def visit(self, op, *args):
self.computation.set_op_rank(op)
# op.args[0] : delta
# op.args[1] : data batch
# op.args[2] (optional) : dbias
# op.fprop.args[0] : data batch
# op.fprop.args[1] : filters
# op.fprop.conv_params : forward params
delta = args[0]
data = args[1]
filters = op.fprop.args[1]
conv_params = op.fprop.conv_params
"""
print(delta.axes)
print(filters.axes)
print(data.axes)
print(conv_params)
"""
ngraph_update_conv = PyngConvolutionBackpropFilters(
self.computation.lookup_cpp_op(data),
Shape(list(filters.axes.lengths)),
self.computation.lookup_cpp_op(delta),
Strides([conv_params['str_h'], conv_params['str_w']]),
Strides([conv_params['dil_h'], conv_params['dil_w']]),
CoordinateDiff([conv_params['pad_h'], conv_params['pad_w']]),
CoordinateDiff([conv_params['pad_h'], conv_params['pad_w']]),
Strides([1, 1]))
ngraph_update_conv.name = op.name.replace(
'/', '_') + "_ConvolutionBackpropFilters"
self.computation.register_cpp_op(op,
ngraph_update_conv,
set_name=False)
def visit(self, op, *args):
self.computation.set_op_rank(op)
# op.args[0] : inputs
# op.args[1] : filters
# op.args[2] (optional): bias
# op.conv_params
# op.channel_axes
# op.spatial_axes
if len(args) == 2:
inputs = args[0]
filters = args[1]
else:
raise RuntimeError("Not Implemented: Convolution with bias")
"""
{'K': 16, 'T': 1, 'R': 5, 'S': 5, 'str_d': 1, 'pad_d': 0, 'dil_d': 1,
'str_h': 1, 'pad_h': 0, 'dil_h': 1, 'str_w': 1, 'pad_w': 0, 'dil_w': 1}
"""
"""
print(inputs.axes)
print(op.axes)
print(filters.axes)
"""
# print(op_element_type.get_output_shape(0))
ngraph_conv = PyngConvolution(
self.computation.lookup_cpp_op(inputs),
self.computation.lookup_cpp_op(filters),
Strides([op.conv_params['str_h'], op.conv_params['str_w']]),
Strides([op.conv_params['dil_h'], op.conv_params['dil_w']]),
CoordinateDiff([op.conv_params['pad_h'], op.conv_params['pad_w']]),
CoordinateDiff([op.conv_params['pad_h'], op.conv_params['pad_w']]),
Strides([1, 1]))
ngraph_conv.name = op.name.replace('/', '_') + "_Convolution"
self.computation.register_cpp_op(op, ngraph_conv, set_name=False)
def convolution(
x, # type: Node
weights, # type: Node
strides=None, # type: List[int]
dilation=None, # type: List[int]
padding_above=None, # type: List[int]
padding_below=None, # type: List[int]
name=None, # type: str
):
# type: (...) -> Node
"""Return convolution node."""
if strides is None:
strides = [1] * (
len(x.shape) - 2
) # Default to as many 1s as spatial dimensions of input.
if dilation is None:
dilation = [1] * (len(x.shape) - 2)
if padding_above is None:
padding_above = [0] * (len(x.shape) - 2)
if padding_below is None:
padding_below = [0] * (len(x.shape) - 2)
return Convolution(x, weights, Strides(strides), Strides(dilation),
CoordinateDiff(padding_above),
CoordinateDiff(padding_below))
def test_convolutionBackpropData():
element_type = Type.f32
image_shape = Shape([1, 1, 10, 10])
filter_shape = Shape([1, 1, 3, 3])
output_shape = Shape([1, 1, 17, 17])
image_arr = np.arange(100, dtype=np.float32).reshape(1, 1, 10, 10)
filter_arr = np.ones(9, dtype=np.float32).reshape(1, 1, 3, 3)
window_strides = [1, 1]
window_dilation = [1, 1]
padding_below = [0, 0]
padding_above = [0, 0]
data_dilation = [2, 2]
output_arr = convolution2d(image_arr[0][0], filter_arr[0][0],
window_strides, window_dilation, padding_below,
padding_above,
data_dilation).reshape(1, 1, 17, 17)
A = Parameter(element_type, filter_shape)
B = Parameter(element_type, output_shape)
parameter_list = [A, B]
function = Function(
NodeVector([
ConvolutionBackpropData(image_shape, A, B, Strides(window_strides),
Strides(window_dilation),
CoordinateDiff(padding_below),
CoordinateDiff(padding_above),
Strides(data_dilation))
]), parameter_list, 'test')
backend, cf = make_backend_call_frame(function)
a = backend.make_primary_tensor_view(element_type, filter_shape)
b = backend.make_primary_tensor_view(element_type, output_shape)
a.write(util.numpy_to_c(filter_arr), 0, 3 * 3 * 4)
b.write(util.numpy_to_c(output_arr), 0, 17 * 17 * 4)
result_arr = np.zeros(10 * 10, dtype=np.float32).reshape(1, 1, 10, 10)
result = backend.make_primary_tensor_view(element_type,
Shape([1, 1, 10, 10]))
result.write(util.numpy_to_c(result_arr), 0, 10 * 10 * 4)
cf.call([result], [a, b])
result.read(util.numpy_to_c(result_arr), 0, 10 * 10 * 4)
result_arr_ref = np.array(
[[[[22, 60, 70, 80, 90, 100, 110, 120, 130, 54.],
[105, 275, 300, 325, 350, 375, 400, 425, 450, 185.],
[205, 525, 550, 575, 600, 625, 650, 675, 700, 285.],
[305, 775, 800, 825, 850, 875, 900, 925, 950, 385.],
[405, 1025, 1050, 1075, 1100, 1125, 1150, 1175, 1200, 485.],
[505, 1275, 1300, 1325, 1350, 1375, 1400, 1425, 1450, 585.],
[605, 1525, 1550, 1575, 1600, 1625, 1650, 1675, 1700, 685.],
[705, 1775, 1800, 1825, 1850, 1875, 1900, 1925, 1950, 785.],
[805, 2025, 2050, 2075, 2100, 2125, 2150, 2175, 2200, 885.],
[342, 860, 870, 880, 890, 900, 910, 920, 930, 374.]]]])
assert np.allclose(result_arr, result_arr_ref)
def test_convolutionBackpropFilters():
element_type = Type.f32
image_shape = Shape([1, 1, 10, 10])
filter_shape = Shape([1, 1, 3, 3])
output_shape = Shape([1, 1, 17, 17])
image_arr = np.arange(100, dtype=np.float32).reshape(1, 1, 10, 10)
filter_arr = np.ones(9, dtype=np.float32).reshape(1, 1, 3, 3)
window_strides = [1, 1]
window_dilation = [1, 1]
padding_below = [0, 0]
padding_above = [0, 0]
data_dilation = [2, 2]
output_arr = convolution2d(image_arr[0][0], filter_arr[0][0],
window_strides, window_dilation, padding_below,
padding_above,
data_dilation).reshape(1, 1, 17, 17)
A = Parameter(element_type, image_shape)
B = Parameter(element_type, output_shape)
parameter_list = [A, B]
function = Function(
NodeVector([
ConvolutionBackpropFilters(A, filter_shape, B,
Strides(window_strides),
Strides(window_dilation),
CoordinateDiff(padding_below),
CoordinateDiff(padding_above),
Strides(data_dilation))
]), parameter_list, 'test')
backend, cf = make_backend_call_frame(function)
a = backend.make_primary_tensor_view(element_type, image_shape)
b = backend.make_primary_tensor_view(element_type, output_shape)
a.write(util.numpy_to_c(image_arr), 0, 10 * 10 * 4)
b.write(util.numpy_to_c(output_arr), 0, 17 * 17 * 4)
result_arr = np.zeros(3 * 3, dtype=np.float32).reshape(1, 1, 3, 3)
result = backend.make_primary_tensor_view(element_type, Shape([1, 1, 3,
3]))
result.write(util.numpy_to_c(result_arr), 0, 3 * 3 * 4)
cf.call([result], [a, b])
result.read(util.numpy_to_c(result_arr), 0, 3 * 3 * 4)
result_arr_ref = np.array([[[[923832, 413952, 939870.],
[425832, 190752, 432960.],
[1084212, 485232, 1100250.]]]])
assert np.allclose(result_arr, result_arr_ref)
def test_convolution_with_data_dilation():
element_type = Type.f32
image_shape = Shape([1, 1, 10, 10])
filter_shape = Shape([1, 1, 3, 3])
A = Parameter(element_type, image_shape)
B = Parameter(element_type, filter_shape)
parameter_list = [A, B]
image_arr = np.arange(100, dtype=np.float32).reshape(1, 1, 10, 10)
filter_arr = np.ones(9, dtype=np.float32).reshape(1, 1, 3, 3)
strides = [1, 1]
dilation = [1, 1]
padding_below = [0, 0]
padding_above = [0, 0]
data_dilation = [2, 2]
function = Function(
NodeVector([
Convolution(A, B, Strides(strides), Strides(dilation),
CoordinateDiff(padding_below),
CoordinateDiff(padding_above), Strides(data_dilation))
]), parameter_list, 'test')
backend, cf = make_backend_call_frame(function)
a = backend.make_primary_tensor_view(element_type, image_shape)
b = backend.make_primary_tensor_view(element_type, filter_shape)
a.write(util.numpy_to_c(image_arr), 0, 10 * 10 * 4)
b.write(util.numpy_to_c(filter_arr), 0, 3 * 3 * 4)
result_arr = np.zeros(17 * 17, dtype=np.float32).reshape(1, 1, 17, 17)
result = backend.make_primary_tensor_view(element_type,
Shape([1, 1, 17, 17]))
result.write(util.numpy_to_c(result_arr), 0, 17 * 17 * 4)
cf.call([result], [a, b])
result.read(util.numpy_to_c(result_arr), 0, 17 * 17 * 4)
result_arr_ref = convolution2d(image_arr[0][0], filter_arr[0][0], strides,
dilation, padding_below, padding_above,
data_dilation).reshape(1, 1, 17, 17)
assert np.allclose(result_arr, result_arr_ref)
def convolution(
data_batch, # type: Node
filter_weights, # type: Node
filter_strides=None, # type: List[int]
filter_dilation_strides=None, # type: List[int]
padding_below=None, # type: List[int]
padding_above=None, # type: List[int]
data_dilation_strides=None, # type: List[int]
name=None, # type: str
):
# type: (...) -> Node
"""Return node performing batched convolution operation.
:param data_batch: The node providing data batch tensor.
:param filter_weights: The node providing filters tensor.
:param filter_strides: The kernel window movement strides.
:param filter_dilation_strides: The filters dilation strides.
:param padding_below: The number of zero padding elements to add on each axis below 0
coordinate.
:param padding_above: The number of zero padding elements to add on each axis above max
coordinate.
:param data_dilation_strides: The data batch dilation strides.
:param name: The optional new name for output node.
:return: New node performing batched convolution operation.
"""
spatial_dim_count = len(data_batch.shape) - 2
if filter_strides is None:
filter_strides = [1] * spatial_dim_count
if filter_dilation_strides is None:
filter_dilation_strides = [1] * spatial_dim_count
if padding_above is None:
padding_above = [0] * spatial_dim_count
if padding_below is None:
padding_below = [0] * spatial_dim_count
if data_dilation_strides is None:
data_dilation_strides = [1] * spatial_dim_count
return Convolution(data_batch, filter_weights, Strides(filter_strides),
Strides(filter_dilation_strides),
CoordinateDiff(padding_below),
CoordinateDiff(padding_above),
Strides(data_dilation_strides))
def convolution_backprop_data(
data_batch_shape, # type: TensorShape
filters, # type: Node
output_delta, # type: Node
window_movement_strides_forward=None, # type: List[int]
window_dilation_strides_forward=None, # type: List[int]
padding_below_forward=None, # type: List[int]
padding_above_forward=None, # type: List[int]
data_dilation_strides_forward=None, # type: List[int]
name=None, # type: str
):
# type: (...) -> Node
"""Return node performing a batched-convolution data batch-backprop operation.
:param data_batch_shape: The shape of the data batch from forward-prop.
:param filters: The node producing the filters from forward-prop.
:param output_delta: The node producing output delta.
:param window_movement_strides_forward: The window movement strides from forward-prop.
:param window_dilation_strides_forward: The window dilation strides from forward-prop.
:param padding_below_forward: The padding-below sizes from forward-prop.
:param padding_above_forward: The padding-above sizes from forward-prop.
:param data_dilation_strides_forward: The data dilation strides from forward-prop.
"""
spatial_dim_count = len(data_batch_shape) - 2
if window_movement_strides_forward is None:
window_movement_strides_forward = [1] * spatial_dim_count
if window_dilation_strides_forward is None:
window_dilation_strides_forward = [1] * spatial_dim_count
if padding_below_forward is None:
padding_below_forward = [0] * spatial_dim_count
if padding_above_forward is None:
padding_above_forward = [0] * spatial_dim_count
if data_dilation_strides_forward is None:
data_dilation_strides_forward = [1] * spatial_dim_count
return ConvolutionBackpropData(Shape(data_batch_shape), filters,
output_delta,
Strides(window_movement_strides_forward),
Strides(window_dilation_strides_forward),
CoordinateDiff(padding_below_forward),
CoordinateDiff(padding_above_forward),
Strides(data_dilation_strides_forward))
def test_convolution_with_padding():
element_type = Type.f32
image_shape = Shape([1, 1, 10, 10])
filter_shape = Shape([1, 1, 3, 3])
A = Parameter(element_type, image_shape)
B = Parameter(element_type, filter_shape)
parameter_list = [A, B]
image_arr = np.arange(100, dtype=np.float32).reshape(1, 1, 10, 10)
filter_arr = np.zeros(9, dtype=np.float32).reshape(1, 1, 3, 3)
filter_arr[0][0][1][1] = 1
strides = [1, 1]
dilation = [2, 2]
padding_below = [0, 0]
padding_above = [0, 0]
function = Function([
Convolution(A, B, Strides(strides), Strides(dilation),
CoordinateDiff(padding_below),
CoordinateDiff(padding_above))
], parameter_list, 'test')
backend = Backend.create(test.BACKEND_NAME)
a = backend.create_tensor(element_type, image_shape)
b = backend.create_tensor(element_type, filter_shape)
a.write(util.numpy_to_c(image_arr), 10 * 10 * 4)
b.write(util.numpy_to_c(filter_arr), 3 * 3 * 4)
result_arr = np.zeros(36, dtype=np.float32).reshape(1, 1, 6, 6)
result = backend.create_tensor(element_type, Shape([1, 1, 6, 6]))
result.write(util.numpy_to_c(result_arr), 6 * 6 * 4)
handle = backend.compile(function)
handle.call([result], [a, b])
result.read(util.numpy_to_c(result_arr), 6 * 6 * 4)
result_arr_ref = convolution2d(image_arr[0][0], filter_arr[0][0], strides,
dilation, padding_below,
padding_above).reshape(1, 1, 6, 6)
assert np.allclose(result_arr, result_arr_ref)
# test with non-zero padding
element_type = Type.f32
image_shape = Shape([1, 1, 10, 10])
filter_shape = Shape([1, 1, 3, 3])
A = Parameter(element_type, image_shape)
B = Parameter(element_type, filter_shape)
parameter_list = [A, B]
image_arr = np.arange(100, dtype=np.float32).reshape(1, 1, 10, 10)
filter_arr = (np.ones(9, dtype=np.float32).reshape(1, 1, 3, 3)) * -1
filter_arr[0][0][1][1] = 1
strides = [1, 1]
dilation = [2, 2]
padding_below = [2, 1]
padding_above = [1, 2]
function = Function([
Convolution(A, B, Strides(strides), Strides(dilation),
CoordinateDiff(padding_below),
CoordinateDiff(padding_above))
], parameter_list, 'test')
backend = Backend.create(test.BACKEND_NAME)
a = backend.create_tensor(element_type, image_shape)
b = backend.create_tensor(element_type, filter_shape)
a.write(util.numpy_to_c(image_arr), 10 * 10 * 4)
b.write(util.numpy_to_c(filter_arr), 3 * 3 * 4)
result_arr = np.zeros(81, dtype=np.float32).reshape(1, 1, 9, 9)
result = backend.create_tensor(element_type, Shape([1, 1, 9, 9]))
result.write(util.numpy_to_c(result_arr), 9 * 9 * 4)
handle = backend.compile(function)
handle.call([result], [a, b])
result.read(util.numpy_to_c(result_arr), 9 * 9 * 4)
result_arr_ref = convolution2d(image_arr[0][0], filter_arr[0][0], strides,
dilation, padding_below,
padding_above).reshape(1, 1, 9, 9)
assert np.allclose(result_arr, result_arr_ref)
