Examples with TensorList com.simiacryptus.mindseye.lang.TensorList used on opensource projects

Example 71 with TensorList

use of com.simiacryptus.mindseye.lang.TensorList in project MindsEye by SimiaCryptus.

the class SumInputsLayer method evalAndFree.

@Nullable
@Override
public Result evalAndFree(@Nonnull final Result... inObj) {
    @Nonnull final int[] dimensions = inObj[0].getData().getDimensions();
    if (3 != dimensions.length) {
        throw new IllegalArgumentException("dimensions=" + Arrays.toString(dimensions));
    }
    for (int i = 1; i < inObj.length; i++) {
        if (Tensor.length(dimensions) != Tensor.length(inObj[i].getData().getDimensions())) {
            throw new IllegalArgumentException(Arrays.toString(dimensions) + " != " + Arrays.toString(inObj[i].getData().getDimensions()));
        }
    }
    if (!CudaSystem.isEnabled())
        return getCompatibilityLayer().evalAndFree(inObj);
    Stream<TensorList> tensorListStream = Arrays.stream(inObj).map(x -> x.getData());
    if (!CoreSettings.INSTANCE.isSingleThreaded() && parallel)
        tensorListStream = tensorListStream.parallel();
    return new Result(tensorListStream.reduce((leftData, rightData) -> CudaSystem.run(gpu -> {
        return gpu.addAndFree(precision, leftData, rightData);
    }, leftData, rightData)).get(), (@Nonnull final DeltaSet<Layer> buffer, @Nonnull final TensorList delta) -> {
        @Nonnull Stream<Result> deltaStream = Arrays.stream(inObj);
        if (!CoreSettings.INSTANCE.isSingleThreaded() && parallel)
            deltaStream = deltaStream.parallel();
        deltaStream.filter(Result::isAlive).forEach(obj -> {
            delta.addRef();
            obj.accumulate(buffer, delta);
        });
    }) {

        @Override
        protected void _free() {
            Arrays.stream(inObj).forEach(x -> x.freeRef());
        }

        @Override
        public boolean isAlive() {
            for (@Nonnull final Result element : inObj) if (element.isAlive()) {
                return true;
            }
            return false;
        }
    };
}

Example 72 with TensorList

use of com.simiacryptus.mindseye.lang.TensorList in project MindsEye by SimiaCryptus.

the class SumReducerLayer method evalAndFree.

@Nullable
@Override
public Result evalAndFree(final Result... inObj) {
    if (!CudaSystem.isEnabled())
        return getCompatibilityLayer().evalAndFree(inObj);
    final Result input = inObj[0];
    final TensorList inputData = input.getData();
    @Nonnull final int[] inputSize = inputData.getDimensions();
    int length = inputData.length();
    CudaTensorList result = CudaSystem.run(gpu -> {
        CudaTensor inputTensor = gpu.getTensor(inputData, precision, MemoryType.Device, false);
        inputData.freeRef();
        CudaMemory inputMemory = inputTensor.getMemory(gpu);
        @Nonnull final CudaDevice.CudaTensorDescriptor outputDescriptor = gpu.newTensorDescriptor(precision, length, 1, 1, 1);
        long size = (long) precision.size * outputDescriptor.nStride * length;
        @Nonnull final CudaMemory outputMemory = gpu.allocate(size, MemoryType.Managed, true);
        CudaResource<cudnnReduceTensorDescriptor> reduceTensorDescriptor = gpu.cudnnCreateReduceTensorDescriptor(cudnnReduceTensorOp.CUDNN_REDUCE_TENSOR_ADD, precision.code, cudnnNanPropagation.CUDNN_NOT_PROPAGATE_NAN, cudnnReduceTensorIndices.CUDNN_REDUCE_TENSOR_NO_INDICES, cudnnIndicesType.CUDNN_32BIT_INDICES);
        @Nonnull final CudaMemory workspacePtr = gpu.allocate(inputMemory.size, MemoryType.Device, true);
        @Nonnull final CudaMemory indexPtr = gpu.allocate(12 * length, MemoryType.Device, false);
        // outputPtr.synchronize();
        gpu.cudnnReduceTensor(reduceTensorDescriptor.getPtr(), indexPtr.getPtr(), indexPtr.size, workspacePtr.getPtr(), workspacePtr.size, precision.getPointer(1.0), inputTensor.descriptor.getPtr(), inputMemory.getPtr(), precision.getPointer(0.0), outputDescriptor.getPtr(), outputMemory.getPtr());
        inputMemory.dirty();
        outputMemory.dirty();
        workspacePtr.dirty();
        Stream.of(inputTensor, inputMemory, reduceTensorDescriptor, workspacePtr, indexPtr).forEach(ReferenceCounting::freeRef);
        return CudaTensorList.wrap(CudaTensor.wrap(outputMemory, outputDescriptor, precision), length, new int[] { 1, 1, 1 }, precision);
    });
    return new Result(result, (DeltaSet<Layer> ctx, TensorList delta) -> {
        // Not supported by CuDNN?
        // CudaTensorList passback = CudaSystem.run(gpu -> {
        // CudaTensor deltaTensor = gpu.getTensor(delta, precision, MemoryType.Device, false);
        // CudaMemory deltaMemory = deltaTensor.getMemory(gpu);
        // 
        // @Nonnull final CudaDevice.CudaTensorDescriptor passbackDescriptor1 = gpu.newTensorDescriptor(
        // precision, length, inputSize[2], inputSize[1], inputSize[0]
        // );
        // @Nonnull final CudaMemory passbackPtr1 = gpu.allocate((long) precision.size * passbackDescriptor1.nStride * length, MemoryType.Device, false);
        // gpu.cudnnAddTensor(precision.getPointer(1.0), deltaTensor.descriptor.getPtr(), deltaMemory.getPtr(),
        // precision.getPointer(1.0), passbackDescriptor1.getPtr(), passbackPtr1.getPtr());
        // passbackPtr1.dirty();
        // 
        // Stream.of(deltaTensor, deltaMemory, passbackDescriptor1, passbackPtr1).forEach(ReferenceCounting::freeRef);
        // return CudaTensorList.wrap(CudaTensor.wrap(passbackPtr1, passbackDescriptor1, precision), length, inputSize, precision);
        // });
        TensorList passback = TensorArray.wrap(IntStream.range(0, length).mapToObj(i -> {
            Tensor tensor = delta.get(i);
            Tensor tensor1 = new Tensor(inputSize).setAll(tensor.get(0));
            tensor.freeRef();
            return tensor1;
        }).toArray(i -> new Tensor[i]));
        input.accumulate(ctx, passback);
    }) {

        @Override
        protected void _free() {
            super._free();
            input.freeRef();
        }
    };
}

Example 73 with TensorList

use of com.simiacryptus.mindseye.lang.TensorList in project MindsEye by SimiaCryptus.

the class AvgPoolingLayer method eval.

@Nonnull
@SuppressWarnings("unchecked")
@Override
public Result eval(@Nonnull final Result... inObj) {
    final int kernelSize = Tensor.length(kernelDims);
    final TensorList data = inObj[0].getData();
    @Nonnull final int[] inputDims = data.getDimensions();
    final int[] newDims = IntStream.range(0, inputDims.length).map(i -> {
        assert 0 == inputDims[i] % kernelDims[i] : inputDims[i] + ":" + kernelDims[i];
        return inputDims[i] / kernelDims[i];
    }).toArray();
    final Map<Coordinate, List<int[]>> coordMap = AvgPoolingLayer.getCoordMap(kernelDims, newDims);
    final Tensor[] outputValues = IntStream.range(0, data.length()).mapToObj(dataIndex -> {
        @Nullable final Tensor input = data.get(dataIndex);
        @Nonnull final Tensor output = new Tensor(newDims);
        for (@Nonnull final Entry<Coordinate, List<int[]>> entry : coordMap.entrySet()) {
            double sum = entry.getValue().stream().mapToDouble(inputCoord -> input.get(inputCoord)).sum();
            if (Double.isFinite(sum)) {
                output.add(entry.getKey(), sum / kernelSize);
            }
        }
        input.freeRef();
        return output;
    }).toArray(i -> new Tensor[i]);
    Arrays.stream(inObj).forEach(nnResult -> nnResult.addRef());
    return new Result(TensorArray.wrap(outputValues), (@Nonnull final DeltaSet<Layer> buffer, @Nonnull final TensorList delta) -> {
        if (inObj[0].isAlive()) {
            final Tensor[] passback = IntStream.range(0, delta.length()).mapToObj(dataIndex -> {
                @Nullable Tensor tensor = delta.get(dataIndex);
                @Nonnull final Tensor backSignal = new Tensor(inputDims);
                for (@Nonnull final Entry<Coordinate, List<int[]>> outputMapping : coordMap.entrySet()) {
                    final double outputValue = tensor.get(outputMapping.getKey());
                    for (@Nonnull final int[] inputCoord : outputMapping.getValue()) {
                        backSignal.add(inputCoord, outputValue / kernelSize);
                    }
                }
                tensor.freeRef();
                return backSignal;
            }).toArray(i -> new Tensor[i]);
            @Nonnull TensorArray tensorArray = TensorArray.wrap(passback);
            inObj[0].accumulate(buffer, tensorArray);
        }
    }) {

        @Override
        protected void _free() {
            Arrays.stream(inObj).forEach(nnResult -> nnResult.freeRef());
        }

        @Override
        public boolean isAlive() {
            return inObj[0].isAlive();
        }
    };
}

Example 74 with TensorList

use of com.simiacryptus.mindseye.lang.TensorList in project MindsEye by SimiaCryptus.

the class FullyConnectedReferenceLayer method eval.

@Nonnull
@Override
public Result eval(final Result... inObj) {
    final Result inputResult = inObj[0];
    final TensorList indata = inputResult.getData();
    inputResult.addRef();
    indata.addRef();
    @Nonnull int[] inputDimensions = indata.getDimensions();
    assert Tensor.length(inputDimensions) == Tensor.length(this.inputDims) : Arrays.toString(inputDimensions) + " == " + Arrays.toString(this.inputDims);
    weights.addRef();
    return new Result(TensorArray.wrap(IntStream.range(0, indata.length()).mapToObj(index -> {
        @Nullable final Tensor input = indata.get(index);
        @Nullable final Tensor output = new Tensor(outputDims);
        weights.coordStream(false).forEach(c -> {
            int[] coords = c.getCoords();
            double prev = output.get(coords[1]);
            double w = weights.get(c);
            double i = input.get(coords[0]);
            double value = prev + w * i;
            output.set(coords[1], value);
        });
        input.freeRef();
        return output;
    }).toArray(i -> new Tensor[i])), (@Nonnull final DeltaSet<Layer> buffer, @Nonnull final TensorList delta) -> {
        if (!isFrozen()) {
            final Delta<Layer> deltaBuffer = buffer.get(FullyConnectedReferenceLayer.this, getWeights().getData());
            Tensor[] array = IntStream.range(0, indata.length()).mapToObj(i -> {
                @Nullable final Tensor inputTensor = indata.get(i);
                @Nullable final Tensor deltaTensor = delta.get(i);
                @Nonnull Tensor weights = new Tensor(FullyConnectedReferenceLayer.this.weights.getDimensions());
                weights.coordStream(false).forEach(c -> {
                    int[] coords = c.getCoords();
                    weights.set(c, inputTensor.get(coords[0]) * deltaTensor.get(coords[1]));
                });
                inputTensor.freeRef();
                deltaTensor.freeRef();
                return weights;
            }).toArray(i -> new Tensor[i]);
            Tensor tensor = Arrays.stream(array).reduce((a, b) -> {
                Tensor c = a.addAndFree(b);
                b.freeRef();
                return c;
            }).get();
            deltaBuffer.addInPlace(tensor.getData()).freeRef();
            tensor.freeRef();
        }
        if (inputResult.isAlive()) {
            @Nonnull final TensorList tensorList = TensorArray.wrap(IntStream.range(0, indata.length()).mapToObj(i -> {
                @Nullable final Tensor inputTensor = new Tensor(inputDims);
                @Nullable final Tensor deltaTensor = delta.get(i);
                weights.coordStream(false).forEach(c -> {
                    int[] coords = c.getCoords();
                    inputTensor.set(coords[0], inputTensor.get(coords[0]) + weights.get(c) * deltaTensor.get(coords[1]));
                });
                deltaTensor.freeRef();
                return inputTensor;
            }).toArray(i -> new Tensor[i]));
            inputResult.accumulate(buffer, tensorList);
        }
    }) {

        @Override
        protected void _free() {
            indata.freeRef();
            inputResult.freeRef();
            weights.freeRef();
        }

        @Override
        public boolean isAlive() {
            return inputResult.isAlive() || !isFrozen();
        }
    };
}

Example 75 with TensorList

use of com.simiacryptus.mindseye.lang.TensorList in project MindsEye by SimiaCryptus.

the class GaussianNoiseLayer method eval.

@Nonnull
@Override
public Result eval(final Result... inObj) {
    final Result in0 = inObj[0];
    final TensorList inputData = in0.getData();
    final int itemCnt = inputData.length();
    in0.addRef();
    inputData.addRef();
    final Tensor[] outputA = IntStream.range(0, itemCnt).mapToObj(dataIndex -> {
        @Nonnull final Random random = new Random(seed);
        @Nullable final Tensor input = inputData.get(dataIndex);
        @Nullable final Tensor output = input.map(x -> {
            return x + random.nextGaussian() * getValue();
        });
        input.freeRef();
        return output;
    }).toArray(i -> new Tensor[i]);
    return new Result(TensorArray.wrap(outputA), (@Nonnull final DeltaSet<Layer> buffer, @Nonnull final TensorList delta) -> {
        if (in0.isAlive()) {
            @Nonnull TensorArray tensorArray = TensorArray.wrap(IntStream.range(0, delta.length()).mapToObj(dataIndex -> {
                Tensor tensor = delta.get(dataIndex);
                @Nullable final double[] deltaData = tensor.getData();
                @Nonnull final Tensor passback = new Tensor(inputData.getDimensions());
                for (int i = 0; i < passback.length(); i++) {
                    passback.set(i, deltaData[i]);
                }
                tensor.freeRef();
                return passback;
            }).toArray(i -> new Tensor[i]));
            in0.accumulate(buffer, tensorArray);
        }
    }) {

        @Override
        protected void _free() {
            inputData.freeRef();
            in0.freeRef();
        }

        @Override
        public boolean isAlive() {
            return in0.isAlive() || !isFrozen();
        }
    };
}