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

Example 21 with TensorArray

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

the class SumMetaLayer method eval.

@Nullable
@Override
public Result eval(@Nonnull final Result... inObj) {
    final Result input = inObj[0];
    Arrays.stream(inObj).forEach(nnResult -> nnResult.addRef());
    final int itemCnt = input.getData().length();
    if (null == lastResult || minBatches < itemCnt) {
        @Nonnull final ToDoubleFunction<Coordinate> f = (c) -> IntStream.range(0, itemCnt).mapToDouble(dataIndex -> input.getData().get(dataIndex).get(c)).sum();
        lastResult = input.getData().get(0).mapCoords(f);
    }
    return new Result(TensorArray.wrap(lastResult), (@Nonnull final DeltaSet<Layer> buffer, @Nonnull final TensorList data) -> {
        if (input.isAlive()) {
            @Nullable final Tensor delta = data.get(0);
            @Nonnull final Tensor[] feedback = new Tensor[itemCnt];
            Arrays.parallelSetAll(feedback, i -> new Tensor(delta.getDimensions()));
            @Nonnull final ToDoubleFunction<Coordinate> f = (inputCoord) -> {
                for (int inputItem = 0; inputItem < itemCnt; inputItem++) {
                    feedback[inputItem].add(inputCoord, delta.get(inputCoord));
                }
                return 0;
            };
            delta.mapCoords(f);
            @Nonnull TensorArray tensorArray = TensorArray.wrap(feedback);
            input.accumulate(buffer, tensorArray);
        }
    }) {

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

        @Override
        public boolean isAlive() {
            return input.isAlive();
        }
    };
}

Example 22 with TensorArray

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

the class ProductInputsLayer method eval.

@Nonnull
@Override
public Result eval(@Nonnull final Result... inObj) {
    assert inObj.length > 1;
    Arrays.stream(inObj).forEach(x -> x.getData().addRef());
    Arrays.stream(inObj).forEach(nnResult -> nnResult.addRef());
    for (int i = 1; i < inObj.length; i++) {
        final int dim0 = Tensor.length(inObj[0].getData().getDimensions());
        final int dimI = Tensor.length(inObj[i].getData().getDimensions());
        if (dim0 != 1 && dimI != 1 && dim0 != dimI) {
            throw new IllegalArgumentException(Arrays.toString(inObj[0].getData().getDimensions()) + " != " + Arrays.toString(inObj[i].getData().getDimensions()));
        }
    }
    return new Result(Arrays.stream(inObj).parallel().map(x -> {
        TensorList data = x.getData();
        data.addRef();
        return data;
    }).reduce((l, r) -> {
        TensorArray productArray = TensorArray.wrap(IntStream.range(0, Math.max(l.length(), r.length())).parallel().mapToObj(i1 -> {
            @Nullable final Tensor left = l.get(1 == l.length() ? 0 : i1);
            @Nullable final Tensor right = r.get(1 == r.length() ? 0 : i1);
            Tensor product = Tensor.product(left, right);
            left.freeRef();
            right.freeRef();
            return product;
        }).toArray(i -> new Tensor[i]));
        l.freeRef();
        r.freeRef();
        return productArray;
    }).get(), (@Nonnull final DeltaSet<Layer> buffer, @Nonnull final TensorList delta) -> {
        for (@Nonnull final Result input : inObj) {
            if (input.isAlive()) {
                @Nonnull TensorList passback = Arrays.stream(inObj).parallel().map(x -> {
                    TensorList tensorList = x == input ? delta : x.getData();
                    tensorList.addRef();
                    return tensorList;
                }).reduce((l, r) -> {
                    TensorArray productList = TensorArray.wrap(IntStream.range(0, Math.max(l.length(), r.length())).parallel().mapToObj(j -> {
                        @Nullable final Tensor left = l.get(1 == l.length() ? 0 : j);
                        @Nullable final Tensor right = r.get(1 == r.length() ? 0 : j);
                        Tensor product = Tensor.product(left, right);
                        left.freeRef();
                        right.freeRef();
                        return product;
                    }).toArray(j -> new Tensor[j]));
                    l.freeRef();
                    r.freeRef();
                    return productList;
                }).get();
                final TensorList inputData = input.getData();
                if (1 == inputData.length() && 1 < passback.length()) {
                    TensorArray newValue = TensorArray.wrap(passback.stream().reduce((a, b) -> {
                        @Nullable Tensor c = a.addAndFree(b);
                        b.freeRef();
                        return c;
                    }).get());
                    passback.freeRef();
                    passback = newValue;
                }
                if (1 == Tensor.length(inputData.getDimensions()) && 1 < Tensor.length(passback.getDimensions())) {
                    TensorArray newValue = TensorArray.wrap(passback.stream().map((a) -> {
                        @Nonnull Tensor b = new Tensor(a.sum());
                        a.freeRef();
                        return b;
                    }).toArray(i -> new Tensor[i]));
                    passback.freeRef();
                    passback = newValue;
                }
                input.accumulate(buffer, passback);
            }
        }
    }) {

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

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

Example 23 with TensorArray

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

the class ReLuActivationLayer method eval.

@Nonnull
@Override
public Result eval(final Result... inObj) {
    final Result input = inObj[0];
    final TensorList indata = input.getData();
    input.addRef();
    indata.addRef();
    weights.addRef();
    final int itemCnt = indata.length();
    return new Result(TensorArray.wrap(IntStream.range(0, itemCnt).parallel().mapToObj(dataIndex -> {
        @Nullable Tensor tensorElement = indata.get(dataIndex);
        @Nonnull final Tensor tensor = tensorElement.multiply(weights.get(0));
        tensorElement.freeRef();
        @Nullable final double[] outputData = tensor.getData();
        for (int i = 0; i < outputData.length; i++) {
            if (outputData[i] < 0) {
                outputData[i] = 0;
            }
        }
        return tensor;
    }).toArray(i -> new Tensor[i])), (@Nonnull final DeltaSet<Layer> buffer, @Nonnull final TensorList delta) -> {
        if (!isFrozen()) {
            IntStream.range(0, delta.length()).parallel().forEach(dataIndex -> {
                @Nullable Tensor deltaTensor = delta.get(dataIndex);
                @Nullable final double[] deltaData = deltaTensor.getData();
                @Nullable Tensor inputTensor = indata.get(dataIndex);
                @Nullable final double[] inputData = inputTensor.getData();
                @Nonnull final Tensor weightDelta = new Tensor(weights.getDimensions());
                @Nullable final double[] weightDeltaData = weightDelta.getData();
                for (int i = 0; i < deltaData.length; i++) {
                    weightDeltaData[0] += inputData[i] < 0 ? 0 : deltaData[i] * inputData[i];
                }
                buffer.get(ReLuActivationLayer.this, weights.getData()).addInPlace(weightDeltaData).freeRef();
                deltaTensor.freeRef();
                inputTensor.freeRef();
                weightDelta.freeRef();
            });
        }
        if (input.isAlive()) {
            final double weight = weights.getData()[0];
            @Nonnull TensorArray tensorArray = TensorArray.wrap(IntStream.range(0, delta.length()).parallel().mapToObj(dataIndex -> {
                @Nullable Tensor deltaTensor = delta.get(dataIndex);
                @Nullable final double[] deltaData = deltaTensor.getData();
                @Nullable Tensor inTensor = indata.get(dataIndex);
                @Nullable final double[] inputData = inTensor.getData();
                @Nonnull final int[] dims = inTensor.getDimensions();
                @Nonnull final Tensor passback = new Tensor(dims);
                for (int i = 0; i < passback.length(); i++) {
                    passback.set(i, inputData[i] < 0 ? 0 : deltaData[i] * weight);
                }
                inTensor.freeRef();
                deltaTensor.freeRef();
                return passback;
            }).toArray(i -> new Tensor[i]));
            input.accumulate(buffer, tensorArray);
        }
    }) {

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

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

Example 24 with TensorArray

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

the class ScaleMetaLayer method eval.

@Nullable
@Override
public Result eval(@Nonnull final Result... inObj) {
    final int itemCnt = inObj[0].getData().length();
    Arrays.stream(inObj).forEach(nnResult -> nnResult.addRef());
    Arrays.stream(inObj).forEach(x -> x.getData().addRef());
    final Tensor[] tensors = IntStream.range(0, itemCnt).mapToObj(dataIndex -> inObj[0].getData().get(dataIndex).mapIndex((v, c) -> v * inObj[1].getData().get(0).get(c))).toArray(i -> new Tensor[i]);
    Tensor tensor0 = tensors[0];
    tensor0.addRef();
    return new Result(TensorArray.wrap(tensors), (@Nonnull final DeltaSet<Layer> buffer, @Nonnull final TensorList data) -> {
        if (inObj[0].isAlive()) {
            @Nonnull TensorArray tensorArray = TensorArray.wrap(data.stream().map(t -> {
                @Nullable Tensor t1 = inObj[1].getData().get(0);
                @Nullable Tensor tensor = t.mapIndex((v, c) -> {
                    return v * t1.get(c);
                });
                t.freeRef();
                t1.freeRef();
                return tensor;
            }).toArray(i -> new Tensor[i]));
            inObj[0].accumulate(buffer, tensorArray);
        }
        if (inObj[1].isAlive()) {
            @Nullable final Tensor passback = tensor0.mapIndex((v, c) -> {
                return IntStream.range(0, itemCnt).mapToDouble(i -> data.get(i).get(c) * inObj[0].getData().get(i).get(c)).sum();
            });
            tensor0.freeRef();
            @Nonnull TensorArray tensorArray = TensorArray.wrap(IntStream.range(0, inObj[1].getData().length()).mapToObj(i -> i == 0 ? passback : passback.map(v -> 0)).toArray(i -> new Tensor[i]));
            inObj[1].accumulate(buffer, tensorArray);
        }
        Arrays.stream(inObj).forEach(x -> x.getData().addRef());
    }) {

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

        @Override
        public boolean isAlive() {
            return inObj[0].isAlive() || inObj[1].isAlive();
        }
    };
}

Example 25 with TensorArray

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

the class SimpleActivationLayer method eval.

@Nonnull
@Override
public Result eval(@Nonnull final Result... inObj) {
    final TensorList indata0 = inObj[0].getData();
    final int itemCnt = indata0.length();
    assert 0 < itemCnt;
    Arrays.stream(inObj).forEach(nnResult -> nnResult.addRef());
    Arrays.stream(inObj).forEach(nnResult -> nnResult.getData().addRef());
    @Nonnull final Tensor[] inputGradientA = new Tensor[itemCnt];
    return new Result(TensorArray.wrap(IntStream.range(0, itemCnt).parallel().mapToObj(dataIndex -> {
        @Nullable final Tensor input = indata0.get(dataIndex);
        @Nonnull final Tensor output = new Tensor(indata0.getDimensions());
        @Nonnull final Tensor inputGradient = new Tensor(input.length());
        inputGradientA[dataIndex] = inputGradient;
        @Nonnull final double[] results = new double[2];
        for (int i = 0; i < input.length(); i++) {
            eval(input.getData()[i], results);
            inputGradient.set(i, results[1]);
            output.set(i, results[0]);
        }
        input.freeRef();
        return output;
    }).toArray(i -> new Tensor[i])), (@Nonnull final DeltaSet<Layer> buffer, @Nonnull final TensorList data) -> {
        if (inObj[0].isAlive()) {
            @Nonnull TensorArray tensorArray = TensorArray.wrap(IntStream.range(0, itemCnt).parallel().mapToObj(dataIndex -> {
                @Nonnull final Tensor passback = new Tensor(data.getDimensions());
                @Nullable final double[] gradientData = inputGradientA[dataIndex].getData();
                @Nullable Tensor tensor = data.get(dataIndex);
                IntStream.range(0, passback.length()).forEach(i -> {
                    final double v = gradientData[i];
                    if (Double.isFinite(v)) {
                        passback.set(i, tensor.get(i) * v);
                    }
                });
                tensor.freeRef();
                return passback;
            }).toArray(i -> new Tensor[i]));
            inObj[0].accumulate(buffer, tensorArray);
        }
    }) {

        @Override
        protected void _free() {
            Arrays.stream(inObj).forEach(nnResult -> nnResult.freeRef());
            Arrays.stream(inObj).forEach(nnResult -> nnResult.getData().freeRef());
            for (@Nonnull Tensor tensor : inputGradientA) {
                tensor.freeRef();
            }
        }

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