1. Test Modules
2. Differential Validation
   1. Feedback Validation
   2. Learning Validation
   3. Total Accuracy
   4. Frozen and Alive Status
3. Results

Subreport: Logs for com.simiacryptus.ref.lang.ReferenceCountingBase

Test Modules

Using Seed 7850499601193346048

Differential Validation

SingleDerivativeTester.java:101 executed in 0.00 seconds (0.000 gc):

        log.info(RefString.format("Inputs: %s", prettyPrint(inputPrototype)));
        log.info(RefString.format("Inputs Statistics: %s", printStats(inputPrototype)));
        log.info(RefString.format("Output: %s", outputPrototype.prettyPrint()));
        assert outputPrototype != null;
        log.info(RefString.format("Outputs Statistics: %s", outputPrototype.getScalarStatistics()));
      },
      outputPrototype.addRef(),
      RefUtil.addRef(inputPrototype)));

Inputs: [
[ [ 0.08, -0.804 ], [ 0.496, -0.712 ], [ 0.048, 1.356 ], [ -1.72, 1.032 ], [ -0.852, -0.176 ] ],
[ [ 0.7, 1.108 ], [ -0.608, 1.048 ], [ 1.524, 1.512 ], [ -1.028, 0.3 ], [ 1.912, 1.556 ] ],
[ [ -0.128, 0.028 ], [ 1.764, -1.616 ], [ 1.208, 1.556 ], [ -0.384, 0.636 ], [ -1.688, 0.788 ] ]
]
Inputs Statistics: {meanExponent=-0.1863636075006468, negative=11, min=-1.72, max=1.912, mean=0.29786666666666667, count=30, sum=8.936, positive=19, stdDev=1.064993888349704, zeros=0}
Output: [
[ [ 0.08826666666666666, 0.5074666666666666 ] ]
]
Outputs Statistics: {meanExponent=-0.6743978766963493, negative=0, min=0.08826666666666666, max=0.5074666666666666, mean=0.2978666666666666, count=2, sum=0.5957333333333332, positive=2, stdDev=0.2096, zeros=0}

Feedback Validation

We validate the agreement between the implemented derivative of the inputs apply finite difference estimations:

SingleDerivativeTester.java:117 executed in 0.12 seconds (0.000 gc):

        return testFeedback(
            statistics,
            component.addRef(),
            RefUtil.addRef(inputPrototype),
            outputPrototype.addRef());
      },
      outputPrototype.addRef(),
      RefUtil.addRef(inputPrototype),
      component.addRef()));

Feedback for input 0
Inputs Values: [
[ [ 0.08, -0.804 ], [ 0.496, -0.712 ], [ 0.048, 1.356 ], [ -1.72, 1.032 ], [ -0.852, -0.176 ] ],
[ [ 0.7, 1.108 ], [ -0.608, 1.048 ], [ 1.524, 1.512 ], [ -1.028, 0.3 ], [ 1.912, 1.556 ] ],
[ [ -0.128, 0.028 ], [ 1.764, -1.616 ], [ 1.208, 1.556 ], [ -0.384, 0.636 ], [ -1.688, 0.788 ] ]
]
Value Statistics: {meanExponent=-0.1863636075006468, negative=11, min=-1.72, max=1.912, mean=0.29786666666666667, count=30, sum=8.936, positive=19, stdDev=1.064993888349704, zeros=0}
Implemented Feedback: [ [ 0.06666666666666667, 0.0 ], [ 0.06666666666666667, 0.0 ], [ 0.06666666666666667, 0.0 ], [ 0.06666666666666667, 0.0 ], [ 0.06666666666666667, 0.0 ], [ 0.06666666666666667, 0.0 ], [ 0.06666666666666667, 0.0 ], [ 0.06666666666666667, 0.0 ], ... ]
Implemented Statistics: {meanExponent=-1.1760912590556813, negative=0, min=0.0, max=0.06666666666666667, mean=0.03333333333333333, count=60, sum=1.9999999999999998, positive=30, stdDev=0.03333333333333333, zeros=30}
Measured Feedback: [ [ 0.06666666666640952, 0.0 ], [ 0.06666666666640952, 0.0 ], [ 0.06666666666640952, 0.0 ], [ 0.06666666666682586, 0.0 ], [ 0.06666666666682586, 0.0 ], [ 0.06666666666682586, 0.0 ], [ 0.06666666666640952, 0.0 ], [ 0.06666666666640952, 0.0 ], ... ]
Measured Statistics: {meanExponent=-1.176091259056482, negative=0, min=0.0, max=0.06666666666710341, mean=0.03333333333327184, count=60, sum=1.9999999999963103, positive=30, stdDev=0.03333333333327183, zeros=30}
Feedback Error: [ [ -2.5714153029099407E-13, 0.0 ], [ -2.5714153029099407E-13, 0.0 ], [ -2.5714153029099407E-13, 0.0 ], [ 1.5919210394343963E-13, 0.0 ], [ 1.5919210394343963E-13, 0.0 ], [ 1.5919210394343963E-13, 0.0 ], [ -2.5714153029099407E-13, 0.0 ], [ -2.5714153029099407E-13, 0.0 ], ... ]
Error Statistics: {meanExponent=-12.46629108888206, negative=17, min=-6.734751645254278E-13, max=4.3674786009972877E-13, mean=-6.14947907410605E-14, count=60, sum=-3.68968744446363E-12, positive=13, stdDev=3.0232552347941765E-13, zeros=30}

Returns

    {
      "absoluteTol" : {
        "count" : 60,
        "sum" : 1.1714462733181108E-11,
        "min" : 0.0,
        "max" : 6.734751645254278E-13,
        "sumOfSquare" : 5.710939886123443E-24,
        "standardDeviation" : 2.3887918613130545E-13,
        "average" : 1.9524104555301847E-13
      },
      "relativeTol" : {
        "count" : 30,
        "sum" : 8.785847049901224E-11,
        "min" : 1.193940779574372E-12,
        "max" : 5.0510637339662224E-12,
        "sumOfSquare" : 3.2124036859612233E-22,
        "standardDeviation" : 1.459870701830917E-12,
        "average" : 2.928615683300408E-12
      }
    }

Learning Validation

We validate the agreement between the implemented derivative of the internal weights apply finite difference estimations:

SingleDerivativeTester.java:133 executed in 0.00 seconds (0.000 gc):

        return testLearning(
            statistics,
            component.addRef(),
            RefUtil.addRef(inputPrototype),
            outputPrototype.addRef());
      },
      outputPrototype.addRef(),
      RefUtil.addRef(inputPrototype),
      component.addRef()));

Returns

    {
      "absoluteTol" : {
        "count" : 60,
        "sum" : 1.1714462733181108E-11,
        "min" : 0.0,
        "max" : 6.734751645254278E-13,
        "sumOfSquare" : 5.710939886123443E-24,
        "standardDeviation" : 2.3887918613130545E-13,
        "average" : 1.9524104555301847E-13
      },
      "relativeTol" : {
        "count" : 30,
        "sum" : 8.785847049901224E-11,
        "min" : 1.193940779574372E-12,
        "max" : 5.0510637339662224E-12,
        "sumOfSquare" : 3.2124036859612233E-22,
        "standardDeviation" : 1.459870701830917E-12,
        "average" : 2.928615683300408E-12
      }
    }

Total Accuracy

The overall agreement accuracy between the implemented derivative and the finite difference estimations:

SingleDerivativeTester.java:148 executed in 0.00 seconds (0.000 gc):

    //log.info(String.format("Component: %s\nInputs: %s\noutput=%s", component, Arrays.toStream(inputPrototype), outputPrototype));
    log.info(RefString.format("Finite-Difference Derivative Accuracy:"));
    log.info(RefString.format("absoluteTol: %s", statistics.absoluteTol));
    log.info(RefString.format("relativeTol: %s", statistics.relativeTol));

Finite-Difference Derivative Accuracy:
absoluteTol: 1.9524e-13 +- 2.3888e-13 [0.0000e+00 - 6.7348e-13] (60#)
relativeTol: 2.9286e-12 +- 1.4599e-12 [1.1939e-12 - 5.0511e-12] (30#)

Frozen and Alive Status

SingleDerivativeTester.java:156 executed in 0.01 seconds (0.000 gc):

    testFrozen(component.addRef(), RefUtil.addRef(inputPrototype));
    testUnFrozen(component.addRef(), RefUtil.addRef(inputPrototype));

LayerTests.java:425 executed in 0.00 seconds (0.000 gc):

    throwException(exceptions.addRef());

Results

  {
    "result": "OK",
    "performance": {
      "execution_time": "0.325",
      "gc_time": "0.141"
    },
    "created_on": 1586743677735,
    "file_name": "derivativeTest",
    "report": {
      "simpleName": "Asymmetric",
      "canonicalName": "com.simiacryptus.mindseye.layers.cudnn.BandAvgReducerLayerTest.Asymmetric",
      "link": "https://github.com/SimiaCryptus/mindseye-cudnn/tree/59d5b3318556370acb2d83ee6ec123ce0fc6974f/src/test/java/com/simiacryptus/mindseye/layers/cudnn/BandAvgReducerLayerTest.java",
      "javaDoc": ""
    },
    "archive": "s3://code.simiacrypt.us/tests/com/simiacryptus/mindseye/layers/cudnn/BandAvgReducerLayer/Asymmetric/derivativeTest/202004130757",
    "id": "9d2e8431-335c-4c12-a8d6-0dac26f52e0e",
    "report_type": "Components",
    "display_name": "Derivative Validation",
    "target": {
      "simpleName": "BandAvgReducerLayer",
      "canonicalName": "com.simiacryptus.mindseye.layers.cudnn.BandAvgReducerLayer",
      "link": "https://github.com/SimiaCryptus/mindseye-cudnn/tree/59d5b3318556370acb2d83ee6ec123ce0fc6974f/src/main/java/com/simiacryptus/mindseye/layers/cudnn/BandAvgReducerLayer.java",
      "javaDoc": ""
    }
  }