mirror of
https://github.com/saymrwulf/pytorch.git
synced 2026-05-15 21:00:47 +00:00
ae1ed27756
Summary: `np.str` is removed from numpy 1.20.0. It was an alias to builtin `str` and it's safe to do the replacement. The whole changes is mechanical, generated using the following onliner: ``` fbgr -sl 'np\.str\b' | xargs perl -pi -e 's,\bnp\.str\b,str,g' ``` Test Plan: sandcastle Differential Revision: D46586144 Pull Request resolved: https://github.com/pytorch/pytorch/pull/103931 Approved by: https://github.com/huydhn
117 lines
3.8 KiB
Python
117 lines
3.8 KiB
Python
|
|
|
|
|
|
|
|
|
|
import numpy as np
|
|
|
|
from hypothesis import given, settings
|
|
import hypothesis.strategies as st
|
|
import unittest
|
|
|
|
from caffe2.python import core, workspace
|
|
import caffe2.python.hypothesis_test_util as hu
|
|
import caffe2.python.serialized_test.serialized_test_util as serial
|
|
|
|
|
|
class TestTile(serial.SerializedTestCase):
|
|
@given(M=st.integers(min_value=1, max_value=10),
|
|
K=st.integers(min_value=1, max_value=10),
|
|
N=st.integers(min_value=1, max_value=10),
|
|
tiles=st.integers(min_value=1, max_value=3),
|
|
axis=st.integers(min_value=0, max_value=2),
|
|
**hu.gcs)
|
|
@settings(deadline=10000)
|
|
def test_tile(self, M, K, N, tiles, axis, gc, dc):
|
|
X = np.random.rand(M, K, N).astype(np.float32)
|
|
|
|
op = core.CreateOperator(
|
|
'Tile', ['X'], 'out',
|
|
tiles=tiles,
|
|
axis=axis,
|
|
)
|
|
|
|
def tile_ref(X, tiles, axis):
|
|
dims = np.asarray([1, 1, 1], dtype=int)
|
|
dims[axis] = tiles
|
|
tiled_data = np.tile(X, dims)
|
|
return (tiled_data,)
|
|
|
|
# Check against numpy reference
|
|
self.assertReferenceChecks(gc, op, [X, tiles, axis],
|
|
tile_ref)
|
|
# Check over multiple devices
|
|
self.assertDeviceChecks(dc, op, [X], [0])
|
|
# Gradient check wrt X
|
|
self.assertGradientChecks(gc, op, [X], 0, [0])
|
|
|
|
@unittest.skipIf(not workspace.has_gpu_support, "No gpu support")
|
|
@given(M=st.integers(min_value=1, max_value=200),
|
|
N=st.integers(min_value=1, max_value=200),
|
|
tiles=st.integers(min_value=50, max_value=100),
|
|
**hu.gcs)
|
|
def test_tile_grad(self, M, N, tiles, gc, dc):
|
|
X = np.random.rand(M, N).astype(np.float32)
|
|
axis = 1
|
|
|
|
op = core.CreateOperator(
|
|
'Tile', ['X'], 'out',
|
|
tiles=tiles,
|
|
axis=axis,
|
|
)
|
|
|
|
def tile_ref(X, tiles, axis):
|
|
dims = np.asarray([1, 1], dtype=int)
|
|
dims[axis] = tiles
|
|
tiled_data = np.tile(X, dims)
|
|
return (tiled_data,)
|
|
|
|
# Check against numpy reference
|
|
self.assertReferenceChecks(gc, op, [X, tiles, axis],
|
|
tile_ref)
|
|
# Check over multiple devices
|
|
self.assertDeviceChecks(dc, op, [X], [0])
|
|
|
|
# Gradient check wrt X
|
|
grad_op = core.CreateOperator(
|
|
'TileGradient', ['dOut'], 'dX',
|
|
tiles=tiles,
|
|
axis=axis,
|
|
)
|
|
dX = np.random.rand(M, N * tiles).astype(np.float32)
|
|
self.assertDeviceChecks(dc, grad_op, [dX], [0])
|
|
|
|
@given(M=st.integers(min_value=1, max_value=10),
|
|
K=st.integers(min_value=1, max_value=10),
|
|
N=st.integers(min_value=1, max_value=10),
|
|
tiles=st.integers(min_value=1, max_value=3),
|
|
axis=st.integers(min_value=0, max_value=2),
|
|
**hu.gcs)
|
|
@settings(deadline=10000)
|
|
def test_tilewinput(self, M, K, N, tiles, axis, gc, dc):
|
|
X = np.random.rand(M, K, N).astype(np.float32)
|
|
|
|
tiles_arg = np.array([tiles], dtype=np.int32)
|
|
axis_arg = np.array([axis], dtype=np.int32)
|
|
|
|
op = core.CreateOperator(
|
|
'Tile', ['X', 'tiles', 'axis'], 'out',
|
|
)
|
|
|
|
def tile_ref(X, tiles, axis):
|
|
dims = np.asarray([1, 1, 1], dtype=int)
|
|
dims[axis] = tiles
|
|
tiled_data = np.tile(X, dims)
|
|
return (tiled_data,)
|
|
|
|
# Check against numpy reference
|
|
self.assertReferenceChecks(gc, op, [X, tiles_arg, axis_arg],
|
|
tile_ref)
|
|
# Check over multiple devices
|
|
self.assertDeviceChecks(dc, op, [X, tiles_arg, axis_arg], [0])
|
|
# Gradient check wrt X
|
|
self.assertGradientChecks(gc, op, [X, tiles_arg, axis_arg], 0, [0])
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|