Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Sign up
Appearance settings

Commit da192a7

Browse files
9prady9syurkevi
authored andcommitted
Refactor cov, var, stdev APIs to reflect 3.8 release
Also adds set_cublas_mode utility function to explicitly enable tensor ops for blas functions
1 parent 8ac4a8d commit da192a7

File tree

5 files changed

+71
-32
lines changed

5 files changed

+71
-32
lines changed

‎arrayfire/algorithm.py‎

Lines changed: 28 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -250,6 +250,34 @@ def max(a, dim=None):
250250
else:
251251
return _reduce_all(a, backend.get().af_max_all)
252252

253+
def maxRagged(vals, lens, dim):
254+
"""
255+
Find the maximum value of a subset of elements along a specified dimension.
256+
257+
The size of the subset of elements along the given dimension are decided based on the lengths
258+
provided in the `lens` array.
259+
260+
Parameters
261+
----------
262+
vals : af.Array
263+
Multi dimensional arrayfire array.
264+
lens : af.Array
265+
Multi dimensional arrayfire array containing number of elements to reduce along given `dim`
266+
dim: optional: int. default: None
267+
Dimension along which the maximum value is required.
268+
269+
Returns
270+
-------
271+
(values, indices): A tuple of af.Array(s)
272+
`values` af.Array will have the maximum values along given dimension for
273+
subsets determined by lengths provided in `lens`
274+
`idx` contains the locations of the maximum values as per the lengths provided in `lens`
275+
"""
276+
out_vals = Array()
277+
out_idx = Array()
278+
safe_call(backend().get().af_max_ragged(c_pointer(out_vals.arr), c_pointer(out_idx.arr), c_pointer(vals.arr), c_pointer(lens.arr), c_int_t(dim)))
279+
return out_vals, out_idx
280+
253281
def maxByKey(keys, vals, dim=-1):
254282
"""
255283
Calculate the max of elements along a specified dimension according to a key.

‎arrayfire/cuda.py‎

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -85,3 +85,10 @@ def set_native_id(idx):
8585

8686
safe_call(backend.get().afcu_set_native_id(idx))
8787
return
88+
89+
def set_cublas_mode(mode=CUBLAS_MATH_MODE.DEFAULT):
90+
"""
91+
Set's cuBLAS math mode for CUDA backend. In other backends, this has no effect.
92+
"""
93+
safe_call(backend().get().afcu_cublasSetMathMode(mode.value))
94+
return

‎arrayfire/library.py‎

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -490,6 +490,13 @@ class VARIANCE(_Enum):
490490
SAMPLE = _Enum_Type(1)
491491
POPULATION = _Enum_Type(2)
492492

493+
class CUBLAS_MATH_MODE(_Enum):
494+
"""
495+
Enable Tensor Core usage if available on CUDA backend GPUs
496+
"""
497+
DEFAULT = _Enum_Type(0)
498+
TENSOR_OP = _Enum_Type(1)
499+
493500
_VER_MAJOR_PLACEHOLDER = "__VER_MAJOR__"
494501

495502
def _setup():

‎arrayfire/statistics.py‎

Lines changed: 25 additions & 28 deletions
Original file line numberDiff line numberDiff line change
@@ -59,7 +59,7 @@ def mean(a, weights=None, dim=None):
5959

6060
return real if imag == 0 else real + imag * 1j
6161

62-
def var(a, isbiased=False, weights=None, dim=None):
62+
def var(a, bias=VARIANCE.DEFAULT, weights=None, dim=None):
6363
"""
6464
Calculate variance along a given dimension.
6565
@@ -68,9 +68,9 @@ def var(a, isbiased=False, weights=None, dim=None):
6868
a: af.Array
6969
The input array.
7070
71-
isbiased: optional: Boolean. default: False.
72-
Boolean denoting population variance (false) or sample
73-
variance (true).
71+
bias: optional: af.VARIANCE. default: DEFAULT.
72+
population variance(VARIANCE.POPULATION) or sample variance(VARIANCE.SAMPLE).
73+
This is ignored if weights are provided.
7474
7575
weights: optional: af.Array. default: None.
7676
Array to calculate for the weighted mean. Must match size of
@@ -89,7 +89,7 @@ def var(a, isbiased=False, weights=None, dim=None):
8989
out = Array()
9090

9191
if weights is None:
92-
safe_call(backend.get().af_var(c_pointer(out.arr), a.arr, isbiased, c_int_t(dim)))
92+
safe_call(backend.get().af_var_v2(c_pointer(out.arr), a.arr, bias.value, c_int_t(dim)))
9393
else:
9494
safe_call(backend.get().af_var_weighted(c_pointer(out.arr), a.arr, weights.arr, c_int_t(dim)))
9595

@@ -99,7 +99,7 @@ def var(a, isbiased=False, weights=None, dim=None):
9999
imag = c_double_t(0)
100100

101101
if weights is None:
102-
safe_call(backend.get().af_var_all(c_pointer(real), c_pointer(imag), a.arr, isbiased))
102+
safe_call(backend.get().af_var_all_v2(c_pointer(real), c_pointer(imag), a.arr, bias.value))
103103
else:
104104
safe_call(backend.get().af_var_all_weighted(c_pointer(real), c_pointer(imag), a.arr, weights.arr))
105105

@@ -150,7 +150,7 @@ def meanvar(a, weights=None, bias=VARIANCE.DEFAULT, dim=-1):
150150
return mean_out, var_out
151151

152152

153-
def stdev(a, dim=None):
153+
def stdev(a, bias=VARIANCE.DEFAULT, dim=None):
154154
"""
155155
Calculate standard deviation along a given dimension.
156156
@@ -159,6 +159,10 @@ def stdev(a, dim=None):
159159
a: af.Array
160160
The input array.
161161
162+
bias: optional: af.VARIANCE. default: DEFAULT.
163+
population variance(VARIANCE.POPULATION) or sample variance(VARIANCE.SAMPLE).
164+
This is ignored if weights are provided.
165+
162166
dim: optional: int. default: None.
163167
The dimension for which to obtain the standard deviation from
164168
input data.
@@ -171,48 +175,41 @@ def stdev(a, dim=None):
171175
"""
172176
if dim is not None:
173177
out = Array()
174-
safe_call(backend.get().af_stdev(c_pointer(out.arr), a.arr, c_int_t(dim)))
178+
safe_call(backend.get().af_stdev_v2(c_pointer(out.arr), a.arr, bias.value,
179+
c_int_t(dim)))
175180
return out
176181
else:
177182
real = c_double_t(0)
178183
imag = c_double_t(0)
179-
safe_call(backend.get().af_stdev_all(c_pointer(real), c_pointer(imag), a.arr))
184+
safe_call(backend.get().af_stdev_all_v2(c_pointer(real), c_pointer(imag), a.arr,
185+
bias.value))
180186
real = real.value
181187
imag = imag.value
182188
return real if imag == 0 else real + imag * 1j
183189

184-
def cov(a, isbiased=False, dim=None):
190+
def cov(a, b, bias=VARIANCE.DEFAULT):
185191
"""
186192
Calculate covariance along a given dimension.
187193
188194
Parameters
189195
----------
190196
a: af.Array
191-
The input array.
197+
Input array.
192198
193-
isbiased: optional: Boolean. default: False.
194-
Boolean denoting whether biased estimate should be taken.
199+
b: af.Array
200+
Input array.
195201
196-
dim: optional: int. default: None.
197-
The dimension for which to obtain the covariance from input data.
202+
bias: optional: af.VARIANCE. default: DEFAULT.
203+
population variance(VARIANCE.POPULATION) or sample variance(VARIANCE.SAMPLE).
198204
199205
Returns
200206
-------
201207
output: af.Array
202-
Array containing the covariance of the input array along a
203-
given dimension.
208+
Array containing the covariance of the input array along a given dimension.
204209
"""
205-
if dim is not None:
206-
out = Array()
207-
safe_call(backend.get().af_cov(c_pointer(out.arr), a.arr, isbiased, c_int_t(dim)))
208-
return out
209-
else:
210-
real = c_double_t(0)
211-
imag = c_double_t(0)
212-
safe_call(backend.get().af_cov_all(c_pointer(real), c_pointer(imag), a.arr, isbiased))
213-
real = real.value
214-
imag = imag.value
215-
return real if imag == 0 else real + imag * 1j
210+
out = Array()
211+
safe_call(backend.get().af_cov_v2(c_pointer(out.arr), a.arr, b.arr, bias.value))
212+
return out
216213

217214
def median(a, dim=None):
218215
"""

‎tests/simple/statistics.py‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -28,10 +28,10 @@ def simple_statistics(verbose=False):
2828
print_func(af.mean(a, weights=w))
2929

3030
display_func(af.var(a, dim=0))
31-
display_func(af.var(a, isbiased=True, dim=0))
31+
display_func(af.var(a, bias=af.VARIANCE.SAMPLE, dim=0))
3232
display_func(af.var(a, weights=w, dim=0))
3333
print_func(af.var(a))
34-
print_func(af.var(a, isbiased=True))
34+
print_func(af.var(a, bias=af.VARIANCE.SAMPLE))
3535
print_func(af.var(a, weights=w))
3636

3737
mean, var = af.meanvar(a, dim=0)
@@ -45,9 +45,9 @@ def simple_statistics(verbose=False):
4545
print_func(af.stdev(a))
4646

4747
display_func(af.var(a, dim=0))
48-
display_func(af.var(a, isbiased=True, dim=0))
48+
display_func(af.var(a, bias=af.VARIANCE.SAMPLE, dim=0))
4949
print_func(af.var(a))
50-
print_func(af.var(a, isbiased=True))
50+
print_func(af.var(a, bias=af.VARIANCE.SAMPLE))
5151

5252
display_func(af.median(a, dim=0))
5353
print_func(af.median(w))

0 commit comments

Comments
(0)

AltStyle によって変換されたページ (->オリジナル) /