blob: 243922d96943d2c10a32d41eca1908124f92c3ce (
plain)
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
|
op {
graph_op_name: "ExperimentalNumaMapAndBatchDataset"
visibility: HIDDEN
in_arg {
name: "input_dataset"
description: <<END
A variant tensor representing the input dataset.
END
}
in_arg {
name: "other_arguments"
description: <<END
A list of tensors, typically values that were captured when building a closure
for `f`.
END
}
in_arg {
name: "batch_size"
description: <<END
A scalar representing the number of elements to accumulate in a
batch. It determines the number of concurrent invocations of `f` that process
elements from `input_dataset` in parallel.
END
}
in_arg {
name: "num_parallel_calls"
description: <<END
A scalar representing the maximum number of parallel invocations of the `map_fn`
function. Applying the `map_fn` on consecutive input elements in parallel has
the potential to improve input pipeline throughput.
END
}
in_arg {
name: "drop_remainder"
description: <<END
A scalar representing whether the last batch should be dropped in case its size
is smaller than desired.
END
}
attr {
name: "f"
description: <<END
A function to apply to the outputs of `input_dataset`.
END
}
summary: "Creates a dataset that fuses mapping with batching."
description: <<END
Creates a dataset that applies `f` to the outputs of `input_dataset` and then
batches `batch_size` of them.
Unlike a "MapDataset", which applies `f` sequentially, this dataset invokes up
to `batch_size * num_parallel_batches` copies of `f` in parallel.
Unlike "MapAndBatchDatasetV2", this dataset uses a NUMA-aware thread scheduling
policy. Because it uses the single-threaded executor, it only supports the
function-based control flow ops.
END
}
|
