Examples with Vector com.tencent.angel.ml.math2.vector.Vector used on opensource projects

Example 6 with Vector

use of com.tencent.angel.ml.math2.vector.Vector in project angel by Tencent.

the class RangeRouterUtils method splitIntIntVector.

public static KeyValuePart[] splitIntIntVector(MatrixMeta matrixMeta, IntIntVector vector) {
    IntIntVectorStorage storage = vector.getStorage();
    if (storage.isSparse()) {
        // Get keys and values
        IntIntSparseVectorStorage sparseStorage = (IntIntSparseVectorStorage) storage;
        int[] keys = sparseStorage.getIndices();
        int[] values = sparseStorage.getValues();
        return split(matrixMeta, vector.getRowId(), keys, values, false);
    } else if (storage.isDense()) {
        // Get values
        IntIntDenseVectorStorage denseStorage = (IntIntDenseVectorStorage) storage;
        int[] values = denseStorage.getValues();
        return split(matrixMeta, vector.getRowId(), values);
    } else {
        // Key and value array pair
        IntIntSortedVectorStorage sortStorage = (IntIntSortedVectorStorage) storage;
        int[] keys = sortStorage.getIndices();
        int[] values = sortStorage.getValues();
        return split(matrixMeta, vector.getRowId(), keys, values, true);
    }
}

Example 7 with Vector

use of com.tencent.angel.ml.math2.vector.Vector in project angel by Tencent.

the class RangeRouterUtils method splitLongDoubleVector.

public static KeyValuePart[] splitLongDoubleVector(MatrixMeta matrixMeta, LongDoubleVector vector) {
    LongDoubleVectorStorage storage = vector.getStorage();
    if (storage.isSparse()) {
        // Get keys and values
        LongDoubleSparseVectorStorage sparseStorage = (LongDoubleSparseVectorStorage) storage;
        long[] keys = sparseStorage.getIndices();
        double[] values = sparseStorage.getValues();
        return split(matrixMeta, vector.getRowId(), keys, values, false);
    } else {
        // Key and value array pair
        LongDoubleSortedVectorStorage sortStorage = (LongDoubleSortedVectorStorage) storage;
        long[] keys = sortStorage.getIndices();
        double[] values = sortStorage.getValues();
        return split(matrixMeta, vector.getRowId(), keys, values, true);
    }
}

Example 8 with Vector

use of com.tencent.angel.ml.math2.vector.Vector in project angel by Tencent.

the class Min method partitionUpdate.

@Override
public void partitionUpdate(PartitionUpdateParam partParam) {
    PartIncrementRowsParam param = (PartIncrementRowsParam) partParam;
    List<RowUpdateSplit> updates = param.getUpdates();
    for (RowUpdateSplit update : updates) {
        ServerRow row = psContext.getMatrixStorageManager().getRow(param.getPartKey(), update.getRowId());
        row.startWrite();
        try {
            Vector vector = getVector(param.getMatrixId(), update.getRowId(), param.getPartKey());
            Ufuncs.imin(vector, update.getVector());
        } finally {
            row.endWrite();
        }
    }
}

Example 9 with Vector

use of com.tencent.angel.ml.math2.vector.Vector in project angel by Tencent.

the class GetRowHandler method handle.

@Override
public void handle(FutureResult finalResult, UserRequest userRequest, ResponseCache responseCache) {
    GetRowRequest getRowRequest = (GetRowRequest) userRequest;
    MapResponseCache cache = (MapResponseCache) responseCache;
    // Merge the sub-response
    List<ServerRow> serverRows = new ArrayList<>(cache.expectedResponseNum);
    for (Response response : cache.getResponses().values()) {
        serverRows.add(((GetRowSplitResponse) (response.getData())).getRowSplit());
    }
    Vector vector = MergeUtils.combineServerRowSplits(serverRows, getRowRequest.getMatrixId(), getRowRequest.getRowId());
    // Set matrix/row information
    vector.setMatrixId(getRowRequest.getMatrixId());
    vector.setRowId(getRowRequest.getRowId());
    // Set result
    finalResult.set(vector);
}

Example 10 with Vector

use of com.tencent.angel.ml.math2.vector.Vector in project angel by Tencent.

the class AdaGradUpdateFunc method update.

@Override
public void update(RowBasedPartition partition, int factor, double[] scalars) {
    double epsilon = scalars[0];
    double beta = scalars[1];
    double lr = scalars[2];
    double l1RegParam = scalars[3];
    double l2RegParam = scalars[4];
    double epoch = (int) scalars[5];
    double batchSize = (int) scalars[6];
    for (int f = 0; f < factor; f++) {
        ServerRow gradientServerRow = partition.getRow(f + 2 * factor);
        try {
            gradientServerRow.startWrite();
            Vector weight = ServerRowUtils.getVector(partition.getRow(f));
            Vector square = ServerRowUtils.getVector(partition.getRow(f + factor));
            Vector gradient = ServerRowUtils.getVector(gradientServerRow);
            if (batchSize > 1) {
                gradient.idiv(batchSize);
            }
            OptFuncs.iexpsmoothing2(square, gradient, beta);
            if (l2RegParam != 0) {
                gradient.iaxpy(weight, l2RegParam);
            }
            OptFuncs.iadagraddelta(gradient, square, l2RegParam, lr);
            weight.isub(gradient);
            if (l1RegParam != 0) {
                OptFuncs.iadagradthredshold(weight, square, l1RegParam, l2RegParam, lr);
            }
            gradient.clear();
        } finally {
            gradientServerRow.endWrite();
        }
    }
}