Implement per-list sequence

[ttnghia]

This PR adds lists::sequences API, allowing to generate per-list sequence. In particular, it allows generating a lists column in which each list is a sequence of numbers/durations. These sequences are generated individually from separate sets of (start, step, size) input values.

Closes #9424.

Note: lists::sequences supports only numeric types (integer types + floating-point types) and duration types.

[vuule]

rerun tests

[wbo4958]

Hi @ttnghia

Looks like when size has some negative values, the sequences result will have some issues.

ColumnVector start = ColumnVector.fromBoxedInts(1, 2, 3, 4, 5, 6);
ColumnVector size = ColumnVector.fromBoxedInts(2, 7, -4, 2, 5, 2);

sequences(start, size) will get below result, the second row is not correct

1 2 
2 3 4 4 5 5 6 
null
4 5 
5 6 7 8 9 
6 7 

[wbo4958]

BTW, When I tested the sequences with float start and float-step. The result returned from CUDF seems to be INT32 type and the result seems not correct.

float[] x = {1.2f, 2.2f, 3.3f};
Integer[] sizeV = new Integer[]{1, 2, 3};
ColumnVector start = ColumnVector.fromFloats(x);
ColumnVector size = ColumnVector.fromBoxedInts(sizeV);
ColumnVector step = ColumnVector.fromFloats(x);

The output is like that

1 2 
2 3 4 5 6 7 8 
3 4 5 6 
4 5 
5 6 7 8 9 
6 7 

[ttnghia]

Hi @ttnghia

Looks like when size has some negative values, the sequences result will have some issues.

Yes, the API explicitly says that "if the input size is negative then output is undefined". Thus, the caller must pass in size column containing values that are at least 0.

[ttnghia]

BTW, When I tested the sequences with float start and float-step. The result returned from CUDF seems to be INT32 type and the result seems not correct.

Sorry I can't reproduce your issue. I ran your example and get this:

List<float>:
Length : 3
Offsets : 0, 1, 3, 6
   1.20000005, 2.20000005, 4.4000001, 3.29999995, 6.5999999, 9.89999962

So maybe something else is wrong with your test?

[wbo4958]

Okay, Got it. Thx for the explanation

[wbo4958]

BTW, When I tested the sequences with float start and float-step. The result returned from CUDF seems to be INT32 type and the result seems not correct.

Sorry I can't reproduce your issue. I ran your example and get this:

List<float>:
Length : 3
Offsets : 0, 1, 3, 6
   1.20000005, 2.20000005, 4.4000001, 3.29999995, 6.5999999, 9.89999962

So maybe something else is wrong with your test?

Sorry, it turned out my test was wrong.

[ttnghia]

@gpucibot merge