Examples of com.github.neuralnetworks.tensor.Tensor

89
90
91
92
93
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
  int size = Arrays.stream(dimensions).reduce(1, (a, b) -> a * b);
  float[] elements = null;
  int offset = 0;

  if (useSharedMemory) {
      Tensor sibling = set.stream().filter(t -> t.getSize() == size).findFirst().orElse(null);
      if (sibling != null) {
    elements = sibling.getElements();
    offset = sibling.getStartOffset();
      }
  }

  if (elements == null && useSharedMemory()) {
      float[] oldElements = getElements();
      if (oldElements == null) {
    oldElements = new float[0];
      }

      offset = oldElements.length;
      elements = Arrays.copyOf(oldElements, offset + size);
      for (Tensor t : tensors)
    t.setElements(elements);
  }

  Tensor newTensor = null;

      if (elements != null) {
          tensors.add(newTensor = TensorFactory.tensor(elements, offset, dimensions));
      } else {
          tensors.add(newTensor = TensorFactory.tensor(dimensions));
View Full Code Here
36
37
38
39
40
41
42
43
44
45
46
47
48
    chunk.add(c.connection);

    if (i == connections.size() - 1 || connections.get(i + 1).target != c.target) {
        ConnectionCalculator cc = getConnectionCalculator(c.target);
        if (cc != null) {
      Tensor t = TensorFactory.tensor(c.target, chunk, valuesProvider);
      float[] elements = t.getElements();
      IntStream.range(t.getStartIndex(), t.getStartIndex() + t.getSize()).forEach(j -> elements[j] = 0);
      cc.calculate(chunk, valuesProvider, c.target);
        }

        chunk.clear();
View Full Code Here
48
49
50
51
52
53
54
55
56
57
58
  connections.add(result);
  return result;
    }

    public Conv2DConnection conv2d(Layer inputLayer, Layer outputLayer, int inputFeatureMapRows, int inputFeatureMapColumns, int inputFilters, int kernelRows, int kernelColumns, int outputFilters, int stride) {
  Tensor weights = null;
  if (useSharedWeights()) {
      int l = sharedWeights.length;
      sharedWeights = Arrays.copyOf(sharedWeights, l + outputFilters * inputFilters * kernelRows * kernelColumns);
      updateSharedWeights();
      weights = TensorFactory.tensor(sharedWeights, l, outputFilters, inputFilters, kernelRows, kernelColumns);
View Full Code Here
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
129
130
  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());

  ValuesProvider results = TensorFactory.tensorProvider(dbn, 1, true);
View Full Code Here
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
176
177
178
179
180
181
182
183
184
  Environment.getInstance().setUseWeightsSharedMemory(true);
  StackedAutoencoder sae = NNFactory.sae(new int [] {3, 2, 2}, true);
  sae.setLayerCalculator(NNFactory.lcWeightedSum(sae, null));

  Autoencoder firstAE = sae.getFirstNeuralNetwork();
  Tensor t = ((FullyConnected) firstAE.getConnection(firstAE.getInputLayer(), firstAE.getHiddenLayer())).getWeights();
  float[] e1 = t.getElements();
  t.forEach(i -> e1[i] = 0.2f);

  t = ((FullyConnected) firstAE.getConnection(firstAE.getHiddenBiasLayer(), firstAE.getHiddenLayer())).getWeights();
  float[] e2 = t.getElements();
  t.forEach(i -> e2[i] = 0.3f);

  t = ((FullyConnected) firstAE.getConnection(firstAE.getHiddenLayer(), firstAE.getOutputLayer())).getWeights();
  float[] e3 = t.getElements();
  t.forEach(i -> e3[i] = 0.8f);

  t = ((FullyConnected) firstAE.getConnection(firstAE.getOutputBiasLayer(), firstAE.getOutputLayer())).getWeights();
  float[] e4 = t.getElements();
  t.forEach(i -> e4[i] = 0.9f);

  Autoencoder secondAE = sae.getLastNeuralNetwork();

  t = ((FullyConnected) secondAE.getConnection(secondAE.getInputLayer(), secondAE.getHiddenLayer())).getWeights();
  float[] e5 = t.getElements();
  t.forEach(i -> e5[i] = 0.4f);
 
  t = ((FullyConnected) secondAE.getConnection(secondAE.getHiddenBiasLayer(), secondAE.getHiddenLayer())).getWeights();
  float[] e6 = t.getElements();
  t.forEach(i -> e6[i] = 0.5f);

  t = ((FullyConnected) secondAE.getConnection(secondAE.getHiddenLayer(), secondAE.getOutputLayer())).getWeights();
  float[] e7 = t.getElements();
  t.forEach(i -> e7[i] = 0.7f);

  t = ((FullyConnected) secondAE.getConnection(secondAE.getOutputBiasLayer(), secondAE.getOutputLayer())).getWeights();
  float[] e8 = t.getElements();
  t.forEach(i -> e8[i] = 0.9f);

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

  ValuesProvider results = TensorFactory.tensorProvider(sae, 1, true);
View Full Code Here
28
29
30
31
32
33
34
35
  this.value = value;
    }

    @Override
    public void calculate(List<Connections> connections, ValuesProvider valuesProvider, Layer targetLayer) {
  Tensor t = TensorFactory.tensor(targetLayer, connections, valuesProvider);
  t.forEach(i -> t.getElements()[i] = value);
    }
View Full Code Here
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
  if (c.getWeights().getSize() != weightUpdates.getSize()) {
      throw new IllegalArgumentException("weights and weightUpdates must have the same size");
  }

  Tensor t = TensorFactory.tensor(targetLayer, c, activations);
  this.ffActivation = t.getElements();
  this.activationStartIndex = t.getStartIndex();
  this.activationFeatureMapRowsDistance = t.getDimensionElementsDistance(1);
  this.activationFeatureMapColumnsDistance = t.getDimensionElementsDistance(2);

  this.weightUpdatesTensor = weightUpdates;
  this.weightUpdates = weightUpdates.getElements();
  this.weightUpdatesMomentum = new float[weightUpdates.getSize()];
    }
View Full Code Here
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
    protected final int stride;

    public AparapiConv2D(Conv2DConnection c, ValuesProvider valuesProvider, Layer targetLayer) {
  super();

  Tensor input = null, output = null;
  if (targetLayer == c.getOutputLayer()) {
      input = TensorFactory.tensor(Util.getOppositeLayer(c, targetLayer), c, valuesProvider);
      output = TensorFactory.tensor(targetLayer, c, valuesProvider);
  } else {
      input = TensorFactory.tensor(targetLayer, c, valuesProvider);
      output = TensorFactory.tensor(Util.getOppositeLayer(c, targetLayer), c, valuesProvider);
  }

  this.input = input.getElements();
  this.inputStartIndex = input.getStartIndex();
  this.inputFeatureMapRowsDistance = input.getDimensionElementsDistance(1);
  this.inputFeatureMapColumnsDistance = input.getDimensionElementsDistance(2);

  this.output = output.getElements();
  this.outputStartIndex =  output.getStartIndex();
  this.outputFeatureMapsDistance =  output.getDimensionElementsDistance(0);
  this.outputFeatureMapRowsDistance = output.getDimensionElementsDistance(1);
  this.outputFeatureMapColumnsDistance = output.getDimensionElementsDistance(2);
  this.outputMiniBatchDistance = output.getDimensionElementsDistance(3);

  this.weights = c.getWeights().getElements();
  this.weightsStartIndex = c.getWeights().getStartIndex();

  this.miniBatchSize = TensorFactory.batchSize(valuesProvider);
  this.outputColumns = c.getOutputFeatureMapColumns();
  this.outputFeatureMapLength = c.getOutputFeatureMapLength();
  this.stride = c.getStride();
  this.featureMapWeights = c.getKernelColumns() * c.getKernelRows() * c.getInputFilters();
  this.featureMapOffsets = new int[featureMapWeights * miniBatchSize];

  int inputMiniBatchDistance = input.getDimensionElementsDistance(3);
  int inputFeatureMapsDistance =  input.getDimensionElementsDistance(0);

  for (int m = 0, offset = 0; m < miniBatchSize; m++) {
      for (int i = 0; i < c.getInputFilters(); i++) {
    for (int j = 0; j < c.getKernelRows(); j++) {
        for (int k = 0; k < c.getKernelColumns(); k++) {
View Full Code Here
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
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
    protected final int outputFeatureMapLength;
    protected final int outputFeatureMapColumns;
    protected final int outputMiniBatchDistance;

    public AparapiSubsampling2D(Subsampling2DConnection c, ValuesProvider valuesProvider, Layer targetLayer) {
  Tensor input = null, output = null;
  if (targetLayer == c.getOutputLayer()) {
      input = TensorFactory.tensor(Util.getOppositeLayer(c, targetLayer), c, valuesProvider);
      output = TensorFactory.tensor(targetLayer, c, valuesProvider);
  } else {
      input = TensorFactory.tensor(targetLayer, c, valuesProvider);
      output = TensorFactory.tensor(Util.getOppositeLayer(c, targetLayer), c,  valuesProvider);
  }

  this.input = input.getElements();
  this.inputStartIndex = input.getStartIndex();
  this.inputFeatureMapsDistance =  input.getDimensionElementsDistance(0);
  this.inputFeatureMapRowsDistance = input.getDimensionElementsDistance(1);
  this.inputFeatureMapColumnsDistance = input.getDimensionElementsDistance(2);

  this.output = output.getElements();
  this.outputStartIndex =  output.getStartIndex();
  this.outputFeatureMapsDistance =  output.getDimensionElementsDistance(0);
  this.outputFeatureMapRowsDistance = output.getDimensionElementsDistance(1);
  this.outputFeatureMapColumnsDistance = output.getDimensionElementsDistance(2);
  this.outputMiniBatchDistance = output.getDimensionElementsDistance(3);
  this.outputFeatureMapLength = c.getOutputFeatureMapLength();
  this.outputFeatureMapColumns = c.getOutputFeatureMapColumns();

  this.miniBatchSize = input.getDimensions()[3];

  this.subsamplingRows = c.getSubsamplingRegionRows();
  this.subsamplingCols = c.getSubsamplingRegionCols();
  this.regionLength = c.getSubsamplingRegionLength();
  this.ioRowsOffset = (c.getInputFeatureMapRows() % subsamplingRows) / 2;
  this.ioColumnsOffset = (c.getInputFeatureMapColumns() % subsamplingCols) / 2;

  this.featureMapOffsets = new int[regionLength * miniBatchSize];
  int inputMiniBatchDistance = input.getDimensionElementsDistance(3);
  for (int m = 0, i = 0; m < miniBatchSize; m++) {
      for (int j = 0; j < subsamplingRows; j++) {
    for (int k = 0; k < subsamplingCols; k++) {
        featureMapOffsets[i++] = j * inputFeatureMapRowsDistance + k * inputFeatureMapColumnsDistance + m * inputMiniBatchDistance;
    }
 
View Full Code Here
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
  return new AparapiConv2DFF(c, valuesProvider, targetLayer);
    }

    protected void calculateBias(Conv2DConnection bias, ValuesProvider vp) {
  if (bias != null) {
      Tensor biasValue = TensorFactory.tensor(bias.getInputLayer(), bias, vp);
      if (biasValue.getElements()[biasValue.getStartIndex()] == 0) {
    biasValue.forEach(i -> biasValue.getElements()[i] = 1);
      }

      Tensor v = TensorFactory.tensor(bias.getOutputLayer(), bias, vp);
      Tensor w = bias.getWeights();
      TensorIterator it = v.iterator();

      while (it.hasNext()) {
    v.getElements()[it.next()] = w.get(it.getCurrentPosition()[0], 0, 0, 0);
      }
  }
    }
View Full Code Here
TOP

Related Classes of com.github.neuralnetworks.tensor.Tensor

  • com.github.neuralnetworks.architecture.ConnectionFactory
  • com.github.neuralnetworks.calculation.LayerCalculatorBase
  • com.github.neuralnetworks.calculation.memory.ValuesProvider
  • com.github.neuralnetworks.calculation.neuronfunctions.AparapiConv2D
  • com.github.neuralnetworks.calculation.neuronfunctions.AparapiFullyConnected
  • com.github.neuralnetworks.calculation.neuronfunctions.AparapiSubsampling2D
  • com.github.neuralnetworks.calculation.neuronfunctions.ConnectionCalculatorConv
  • com.github.neuralnetworks.calculation.neuronfunctions.ConnectionCalculatorFullyConnected
  • com.github.neuralnetworks.calculation.neuronfunctions.ConstantConnectionCalculator
  • com.github.neuralnetworks.test.CNNTest
    • 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.