def test_struct_initializers(self):
obj = ([4], CenteredGrid([1, 4, 1], box[0:1], content_type=struct.shape), ([9], [8, 2]))
z = math.zeros(obj)
self.assertIsInstance(z, tuple)
np.testing.assert_equal(z[0], np.zeros([4]))
z2 = math.zeros_like(z)
np.testing.assert_equal(math.shape(z)[0], math.shape(z2)[0])
def test_zeros_nonuniform(self):
nonuniform = shape_stack('stack', BATCH_DIM, shape(time=1, x=3, y=3),
shape(x=3, y=4), shape())
self.assertEqual(math.zeros(nonuniform).shape, nonuniform)
self.assertEqual(math.ones(nonuniform).shape, nonuniform)
self.assertEqual(math.random_normal(nonuniform).shape, nonuniform)
self.assertEqual(math.random_uniform(nonuniform).shape, nonuniform)
def test_struct_initializers(self):
bounds = box[0:1] # outside unsafe
with struct.unsafe():
obj = ([4], CenteredGrid([1, 4, 1], bounds), ([9], [8, 2]))
z = math.zeros(obj)
self.assertIsInstance(z, tuple)
np.testing.assert_equal(z[0], np.zeros([4]))
z2 = math.zeros_like(z)
np.testing.assert_equal(math.shape(z)[0], math.shape(z2)[0])
def test_tensor_from_tuple_of_tensor_like(self):
native = ((1, 2, 3), math.zeros(vector=3))
for backend in (NUMPY_BACKEND, TORCH_BACKEND, TF_BACKEND):
with backend:
tens = math.tensor(native, names=['stack', 'vector'], convert=False)
self.assertEqual(math.NUMPY_BACKEND, math.choose_backend(tens))
self.assertEqual(shape(stack=2, vector=3), tens.shape)
tens = math.tensor(native, names=['stack', 'vector'])
self.assertEqual(backend, math.choose_backend(tens))
self.assertEqual(shape(stack=2, vector=3), tens.shape)
def test_stack(self):
stacked = shape_stack('stack', BATCH_DIM, shape(time=1, x=3, y=3),
shape(x=3, y=4), shape())
print(stacked)
self.assertEqual(('stack', 'time', 'x', 'y'), stacked.names)
self.assertEqual(3, stacked.stack)
self.assertEqual(1, stacked.time)
math.assert_close((3, 3, 1), stacked.x)
math.assert_close((3, 4, 1), stacked.y)
print(stacked.shape)
self.assertEqual(('stack', 'dims'), stacked.shape.names)
self.assertEqual(12, stacked.shape.volume)
def test_trace_function(self):
def f(x: math.Tensor, y: math.Tensor):
return x + y
for backend in BACKENDS:
with backend:
ft = math.jit_compile(f)
args1 = math.ones(x=2), math.ones(y=2)
args2 = math.ones(x=3), math.ones(y=3)
res1 = ft(*args1)
self.assertEqual(math.shape(x=2, y=2), res1.shape)
math.assert_close(res1, 2)
res2 = ft(*args2)
self.assertEqual(math.shape(x=3, y=3), res2.shape)
math.assert_close(res2, 2)
def _grid_sample(self, box, resolution):
"""
Samples this field on a regular grid.
:param box: physical dimensions of the grid
:param resolution: grid resolution
:return: CenteredGrid
"""
sample_indices_nd = math.to_int(
math.round(box.global_to_local(self.sample_points) * resolution))
sample_indices_nd = math.minimum(
math.maximum(0, sample_indices_nd), resolution - 1
) # Snap outside points to edges, otherwise scatter raises an error
# Correct format for math.scatter
valid_indices = _batch_indices(sample_indices_nd)
shape = (math.shape(
self.data)[0], ) + tuple(resolution) + (self.data.shape[-1], )
scattered = math.scatter(self.sample_points,
valid_indices,
self.data,
shape,
duplicates_handling=self.mode)
return CenteredGrid(data=scattered,
box=box,
extrapolation='constant',
name=self.name + '_centered')
def at(self, other_field):
if isinstance(other_field, CenteredGrid):
batch_size = other_field._batch_size
if batch_size is None:
if other_field.content_type in (struct.shape,
struct.staticshape):
batch_size = other_field.data[0]
else:
batch_size = math.shape(other_field.data)[0]
array = self.grid_sample(other_field.resolution,
other_field.box.size,
batch_size=batch_size)
return other_field.with_data(array)
if isinstance(other_field, StaggeredGrid):
assert self.channels is None or self.channels == other_field.rank
return other_field.with_data([
self.grid_sample(grid.resolution, grid.box.size,
grid._batch_size, 1)
for grid in other_field.unstack()
])
if isinstance(other_field, Domain):
array = self.grid_sample(other_field.resolution,
other_field.box.size)
return CenteredGrid(array,
box=other_field.box,
extrapolation='boundary')
def sparse_cg(field, A, max_iterations, guess, accuracy, back_prop=False):
div_vec = math.reshape(field, [-1, int(np.prod(field.shape[1:]))])
if guess is not None:
guess = math.reshape(guess, [-1, int(np.prod(field.shape[1:]))])
apply_A = lambda pressure: math.matmul(A, pressure)
result_vec, iterations = conjugate_gradient(div_vec, apply_A, guess, accuracy, max_iterations, back_prop)
return math.reshape(result_vec, math.shape(field)), iterations
def distribute_points(density, particles_per_cell=1, distribution='uniform'):
"""
Distribute points according to the distribution specified in density.
:param density: binary tensor
:param particles_per_cell: integer
:param distribution: 'uniform' or 'center'
:return: tensor of shape (batch_size, point_count, rank)
"""
assert distribution in ('center', 'uniform')
index_array = []
batch_size = math.staticshape(density)[0] if math.staticshape(density)[0] is not None else 1
for batch in range(batch_size):
indices = math.where(density[batch, ..., 0] > 0)
indices = math.to_float(indices)
temp = []
for _ in range(particles_per_cell):
if distribution == 'center':
temp.append(indices + 0.5)
elif distribution == 'uniform':
temp.append(indices + math.random_uniform(math.shape(indices)))
index_array.append(math.concat(temp, axis=0))
try:
index_array = math.stack(index_array)
return index_array
except ValueError:
raise ValueError("all arrays in the batch must have the same number of active cells.")
def solve(self, field, domain, guess, enable_backprop):
assert isinstance(domain, FluidDomain)
active_mask = domain.active_tensor(extend=1)
fluid_mask = domain.accessible_tensor(extend=1)
dimensions = math.staticshape(field)[1:-1]
N = int(np.prod(dimensions))
periodic = Material.periodic(domain.domain.boundaries)
if math.choose_backend([field, active_mask,
fluid_mask]).matches_name('SciPy'):
A = sparse_pressure_matrix(dimensions, active_mask, fluid_mask,
periodic)
else:
sidx, sorting = sparse_indices(dimensions, periodic)
sval_data = sparse_values(dimensions, active_mask, fluid_mask,
sorting, periodic)
backend = math.choose_backend(field)
sval_data = backend.cast(sval_data, field.dtype)
A = backend.sparse_tensor(indices=sidx,
values=sval_data,
shape=[N, N])
div_vec = math.reshape(field, [-1, int(np.prod(field.shape[1:]))])
if guess is not None:
guess = math.reshape(guess, [-1, int(np.prod(field.shape[1:]))])
def apply_A(pressure):
return math.matmul(A, pressure)
result_vec, iterations = conjugate_gradient(div_vec, apply_A, guess,
self.accuracy,
self.max_iterations,
enable_backprop)
return math.reshape(result_vec, math.shape(field)), iterations
def test_trace_function(self):
def f(x: math.Tensor, y: math.Tensor):
return x + y
for backend in [
math.NUMPY_BACKEND, tf.TF_BACKEND, torch.TORCH_BACKEND
]:
with backend:
ft = math.trace_function(f)
args1 = math.ones(x=2), math.ones(y=2)
args2 = math.ones(x=3), math.ones(y=3)
res1 = ft(*args1)
self.assertEqual(math.shape(x=2, y=2), res1.shape)
math.assert_close(res1, 2)
res2 = ft(*args2)
self.assertEqual(math.shape(x=3, y=3), res2.shape)
math.assert_close(res2, 2)
def __init__(self,
center,
unit_distance,
maximum_value=1.0,
data=1.0,
name='harmonic',
**kwargs):
rank = math.shape(center)[-1]
AnalyticField.__init__(self, **struct.kwargs(locals()))
def l1_loss(tensor, batch_norm=True, reduce_batches=True):
if isinstance(tensor, StaggeredGrid):
tensor = tensor.staggered
if reduce_batches:
total_loss = math.sum(math.abs(tensor))
else:
total_loss = math.sum(math.abs(tensor),
axis=list(range(1, len(tensor.shape))))
if batch_norm and reduce_batches:
batch_size = math.shape(tensor)[0]
return total_loss / math.to_float(batch_size)
else:
return total_loss
def l_n_loss(tensor, n, batch_norm=True, reduce_batches=True):
if isinstance(tensor, StaggeredGrid):
tensor = tensor.staggered
if reduce_batches:
total_loss = math.sum(tensor**n) / n
else:
total_loss = math.sum(tensor**n,
axis=list(range(1, len(tensor.shape)))) / n
if batch_norm:
batch_size = math.shape(tensor)[0]
return total_loss / math.to_float(batch_size)
else:
return total_loss
def __init__(self, shape=math.EMPTY_SHAPE, scale=10, smoothness=1.0, **dims):
"""
Generates random noise fluctuations which can be configured in physical size and smoothness.
Each time values are sampled from a Noise field, a new noise field is generated.
Noise is typically used as an initializer for CenteredGrids or StaggeredGrids.
Args:
channels: Number of independent random scalar fields this Field consists of
scale: Size of noise fluctuations in physical units
smoothness: Determines how quickly high frequencies die out
"""
self.scale = scale
self.smoothness = smoothness
self._shape = shape & math.shape(**dims)
def batch_indices(indices):
"""
Reshapes the indices such that, aside from indices, they also contain batch number.
For example the entry (32, 40) as coordinates of batch 2 will become (2, 32, 40).
Transform shape (b, p, d) to (b, p, d+1) where batch size is b, number of particles is p and number of dimensions is d.
"""
batch_size = indices.shape[0]
out_spatial_rank = len(indices.shape) - 2
out_spatial_size = math.shape(indices)[1:-1]
batch_range = math.DYNAMIC_BACKEND.choose_backend(indices).range(batch_size)
batch_ids = math.reshape(batch_range, [batch_size] + [1] * out_spatial_rank)
tile_shape = math.pad(out_spatial_size, [[1,0]], constant_values=1)
batch_ids = math.expand_dims(math.tile(batch_ids, tile_shape), axis=-1)
return math.concat((batch_ids, indices), axis=-1)
def create(parent_directory: str,
shape: math.Shape = math.EMPTY_SHAPE,
copy_calling_script=True,
**dimensions) -> 'Scene':
"""
Creates a new `Scene` or a batch of new scenes inside `parent_directory`.
See Also:
`Scene.at()`, `Scene.list()`.
Args:
parent_directory: Directory to hold the new `Scene`. If it doesn't exist, it will be created.
shape: Determines number of scenes to create. Multiple scenes will be represented by a `Scene` with `is_batch=True`.
copy_calling_script: Whether to copy the Python file that invoked this method into the `src` folder of all created scenes.
See `Scene.copy_calling_script()`.
dimensions: Additional batch dimensions
Returns:
Single `Scene` object representing the new scene(s).
"""
shape = (shape & math.shape(**dimensions)).to_batch()
parent_directory = expanduser(parent_directory)
abs_dir = abspath(parent_directory)
if not isdir(abs_dir):
os.makedirs(abs_dir)
next_id = 0
else:
indices = [
int(name[4:]) for name in os.listdir(abs_dir)
if name.startswith("sim_")
]
next_id = max([-1] + indices) + 1
ids = math.wrap(tuple(range(next_id, next_id +
shape.volume))).vector.split(shape)
paths = math.map(lambda id_: join(parent_directory, f"sim_{id_:06d}"),
ids)
scene = Scene(paths)
scene.mkdir()
if copy_calling_script:
try:
scene.copy_calling_script()
except IOError as err:
warnings.warn(
f"Failed to copy calling script to scene during Scene.create(): {err}"
)
return scene
def test_box_batched(self):
box = Box(math.tensor([(0, 0), (1, 1)], 'boxes,vector'), 1)
self.assertEqual(math.shape(boxes=2, x=1, y=1), box.shape)
def component_count(self):
if math.ndims(self.data) == 0:
return 1
return math.shape(self.data)[-1]
def test_combine(self):
self.assertEqual(shape(batch=2, x=3, y=4),
shape(batch=2) & shape(x=3, y=4))
self.assertEqual(shape(x=3, vector=2), shape(vector=2) & shape(x=3))
self.assertEqual(shape(batch=10, x=3, vector=2),
shape(vector=2) & shape(x=3) & shape(batch=10))
def test_downsample_staggered_2d(self):
grid = Domain(x=32, y=40).staggered_grid(1)
downsampled = field.downsample2x(grid)
self.assertEqual(
math.shape(x=16, y=20, vector=2).alphabetically(),
downsampled.shape.alphabetically())
def wave_vector(self, wave_vector):
if len(math.shape(wave_vector)) == 0:
wave_vector = math.expand_dims(wave_vector, 0)
return wave_vector
def test_sample_at(self):
DOMAIN = Domain(x=4, y=3)
field = AngularVelocity([0, 0])
self.assertEqual(math.shape(vector=2), field.shape.channel)
field >> DOMAIN.vector_grid()
field >> DOMAIN.staggered_grid()
def test_box_constructor(self):
box = Box(0, (1, 1))
math.assert_close(box.size, 1)
self.assertEqual(math.shape(x=1, y=1), box.shape)
def test_subshapes(self):
s = shape(batch=10, x=4, y=3, vector=2)
self.assertEqual(shape(batch=10), s.batch)
self.assertEqual(shape(x=4, y=3), s.spatial)
self.assertEqual(shape(vector=2), s.channel)
def test_indexing(self):
s = shape(batch=10, x=4, y=3, vector=2)
self.assertEqual(shape(batch=10), s[0:1])
self.assertEqual(shape(batch=10), s[[0]])
self.assertEqual(shape(x=4, y=3), s[1:3])
def test_custom_spatial_dims(self):
domain = Domain(a=4, b=3)
grid = domain.scalar_grid(1)
self.assertEqual(math.shape(a=4, b=3), grid.shape)
grid = domain.staggered_grid(1)
self.assertEqual(math.shape(a=4, b=3, vector=2), grid.shape)
def test_after_gather(self):
self.assertEqual(shape(x=2),
shape(x=3).after_gather({'x': slice(None, None, 2)}))
def approximate_fraction_inside(self, location, cell_size):
return math.tile(math.to_float(0), list(math.shape(location)[:-1]) + [1])
