2023-04-30 16:10:07 +00:00
|
|
|
"""
|
|
|
|
segment_sum
|
|
|
|
top_k
|
2023-05-17 03:35:11 +00:00
|
|
|
in_top_k
|
|
|
|
logsumexp
|
2023-04-30 16:10:07 +00:00
|
|
|
"""
|
|
|
|
|
|
|
|
from keras_core import backend
|
2023-05-17 03:35:11 +00:00
|
|
|
from keras_core.api_export import keras_core_export
|
2023-05-19 01:04:25 +00:00
|
|
|
from keras_core.backend import KerasTensor
|
2023-04-30 16:10:07 +00:00
|
|
|
from keras_core.backend import any_symbolic_tensors
|
|
|
|
from keras_core.operations.operation import Operation
|
2023-05-19 01:04:25 +00:00
|
|
|
from keras_core.operations.operation_utils import reduce_shape
|
2023-04-30 16:10:07 +00:00
|
|
|
|
|
|
|
|
|
|
|
class SegmentSum(Operation):
|
2023-05-19 01:04:25 +00:00
|
|
|
def __init__(self, num_segments=None, sorted=False):
|
|
|
|
super().__init__()
|
|
|
|
self.num_segments = num_segments
|
|
|
|
self.sorted = sorted
|
|
|
|
|
|
|
|
def compute_output_spec(self, data, segment_ids):
|
|
|
|
num_segments = self.num_segments
|
|
|
|
output_shape = (num_segments,) + tuple(data.shape[1:])
|
|
|
|
return KerasTensor(shape=output_shape, dtype=data.dtype)
|
|
|
|
|
|
|
|
def call(self, data, segment_ids):
|
|
|
|
return backend.math.segment_sum(
|
|
|
|
data,
|
|
|
|
segment_ids,
|
|
|
|
num_segments=self.num_segments,
|
|
|
|
sorted=self.sorted,
|
|
|
|
)
|
2023-04-30 16:10:07 +00:00
|
|
|
|
|
|
|
|
2023-05-17 03:35:11 +00:00
|
|
|
@keras_core_export("keras_core.operations.segment_sum")
|
2023-05-05 16:53:36 +00:00
|
|
|
def segment_sum(data, segment_ids, num_segments=None, sorted=False):
|
|
|
|
if any_symbolic_tensors((data,)):
|
2023-05-19 01:04:25 +00:00
|
|
|
return SegmentSum(num_segments, sorted).symbolic_call(data, segment_ids)
|
|
|
|
return backend.math.segment_sum(
|
|
|
|
data, segment_ids, num_segments=num_segments, sorted=sorted
|
|
|
|
)
|
2023-04-30 16:10:07 +00:00
|
|
|
|
|
|
|
|
|
|
|
class TopK(Operation):
|
2023-05-19 01:04:25 +00:00
|
|
|
def __init__(self, k, sorted=False):
|
|
|
|
super().__init__()
|
|
|
|
self.k = k
|
|
|
|
self.sorted = sorted
|
|
|
|
|
|
|
|
def compute_output_spec(self, x):
|
|
|
|
output_shape = list(x.shape)
|
|
|
|
output_shape[-1] = self.k
|
|
|
|
# Return a tuple (values, indices).
|
|
|
|
return (
|
|
|
|
KerasTensor(shape=output_shape, dtype=x.dtype),
|
|
|
|
KerasTensor(shape=output_shape, dtype="int32"),
|
|
|
|
)
|
|
|
|
|
|
|
|
def call(self, x):
|
|
|
|
return backend.math.top_k(x, self.k, self.sorted)
|
2023-04-30 16:10:07 +00:00
|
|
|
|
|
|
|
|
2023-05-17 03:35:11 +00:00
|
|
|
@keras_core_export("keras_core.operations.top_k")
|
2023-04-30 16:10:07 +00:00
|
|
|
def top_k(x, k, sorted=True):
|
|
|
|
if any_symbolic_tensors((x,)):
|
2023-05-19 01:04:25 +00:00
|
|
|
return TopK(k, sorted).symbolic_call(x)
|
2023-04-30 16:10:07 +00:00
|
|
|
return backend.math.top_k(x, k, sorted)
|
2023-05-01 18:17:26 +00:00
|
|
|
|
|
|
|
|
|
|
|
class InTopK(Operation):
|
2023-05-19 01:04:25 +00:00
|
|
|
def __init__(self, k):
|
|
|
|
super().__init__()
|
|
|
|
self.k = k
|
|
|
|
|
|
|
|
def compute_output_spec(self, targets, predictions):
|
|
|
|
return KerasTensor(shape=targets.shape, dtype="bool")
|
|
|
|
|
|
|
|
def call(self, targets, predictions):
|
|
|
|
return backend.math.in_top_k(targets, predictions, self.k)
|
2023-05-01 18:17:26 +00:00
|
|
|
|
|
|
|
|
2023-05-17 03:35:11 +00:00
|
|
|
@keras_core_export("keras_core.operations.in_top_k")
|
2023-05-01 18:17:26 +00:00
|
|
|
def in_top_k(targets, predictions, k):
|
|
|
|
if any_symbolic_tensors((targets, predictions)):
|
2023-05-19 01:04:25 +00:00
|
|
|
return InTopK(k).symbolic_call(targets, predictions)
|
2023-05-01 18:17:26 +00:00
|
|
|
return backend.math.in_top_k(targets, predictions, k)
|
2023-05-09 23:03:15 +00:00
|
|
|
|
|
|
|
|
2023-05-17 03:35:11 +00:00
|
|
|
class Logsumexp(Operation):
|
2023-05-19 01:04:25 +00:00
|
|
|
def __init__(self, axis=None, keepdims=False):
|
|
|
|
super().__init__()
|
|
|
|
self.axis = axis
|
|
|
|
self.keepdims = keepdims
|
|
|
|
|
|
|
|
def compute_output_spec(self, x):
|
|
|
|
output_shape = reduce_shape(x.shape, self.axis, self.keepdims)
|
|
|
|
return KerasTensor(shape=output_shape)
|
|
|
|
|
|
|
|
def call(self, x):
|
|
|
|
return backend.math.logsumexp(x, axis=self.axis, keepdims=self.keepdims)
|
2023-05-17 03:35:11 +00:00
|
|
|
|
|
|
|
|
|
|
|
@keras_core_export("keras_core.operations.logsumexp")
|
2023-05-09 23:03:15 +00:00
|
|
|
def logsumexp(x, axis=None, keepdims=False):
|
|
|
|
if any_symbolic_tensors((x,)):
|
2023-05-19 01:04:25 +00:00
|
|
|
return Logsumexp(axis, keepdims).symbolic_call(x)
|
2023-05-09 23:03:15 +00:00
|
|
|
return backend.math.logsumexp(x, axis=axis, keepdims=keepdims)
|