1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
|
# BSD 3-Clause License; see https://github.com/scikit-hep/awkward/blob/main/LICENSE
from __future__ import annotations
import numpy as np
import pytest
import awkward as ak
jax = pytest.importorskip("jax")
ak.jax.register_and_check()
# Define all reducers to test
REDUCERS = [
(ak.argmin, {}),
(ak.argmax, {}),
(ak.min, {}),
(ak.max, {}),
(ak.sum, {}),
(ak.prod, {"mask_identity": True}), # mask_identity for prod to handle empty arrays
(ak.any, {}),
(ak.all, {}),
(ak.count, {}),
(ak.count_nonzero, {}),
]
# Define test arrays (single jagged)
SINGLE_JAGGED = [
# Normal array
[[1, 2, 3], [4, 5], [6, 7, 8, 9]],
# Array with first empty
[[], [1, 2], [3, 4, 5]],
# Array with middle empty
[[1, 2], [], [3, 4, 5]],
# Array with last empty
[[1, 2], [3, 4, 5], []],
# Array with multiple empty elements
[[], [1, 2], [], [3, 4], []],
# Array with negative numbers
[[-1, -2], [-3], [4, 5, -6]],
# Array with zeros
[[0, 0], [1, 0], [2, 3, 4]],
]
# Define test arrays (double jagged)
DOUBLE_JAGGED = [
# Normal double jagged array
[[[1, 2], [3]], [[4, 5, 6]], [[7], [8, 9]]],
# Double jagged with empty at first level
[[], [[1, 2], [3, 4]], [[5, 6]]],
# Double jagged with empty at second level
[[[1, 2], []], [[3, 4], [5]], [[6]]],
# Double jagged with various empty elements
[[[]], [[], [1, 2]], [[], [], [3, 4]]],
# Double jagged with negative numbers
[[[-1, -2], [-3]], [[4, 5, -6]], [[7], [-8, 9]]],
# Double jagged with zeros
[[[0, 0], [1]], [[2, 3]], [[4], [5, 6]]],
]
# Define axes to test
AXES = [1, None] # axis=1 for first dimension, None for flattened reduction
DOUBLE_JAGGED_AXES = [1, 2, None] # axis=1 and axis=2 for double jagged
RTOL = 1e-5 # Relative tolerance for floating point comparison
ATOL = 1e-8 # Absolute tolerance for floating point comparison
def compare_results(cpu_list, jax_list):
"""Compare results with tolerance for numeric values."""
if isinstance(cpu_list, (int, float)) and isinstance(jax_list, (int, float)):
# Direct numeric comparison with tolerance
np.testing.assert_allclose(cpu_list, jax_list, rtol=RTOL, atol=ATOL)
elif isinstance(cpu_list, list) and isinstance(jax_list, list):
# Lists should have the same length
assert len(cpu_list) == len(jax_list), (
f"Lists have different lengths: {len(cpu_list)} vs {len(jax_list)}"
)
# Compare each element
for cpu_item, jax_item in zip(cpu_list, jax_list):
compare_results(cpu_item, jax_item)
else:
# For non-numeric types, use exact equality
assert cpu_list == jax_list
@pytest.mark.parametrize("reducer,kwargs", REDUCERS)
@pytest.mark.parametrize("arr", SINGLE_JAGGED)
@pytest.mark.parametrize("axis", AXES)
def test_single_jagged_arrays(reducer, kwargs, arr, axis):
"""Test reducers on single jagged arrays with different axes."""
# Create arrays with different backends
cpu_array = ak.Array(arr, backend="cpu")
jax_array = ak.Array(arr, backend="jax")
# Apply reducers to each backend's array
cpu_result = reducer(cpu_array, axis=axis, **kwargs)
jax_result = reducer(jax_array, axis=axis, **kwargs)
# Convert to lists for comparison
cpu_list = ak.to_list(cpu_result)
jax_list = ak.to_list(jax_result)
# Handle case where axis=None might result in different structures
if axis is None:
# If one result is a scalar and the other is a list with one element
if (
not isinstance(cpu_list, list)
and isinstance(jax_list, list)
and len(jax_list) == 1
):
jax_list = jax_list[0]
elif (
isinstance(cpu_list, list)
and not isinstance(jax_list, list)
and len(cpu_list) == 1
):
cpu_list = cpu_list[0]
# Compare with tolerance for numeric values
compare_results(cpu_list, jax_list)
@pytest.mark.parametrize("reducer,kwargs", REDUCERS)
@pytest.mark.parametrize("arr", DOUBLE_JAGGED)
@pytest.mark.parametrize("axis", DOUBLE_JAGGED_AXES)
def test_double_jagged_arrays(reducer, kwargs, arr, axis):
"""Test reducers on double jagged arrays with different axes."""
# Create arrays with different backends
cpu_array = ak.Array(arr, backend="cpu")
jax_array = ak.Array(arr, backend="jax")
# Apply reducers to each backend's array
cpu_result = reducer(cpu_array, axis=axis, **kwargs)
jax_result = reducer(jax_array, axis=axis, **kwargs)
# Convert to lists for comparison
cpu_list = ak.to_list(cpu_result)
jax_list = ak.to_list(jax_result)
# Handle case where axis=None might result in different structures
if axis is None:
# If one result is a scalar and the other is a list with one element
if (
not isinstance(cpu_list, list)
and isinstance(jax_list, list)
and len(jax_list) == 1
):
jax_list = jax_list[0]
elif (
isinstance(cpu_list, list)
and not isinstance(jax_list, list)
and len(cpu_list) == 1
):
cpu_list = cpu_list[0]
# Compare with tolerance for numeric values
compare_results(cpu_list, jax_list)
# Additional edge cases
@pytest.mark.parametrize("reducer,kwargs", REDUCERS)
def test_all_empty_arrays(reducer, kwargs):
"""Test with arrays that are entirely empty."""
all_empty_data = [[], [], []]
cpu_array = ak.Array(all_empty_data, backend="cpu")
jax_array = ak.Array(all_empty_data, backend="jax")
cpu_result = reducer(cpu_array, axis=1, **kwargs)
jax_result = reducer(jax_array, axis=1, **kwargs)
# Convert to lists for comparison
cpu_list = ak.to_list(cpu_result)
jax_list = ak.to_list(jax_result)
# Handle case where one might be a scalar and the other a list
if (
not isinstance(cpu_list, list)
and isinstance(jax_list, list)
and len(jax_list) == 1
):
jax_list = jax_list[0]
elif (
isinstance(cpu_list, list)
and not isinstance(jax_list, list)
and len(cpu_list) == 1
):
cpu_list = cpu_list[0]
# Compare with tolerance for numeric values
compare_results(cpu_list, jax_list)
# Test with boolean values
@pytest.mark.parametrize("reducer,kwargs", REDUCERS)
def test_boolean_arrays(reducer, kwargs):
"""Test with boolean arrays."""
bool_data = [[True, False], [], [True, True, False], [False]]
cpu_array = ak.Array(bool_data, backend="cpu")
jax_array = ak.Array(bool_data, backend="jax")
cpu_result = reducer(cpu_array, axis=1, **kwargs)
jax_result = reducer(jax_array, axis=1, **kwargs)
# Convert to lists for comparison
cpu_list = ak.to_list(cpu_result)
jax_list = ak.to_list(jax_result)
# Handle case where one might be a scalar and the other a list
if (
not isinstance(cpu_list, list)
and isinstance(jax_list, list)
and len(jax_list) == 1
):
jax_list = jax_list[0]
elif (
isinstance(cpu_list, list)
and not isinstance(jax_list, list)
and len(cpu_list) == 1
):
cpu_list = cpu_list[0]
# Compare with tolerance for numeric values
compare_results(cpu_list, jax_list)
# Test with None values
@pytest.mark.parametrize("reducer,kwargs", REDUCERS)
def test_none_arrays(reducer, kwargs):
"""Test with arrays containing None values."""
none_data = [[None, 1], [2, None], [None, None], [3, 4]]
cpu_array = ak.Array(none_data, backend="cpu")
jax_array = ak.Array(none_data, backend="jax")
cpu_result = reducer(cpu_array, axis=1, **kwargs)
jax_result = reducer(jax_array, axis=1, **kwargs)
# Convert to lists for comparison
cpu_list = ak.to_list(cpu_result)
jax_list = ak.to_list(jax_result)
# Handle case where one might be a scalar and the other a list
if (
not isinstance(cpu_list, list)
and isinstance(jax_list, list)
and len(jax_list) == 1
):
jax_list = jax_list[0]
elif (
isinstance(cpu_list, list)
and not isinstance(jax_list, list)
and len(cpu_list) == 1
):
cpu_list = cpu_list[0]
# Compare with tolerance for numeric values
compare_results(cpu_list, jax_list)
# test with NaN values
@pytest.mark.skip(
reason="(arg)min/max and any do not work with NaNs in the jax backend"
)
@pytest.mark.parametrize("reducer,kwargs", REDUCERS)
def test_nan_arrays(reducer, kwargs):
"""Test with arrays containing NaN values."""
nan_data = [[np.nan, 1], [2, np.nan], [np.nan, np.nan], [3, 4]]
cpu_array = ak.Array(nan_data, backend="cpu")
jax_array = ak.Array(nan_data, backend="jax")
cpu_result = reducer(cpu_array, axis=1, **kwargs)
jax_result = reducer(jax_array, axis=1, **kwargs)
# Convert to lists for comparison
cpu_list = ak.to_list(cpu_result)
jax_list = ak.to_list(jax_result)
# Handle case where one might be a scalar and the other a list
if (
not isinstance(cpu_list, list)
and isinstance(jax_list, list)
and len(jax_list) == 1
):
jax_list = jax_list[0]
elif (
isinstance(cpu_list, list)
and not isinstance(jax_list, list)
and len(cpu_list) == 1
):
cpu_list = cpu_list[0]
# Compare with tolerance for numeric values
compare_results(cpu_list, jax_list)
|
