mirror of
https://github.com/saymrwulf/pytorch.git
synced 2026-05-15 21:00:47 +00:00
a745981c94
Summary: Implement ReduceBackSum & ReduceBackMean with gradients for CPU & GPU contexts. The reduction happens among the last dimenstions for example if input is a M x N matrix ReduceBackSum will results a vector of dim M x 1 contains the rowwise sums. Differential Revision: D4689768 fbshipit-source-id: 5b0482d4341867ecf23526dc6c4d544420e7d8f7
69 lines
2.2 KiB
Python
69 lines
2.2 KiB
Python
from __future__ import absolute_import
|
|
from __future__ import division
|
|
from __future__ import print_function
|
|
from __future__ import unicode_literals
|
|
|
|
from caffe2.python import core
|
|
from hypothesis import given
|
|
|
|
import caffe2.python.hypothesis_test_util as hu
|
|
import hypothesis.strategies as st
|
|
import numpy as np
|
|
|
|
|
|
class TestReduceFrontSum(hu.HypothesisTestCase):
|
|
|
|
def reduce_op_test(self, op_name, op_ref, in_data, num_reduce_dims, device):
|
|
op = core.CreateOperator(
|
|
op_name,
|
|
["inputs"],
|
|
["outputs"],
|
|
num_reduce_dim=num_reduce_dims
|
|
)
|
|
|
|
self.assertReferenceChecks(
|
|
device_option=device,
|
|
op=op,
|
|
inputs=[in_data],
|
|
reference=op_ref
|
|
)
|
|
|
|
self.assertGradientChecks(
|
|
device, op, [in_data], 0, [0], stepsize=1e-2, threshold=1e-2)
|
|
|
|
@given(num_reduce_dim=st.integers(1, 3), **hu.gcs)
|
|
def test_reduce_front_sum(self, num_reduce_dim, gc, dc):
|
|
X = np.random.rand(7, 4, 3, 5).astype(np.float32)
|
|
|
|
def ref_sum(X):
|
|
return [np.sum(X, axis=(tuple(range(num_reduce_dim))))]
|
|
|
|
self.reduce_op_test("ReduceFrontSum", ref_sum, X, num_reduce_dim, gc)
|
|
|
|
@given(num_reduce_dim=st.integers(1, 3), **hu.gcs)
|
|
def test_reduce_front_mean(self, num_reduce_dim, gc, dc):
|
|
X = np.random.rand(6, 7, 8, 2).astype(np.float32)
|
|
|
|
def ref_mean(X):
|
|
return [np.mean(X, axis=(tuple(range(num_reduce_dim))))]
|
|
|
|
self.reduce_op_test("ReduceFrontMean", ref_mean, X, num_reduce_dim, gc)
|
|
|
|
@given(num_reduce_dim=st.integers(1, 3), **hu.gcs)
|
|
def test_reduce_back_sum(self, num_reduce_dim, dc, gc):
|
|
X = np.random.rand(6, 7, 8, 2).astype(np.float32)
|
|
|
|
def ref_sum(X):
|
|
return [np.sum(X, axis=(0, 1, 2, 3)[4 - num_reduce_dim:])]
|
|
|
|
self.reduce_op_test("ReduceBackSum", ref_sum, X, num_reduce_dim, gc)
|
|
|
|
@given(num_reduce_dim=st.integers(1, 3), **hu.gcs)
|
|
def test_reduce_back_mean(self, num_reduce_dim, dc, gc):
|
|
num_reduce_dim = 2
|
|
X = np.random.rand(6, 7, 8, 2).astype(np.float32)
|
|
|
|
def ref_sum(X):
|
|
return [np.mean(X, axis=(0, 1, 2, 3)[4 - num_reduce_dim:])]
|
|
|
|
self.reduce_op_test("ReduceBackMean", ref_sum, X, num_reduce_dim, gc)
|