|
| 1 | +# |
| 2 | +# Copyright (c) 2023, NVIDIA CORPORATION. |
| 3 | +# |
| 4 | +# Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | +# you may not use this file except in compliance with the License. |
| 6 | +# You may obtain a copy of the License at |
| 7 | +# |
| 8 | +# http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | +# |
| 10 | +# Unless required by applicable law or agreed to in writing, software |
| 11 | +# distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | +# See the License for the specific language governing permissions and |
| 14 | +# limitations under the License. |
| 15 | +# |
| 16 | +import pytest |
| 17 | + |
| 18 | +import nvtabular as nvt |
| 19 | +from merlin.core.compat import cudf, numpy, pandas |
| 20 | +from merlin.core.dispatch import make_df |
| 21 | +from nvtabular import ColumnSelector, ops |
| 22 | +from tests.conftest import assert_eq |
| 23 | + |
| 24 | +if cudf: |
| 25 | + _CPU = [True, False] |
| 26 | +else: |
| 27 | + _CPU = [True] |
| 28 | + |
| 29 | + |
| 30 | +@pytest.mark.parametrize("cpu", _CPU) |
| 31 | +def test_list_slice(cpu): |
| 32 | + DataFrame = pandas.DataFrame if cpu else cudf.DataFrame |
| 33 | + |
| 34 | + df = DataFrame({"y": [[0, 1, 2, 2, 767], [1, 2, 2, 3], [1, 223, 4]]}) |
| 35 | + |
| 36 | + op = ops.ListSlice(0, 2) |
| 37 | + selector = ColumnSelector(["y"]) |
| 38 | + transformed = op.transform(selector, df) |
| 39 | + expected = DataFrame({"y": [[0, 1], [1, 2], [1, 223]]}) |
| 40 | + assert_eq(transformed, expected) |
| 41 | + |
| 42 | + op = ops.ListSlice(3, 5) |
| 43 | + transformed = op.transform(selector, df) |
| 44 | + expected = DataFrame({"y": [[2, 767], [3], []]}) |
| 45 | + assert_eq(transformed, expected) |
| 46 | + |
| 47 | + op = ops.ListSlice(4, 10) |
| 48 | + transformed = op.transform(selector, df) |
| 49 | + expected = DataFrame({"y": [[767], [], []]}) |
| 50 | + assert_eq(transformed, expected) |
| 51 | + |
| 52 | + op = ops.ListSlice(100, 20000) |
| 53 | + transformed = op.transform(selector, df) |
| 54 | + expected = DataFrame({"y": [[], [], []]}) |
| 55 | + assert_eq(transformed, expected) |
| 56 | + |
| 57 | + op = ops.ListSlice(-4) |
| 58 | + transformed = op.transform(selector, df) |
| 59 | + expected = DataFrame({"y": [[1, 2, 2, 767], [1, 2, 2, 3], [1, 223, 4]]}) |
| 60 | + assert_eq(transformed, expected) |
| 61 | + |
| 62 | + op = ops.ListSlice(-3, -1) |
| 63 | + transformed = op.transform(selector, df) |
| 64 | + expected = DataFrame({"y": [[2, 2], [2, 2], [1, 223]]}) |
| 65 | + assert_eq(transformed, expected) |
| 66 | + |
| 67 | + |
| 68 | +@pytest.mark.parametrize("cpu", _CPU) |
| 69 | +def test_list_slice_pad(cpu): |
| 70 | + DataFrame = pandas.DataFrame if cpu else cudf.DataFrame |
| 71 | + df = DataFrame({"y": [[0, 1, 2, 2, 767], [1, 2, 2, 3], [1, 223, 4]]}) |
| 72 | + |
| 73 | + # 0 pad to 5 elements |
| 74 | + op = ops.ListSlice(5, pad=True) |
| 75 | + selector = ColumnSelector(["y"]) |
| 76 | + transformed = op.transform(selector, df) |
| 77 | + expected = DataFrame({"y": [[0, 1, 2, 2, 767], [1, 2, 2, 3, 0], [1, 223, 4, 0, 0]]}) |
| 78 | + assert_eq(transformed, expected) |
| 79 | + |
| 80 | + # make sure we can also pad when start != 0, and when pad_value is set |
| 81 | + op = ops.ListSlice(1, 6, pad=True, pad_value=123) |
| 82 | + selector = ColumnSelector(["y"]) |
| 83 | + transformed = op.transform(selector, df) |
| 84 | + expected = DataFrame({"y": [[1, 2, 2, 767, 123], [2, 2, 3, 123, 123], [223, 4, 123, 123, 123]]}) |
| 85 | + assert_eq(transformed, expected) |
| 86 | + |
| 87 | + # we should be able to do pad out negative offsets as well |
| 88 | + op = ops.ListSlice(-4, pad=True, pad_value=-1) |
| 89 | + selector = ColumnSelector(["y"]) |
| 90 | + transformed = op.transform(selector, df) |
| 91 | + expected = DataFrame({"y": [[1, 2, 2, 767], [1, 2, 2, 3], [1, 223, 4, -1]]}) |
| 92 | + assert_eq(transformed, expected) |
| 93 | + |
| 94 | + op = ops.ListSlice(-4, -1, pad=True, pad_value=-1) |
| 95 | + selector = ColumnSelector(["y"]) |
| 96 | + transformed = op.transform(selector, df) |
| 97 | + expected = DataFrame({"y": [[1, 2, 2], [1, 2, 2], [1, 223, -1]]}) |
| 98 | + assert_eq(transformed, expected) |
| 99 | + |
| 100 | + op = ops.ListSlice(-4, pad=True, pad_value=-1) |
| 101 | + selector = ColumnSelector(["y"]) |
| 102 | + transformed = op.transform(selector, df) |
| 103 | + expected = DataFrame({"y": [[1, 2, 2, 767], [1, 2, 2, 3], [1, 223, 4, -1]]}) |
| 104 | + assert_eq(transformed, expected) |
| 105 | + |
| 106 | + |
| 107 | +def test_slice_ndarrays(): |
| 108 | + out = ["test"] >> nvt.ops.ListSlice(10, pad=True) |
| 109 | + workflow = nvt.Workflow(out) |
| 110 | + df = make_df({"test": [[x for x in numpy.asarray(range(1, 4)).astype(numpy.int32)]]}) |
| 111 | + workflow.fit(nvt.Dataset(df, cpu=True)) |
| 112 | + workflow.transform(nvt.Dataset(df, cpu=True)).compute() |
0 commit comments