Reputation: 1889
I have a double[] with K*N elements. I would like to split this into a stream (list/array/...) of N long[] arrays of size K. I already found a solution, but it looks quite dirty and requires a stateful Mapper (I think they are meant to be stateless):
private class DataToLong implements DoubleFunction<long[]> {
int d = 0;
long[] buf = new long[K];
@Override
public long[] apply(double value) {
buf[d] = BitTools.toSortableLong(value);
d++;
long[] ret = null;
if (d >= K) {
ret = buf;
buf = new long[K];
d = 0;
}
return ret;
}
}
public void load(double[] data, int K) {
Arrays.stream(data).mapToObj(new DataToLong())
.filter((x)-> x != null).forEach((buf)->{
//here we do something with 'buf'
});
}
The above code seems to work, but it's actually longer than the non-streams version of the code and it violates the stateless requirements of the Mapper. It there any better way to achieve the same thing?
Upvotes: 2
Views: 1994
Reputation: 1889
Here is another variant of @Louis Wasserman's answer, integrating a part from @srborlongan's answer. Tt avoids the 'new' and the loop:
IntStream.range(0, N)
.mapToObj(
i -> IntStream.range(0, DIM)
.mapToLong(
j -> pre(data[DIM * i + j])
).toArray()
).forEach(buf -> myFunction(buf));
Unfortunately, it seems to be 10-20% slower than @Louis Wasserman's solution
Upvotes: 1
Reputation: 4579
A variant of @Louis Wasserman's answer:
IntStream.range(0, n)
// for each int i, derive a DoubleStream that maps each
// j in range(0, k) to the double in data[k * i + j]
.mapToObj(
i -> IntStream.range(0, k)
.mapToDouble(j -> data[k * i + j])
)
// flatMap from Stream<DoubleStream> to DoubleStream
.flatMapToDouble(Function.identity())
// call BitTools::toSortableLong on every double
.mapToLong(BitTools::toSortableLong)
// collect into a long[]
.toArray()
;
Upvotes: 0
Reputation: 198033
IntStream.range(0, n).mapToObj(
i -> {
long[] arr = new long[k];
for (int j = 0; j < k; j++) {
arr[j] = BitTools.toSortableLong(data[k * i + j]);
}
return arr;
});
Upvotes: 2