def test_tile_infer_correct_2d_tensor(self):
graph = build_graph(nodes_attributes, edges,
{'data': {'shape': np.array([3, 7])},
'tile_values': {'value': np.array([5, 1])}})
tile_node = Node(graph, 'tile')
Tile.infer(tile_node)
self.assertTrue(np.all(np.array([15, 7]) == graph.node['tile_out']['shape']))
def replace_pattern(graph: Graph, match: dict):
tile = match['tile']
if tile.has_valid('tile_array'):
tile_array = tile.tile_array
assert len(tile_array) == len(tile.in_port(0).data.get_shape())
non_one_tile = np.argwhere(tile_array != 1).flatten()
# We need to add new tiles only in case when we tile more than one dimension
if len(non_one_tile) > 1:
last_tile = None
for i in non_one_tile:
axis = i
tiles = tile_array[i]
new_tile = Tile(graph, {'name': tile.name + '/Tile_{}/'.format(i), 'axis': axis, 'tiles': tiles,
'need_shape_inference': True}).create_node()
if not last_tile:
last_tile = new_tile
tile.in_port(0).get_connection().set_destination(new_tile.in_port(0))
else:
last_tile.out_port(0).connect(new_tile.in_port(0))
last_tile = new_tile
# Reconnect output to new tile node and delete old tile
tile.out_port(0).get_connection().set_source(last_tile.out_port(0))
def test_tile_infer_one_input_correct_missing_tiles(self):
graph = build_graph(nodes_attributes, [('data', 'tile'),
('tile', 'tile_out')],
{'tile': {
'axis': 1
}})
tile_node = Node(graph, 'tile')
Tile.infer(tile_node)
self.assertIsNone(graph.node['tile_out']['shape'])
def test_tile_infer_values_test(self):
input_data = np.arange(-30, 60, 0.25).reshape([2, 4, 3, -1])
tile_values = np.array([3, 1, 1, 1])
graph = build_graph(nodes_attributes, edges,
{'data': {'shape': np.array(input_data.shape), 'value': input_data},
'tile_values': {'value': tile_values}})
tile_node = Node(graph, 'tile')
Tile.infer(tile_node)
self.assertTrue(np.all(np.tile(input_data, tile_values) == graph.node['tile_out']['value']))
def test_tile_infer_correct(self):
graph = build_graph(nodes_attributes, edges,
{'tile_values': {
'value': np.array([7, 1, 1, 1])
}})
tile_node = Node(graph, 'tile')
Tile.infer(tile_node)
self.assertTrue(
np.all(
np.array([70, 20, 30, 40]) == graph.node['tile_out']['shape']))
def test_tile_infer_undefined_tile_values(self):
graph = build_graph(nodes_attributes, [('data', 'tile'),
('tile_values', 'tile'),
('tile', 'tile_out')],
{'tile_values': {
'value': None
}})
tile_node = Node(graph, 'tile')
Tile.infer(tile_node)
self.assertIsNone(graph.node['tile_out']['shape'])
def test_tile_infer_three_non_one(self):
graph = build_graph(nodes_attributes, edges,
{'tile_values': {
'value': np.array([2, 1, 5, 2])
}})
tile_node = Node(graph, 'tile')
Tile.infer(tile_node)
self.assertTrue(
np.all(
np.array([20, 20, 150, 80]) == graph.node['tile_out']
['shape']))
def test_tile_infer_values_const_propagation(self):
"""
Test for constant propagation even if tile with multiple tile indices is not supported
"""
input_data = np.arange(-30, 60, 0.25).reshape([2, 4, 3, -1])
tile_values = np.array([4, 3, 2, 5])
graph = build_graph(nodes_attributes, edges,
{'data': {'shape': np.array(input_data.shape), 'value': input_data},
'tile_values': {'value': tile_values}})
tile_node = Node(graph, 'tile')
Tile.infer(tile_node)
self.assertTrue(np.all(np.tile(input_data, tile_values) == graph.node['tile_out']['value']))
def test_tile_infer_shapes_mismatch(self):
graph = build_graph(nodes_attributes, [('data', 'tile'),
('tile_values', 'tile'),
('tile', 'tile_out')],
{
'tile_values': {
'value': np.array([1, 2, 1]),
'shape': np.array([3])
}
})
tile_node = Node(graph, 'tile')
Tile.infer(tile_node)
self.assertIsNone(graph.node['tile_out']['shape'])
def test_tile_infer_shapes_alignment(self):
graph = build_graph(nodes_attributes, edges, {
'tile_values': {
'value': np.array([1, 2, 3]),
'shape': np.array([3])
}
})
tile_node = Node(graph, 'tile')
Tile.infer(tile_node)
self.assertTrue(
np.all(
np.array([10, 20, 60, 120]) == graph.node['tile_out']
['shape']))
def test_tile_infer_none_input_shape(self):
graph = build_graph(
nodes_attributes, [('data', 'tile'), ('tile_values', 'tile'),
('tile', 'tile_out')], {
'data': {
'shape': None
},
'tile_values': {
'value': np.array([1, 7, 1, 1])
}
})
tile_node = Node(graph, 'tile')
Tile.infer(tile_node)
self.assertIsNone(graph.node['tile_out']['shape'])
def test_tile_infer_all_ones(self):
graph = build_graph(nodes_attributes, [('data', 'tile'),
('tile_values', 'tile'),
('tile', 'tile_out')],
{'tile_values': {
'value': np.array([1, 1, 1, 1])
}})
tile_node = Node(graph, 'tile')
Tile.infer(tile_node)
self.assertTrue(
np.all(
np.array([10, 20, 30, 40]) == graph.node['tile_out']['shape']))
self.assertEqual(tile_node.axis, 0)
self.assertEqual(tile_node.tiles, 1)
def test_tile_infer_two_non_one(self):
graph = build_graph(nodes_attributes, [('data', 'tile'),
('tile_values', 'tile'),
('tile', 'tile_out')],
{'tile_values': {
'value': np.array([2, 1, 1, 2])
}})
tile_node = Node(graph, 'tile')
Tile.infer(tile_node)
self.assertIsNone(graph.node['tile']['type'])
self.assertTrue(
np.all(
np.array([20, 20, 30, 80]) == graph.node['tile_out']['shape']))
self.assertFalse(tile_node.has_and_set('axis'))
self.assertFalse(tile_node.has_and_set('tiles'))
def test_tile_infer_one_input_correct(self):
graph = build_graph(nodes_attributes, [('data', 'tile'),
('tile', 'tile_out')],
{'tile': {
'axis': 1,
'tiles': 7
}})
tile_node = Node(graph, 'tile')
Tile.infer(tile_node)
self.assertTrue(
np.all(
np.array([10, 140, 30, 40]) == graph.node['tile_out']
['shape']))
self.assertEqual(tile_node.axis, 1)
self.assertEqual(tile_node.tiles, 7)
def replace_pattern(graph: Graph, match: dict):
node = match['op']
shapes = [in_node.shape for _, in_node in node.in_nodes().items()]
out_shape = node.out_node().shape
tname = node.name + '/Broadcast/'
tile = Tile(graph, dict(name=tname))
# Working with scalar values
for i, shape in enumerate(shapes):
if len(shape) == 0:
shapes[i] = np.ones(len(out_shape), dtype=np.int64)
node.in_node(i).shape = shapes[i].copy()
if node.in_node(i).value is not None:
node.in_node(i).value = np.reshape(node.in_node(i).value, newshape=shapes[i])
if not all([len(shape) == len(out_shape) for shape in shapes]):
log.warning("Cannot apply broadcast for Eltwise layer {} "
"because not all input shapes {} have the same number of elements "
"as output shape {}.".format(node.soft_get('name'),
shapes,
out_shape
)
)
return
input_idx = 0
for port, old_input in node.in_nodes().items():
# old_input = node.in_node(input_idx)
input = old_input
for i in range(len(out_shape)):
if shapes[input_idx][i] == 1 and out_shape[i] > 1:
new_op = tile.create_node([input], dict(axis=i, tiles=out_shape[i]))
# add a data node following a new operation node
data_id = graph.unique_id(node.name)
graph.add_node(data_id, kind='data', shape=None, value=None)
new_data = Node(graph, data_id)
graph.add_edge(new_op.id, new_data.id, **{'out': 0})
new_op.infer(new_op)
input = new_data
if input != old_input:
# create a new edge from new data node after Tile application to the eltwise
# and copy all edge attributes from the old edge
# [0] is not what we really want
graph.add_edge(input.id, node.id, **graph[old_input.id][node.id][0])
graph.remove_edge(old_input.id, node.id)
input_idx += 1
def replace_pattern(self, graph: Graph, match: dict):
node = match['tile']
name = node.soft_get('name', node.id)
axis = node.axis
tiles = node.tiles
input_shape = node.in_port(0).data.get_shape()
assert input_shape is not None
tiles_input_value = int64_array(np.ones(input_shape.size))
tiles_input_value[axis] = tiles
const = Const(graph, {
'value': tiles_input_value,
'name': name + '/tiles'
}).create_node()
tile = Tile(graph, {'name': name}).create_node()
node.out_port(0).get_connection().set_source(tile.out_port(0))
node.in_port(0).get_connection().set_destination(tile.in_port(0))
const.out_port(0).connect(tile.in_port(1))
def extract(node):
Tile.update_node_stat(node, {})
return __class__.enabled
def mxrepeat_decomposition(node: Node):
graph = node.graph
name = node.soft_get('name', node.id)
rename_node(node, name + '/to_be_removed')
# Unqueeze
input_rank = Rank(graph, {'name': name + '/Rank'}).create_node()
node.in_port(0).get_source().connect(input_rank.in_port(0))
axis = get_canonical_axis_index_node(input_rank, node.axis)
unsqueeze_axis = create_op_node_with_second_input(
graph,
Add,
int64_array([1]), {'name': name + '/Unsqueeze/Axis'},
input_node=axis)
unsqueeze = Unsqueeze(graph, {
'name': name + '/Unsqueeze'
}).create_node()
unsqueeze.in_port(1).connect(unsqueeze_axis.out_port(0))
# Tile (1, 1, ..., repeats, ..., 1)
# we generate tile array according to the following table:
# parts: | first | repeats | second |
# i: | 0, 1, ..., axis,| axis + 1,| ..., rank+1 |
# tile_array: | 1, 1, ..., 1 ,| repeats ,| ..., 1 |
one = Const(graph, {
'name': name + '/Broadcast/One',
'value': int64_array([1])
}).create_node()
first_ones = Broadcast(graph, {
'name': name + '/Broadcast/Ones_first_part'
}).create_node()
first_ones.in_port(0).connect(one.out_port(0))
first_ones.in_port(1).connect(unsqueeze_axis.out_port(0))
repeats = Const(graph, {
'name': name + '/repeats',
'value': int64_array([node.repeats])
}).create_node()
second_ones = Broadcast(graph, {
'name': name + '/Broadcast/Ones_second_part'
}).create_node()
second_part_broadcast_shape = Sub(
graph, {
'name': name + '/Broadcast/Shape/second_part'
}).create_node()
second_part_broadcast_shape.in_port(0).connect(input_rank.out_port(0))
second_part_broadcast_shape.in_port(1).connect(
unsqueeze_axis.out_port(0))
second_ones.in_port(0).connect(one.out_port(0))
second_ones.in_port(1).connect(second_part_broadcast_shape.out_port(0))
tile_repeats = new_shape_node_from_shape_nodes(
[first_ones, repeats, second_ones])
tile = Tile(graph, {'name': name + '/Tile'}).create_node()
tile.in_port(1).connect(tile_repeats.out_port(0))
# Reshape (input_shape[:axis], input_shape[axis] * repeats, input_shape[axis+1:])
# we generate reshape dim array according to the following table:
# parts: | first | rep | second |
# i: | 0, 1, ... ,| axis, | ..., rank |
# dim_array: | inp_sh[i] ,| input_shape[axis] * repeats ,| inp_sh[i] |
input_shape = Shape(graph, {'name': name + '/Shape'}).create_node()
node.in_port(0).get_source().connect(input_shape.in_port(0))
first_input_shape_part = get_shape_values_by_range_idxs(
input_shape,
input_rank,
begin=0,
end=node.axis,
include_begin=True,
include_end=False)
original_axis_dim = create_op_with_const_inputs(
graph,
Gather, {2: int64_array(0)}, {'name': name + '/OriginalDim'},
input_node=input_shape)
original_axis_dim.in_port(1).connect(axis.out_port(0))
repeated_dimention = Mul(graph, {
'name': name + '/RepeatedDim'
}).create_node()
repeated_dimention.in_port(0).connect(original_axis_dim.out_port(0))
repeated_dimention.in_port(1).connect(repeats.out_port(0))
second_input_shape_part = get_shape_values_by_range_idxs(
input_shape,
input_rank,
begin=node.axis,
end=-1,
include_begin=False,
include_end=True)
output_shape = new_shape_node_from_shape_nodes([
first_input_shape_part, repeated_dimention, second_input_shape_part
])
reshape = Reshape(graph, {'name': name}).create_node()
rename_node(reshape, name)
reshape.in_port(1).connect(output_shape.out_port(0))
# Final connections
node.in_port(0).get_connection().set_destination(unsqueeze.in_port(0))
tile.in_port(0).connect(unsqueeze.out_port(0))
reshape.in_port(0).connect(tile.out_port(0))
node.out_port(0).get_connection().set_source(reshape.out_port(0))
def extract(cls, node):
Tile.update_node_stat(node, {})
return cls.enabled
def extract(cls, node: Node):
attrs = get_mxnet_layer_attrs(node.symbol_dict)
Tile.update_node_stat(node, {
'reps': attrs.tuple('reps', int, None),
})
return cls.enabled
