Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Expanding mask converter #108

Merged
merged 4 commits into from
Jul 14, 2022
Merged

Expanding mask converter #108

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
d4baf0f
Adding multiple inputs and outputs to mask converter
bernardopacini Jul 11, 2022
4f13875
Merge branch 'main' into expandingMaskConverter
bernardopacini Jul 11, 2022
2f2c14e
Adding test for multiple masks
bernardopacini Jul 12, 2022
af83d1c
Adding warning for overlapping masks when unmasking
bernardopacini Jul 12, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
87 changes: 70 additions & 17 deletions mphys/mask_converter.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
import numpy as np
import openmdao.api as om


Expand Down Expand Up @@ -48,25 +49,48 @@ def setup(self):
mask = self.options['mask']

self.add_input(input.name, shape=input.shape, tags=input.tags, distributed=distributed)
self.add_output(output.name, shape=output.shape, tags=output.tags, val=self.options['init_output'], distributed=distributed)

if isinstance(output, list):
if len(output) != len(mask):
raise ValueError("Output length and mask length not equal")
for i in range(len(output)):
self.add_output(output[i].name, shape=output[i].shape, tags=output[i].tags, val=self.options['init_output'], distributed=distributed)
else:
self.add_output(output.name, shape=output.shape, tags=output.tags, val=self.options['init_output'], distributed=distributed)

def compute(self, inputs, outputs):
input = self.options['input']
output = self.options['output']
mask = self.options['mask']
outputs[output.name] = inputs[input.name][mask]
output = self.options['output']

if isinstance(output, list):
for i in range(len(output)):
outputs[output[i].name] = inputs[input.name][mask[i]]
else:
outputs[output.name] = inputs[input.name][mask]

def compute_jacvec_product(self, inputs, d_inputs, d_outputs, mode):
input = self.options['input']
output = self.options['output']
mask = self.options['mask']
if mode == 'fwd':
if input.name in d_inputs and output.name in d_outputs:
d_outputs[output.name] += d_inputs[input.name][mask]

if mode == 'rev':
if input.name in d_inputs and output.name in d_outputs:
d_inputs[input.name][mask] += d_outputs[output.name]
if isinstance(output, list):
for i in range(len(output)):
if mode == 'fwd':
if input.name in d_inputs and output[i].name in d_outputs:
d_outputs[output[i].name] += d_inputs[input.name][mask[i]]

if mode == 'rev':
if input.name in d_inputs and output[i].name in d_outputs:
d_inputs[input.name][mask[i]] += d_outputs[output[i].name]
else:
if mode == 'fwd':
if input.name in d_inputs and output.name in d_outputs:
d_outputs[output.name] += d_inputs[input.name][mask]

if mode == 'rev':
if input.name in d_inputs and output.name in d_outputs:
d_inputs[input.name][mask] += d_outputs[output.name]

class UnmaskedConverter(om.ExplicitComponent):
"""
Expand Down Expand Up @@ -103,7 +127,20 @@ def setup(self):
output = self.options['output']
mask = self.options['mask']

self.add_input(input.name, shape=input.shape, tags=input.tags, distributed=distributed)
if isinstance(input, list):
if len(input) != len(mask):
raise ValueError("Input length and mask length not equal")

for i in range(len(input)-1):
for j in range(i+1, len(input)):
if (np.any(np.logical_and(mask[i], mask[j]))):
raise RuntimeWarning("Overlapping masking arrays, values will conflict.")

for i in range(len(input)):
self.add_input(input[i].name, shape=input[i].shape, tags=input[i].tags, distributed=distributed)
else:
self.add_input(input.name, shape=input.shape, tags=input.tags, distributed=distributed)

self.add_output(output.name, shape=output.shape, tags=output.tags, distributed=distributed)

def compute(self, inputs, outputs):
Expand All @@ -112,16 +149,32 @@ def compute(self, inputs, outputs):
mask = self.options['mask']
def_vals = self.options['default_values']
outputs[output.name][:] = def_vals
outputs[output.name][mask] = inputs[input.name]

if isinstance(input, list):
for i in range(len(input)):
outputs[output.name][mask[i]] = inputs[input[i].name]
else:
outputs[output.name][mask] = inputs[input.name]

def compute_jacvec_product(self, inputs, d_inputs, d_outputs, mode):
input = self.options['input']
output = self.options['output']
mask = self.options['mask']
if mode == 'fwd':
if input.name in d_inputs and output.name in d_outputs:
d_outputs[output.name][mask] += d_inputs[input.name]

if mode == 'rev':
if input.name in d_inputs and output.name in d_outputs:
d_inputs[input.name] += d_outputs[output.name][mask]
if isinstance(input, list):
for i in range(len(input)):
if mode == 'fwd':
if input[i].name in d_inputs and output.name in d_outputs:
d_outputs[output.name][mask[i]] += d_inputs[input[i].name]

if mode == 'rev':
if input[i].name in d_inputs and output.name in d_outputs:
d_inputs[input[i].name] += d_outputs[output.name][mask[i]]
else:
if mode == 'fwd':
if input.name in d_inputs and output.name in d_outputs:
d_outputs[output.name][mask] += d_inputs[input.name]

if mode == 'rev':
if input.name in d_inputs and output.name in d_outputs:
d_inputs[input.name] += d_outputs[output.name][mask]
73 changes: 72 additions & 1 deletion tests/unit_tests/test_mask_converter.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,7 @@
from openmdao.utils.assert_utils import assert_near_equal


class TestMaskConverter(unittest.TestCase):
class TestMaskConverterSingle(unittest.TestCase):
N_PROCS = 1 #TODO should be 2 or more but there is a bug in OM currently

def setUp(self):
Expand Down Expand Up @@ -59,5 +59,76 @@ def test_check_partials(self):
assert_near_equal(rel_error.forward, 0.0, tolerance=tol)
assert_near_equal(rel_error.forward_reverse, 0.0, tolerance=tol)

class TestMaskConverterMulti(unittest.TestCase):
N_PROCS = 1 #TODO should be 2 or more but there is a bug in OM currently

def setUp(self):
self.common = CommonMethods()
self.prob = om.Problem()
inputs = self.prob.model.add_subsystem('inputs', om.IndepVarComp())

inputs.add_output('unmasked_input', val=np.ones(10, dtype=float), distributed=True)
inputs.add_output('masked_input_1', val=np.arange(5, dtype=float), distributed=True)
inputs.add_output('masked_input_2', val=np.arange(5, dtype=float), distributed=True)

# Create a mask that masks every other entry of the input array
mask_input = MaskedVariableDescription('unmasked_input', shape=10, tags=['mphys_coordinates'])
mask_output = [
MaskedVariableDescription('masked_output_1', shape=5, tags=['mphys_coordinates']),
MaskedVariableDescription('masked_output_2', shape=5, tags=['mphys_coordinates']),
]
mask = [
np.zeros([10], dtype=bool),
np.zeros([10], dtype=bool),
]
mask[0][0:5] = True
mask[0][5:10] = False
mask[1][0:5] = False
mask[1][5:10] = True
masker = MaskedConverter(input=mask_input, output=mask_output, mask=mask, distributed=True)

self.prob.model.add_subsystem('masker', masker)

unmask_input = [
MaskedVariableDescription('masked_input_1', shape=5, tags=['mphys_coordinates']),
MaskedVariableDescription('masked_input_2', shape=5, tags=['mphys_coordinates']),
]
unmask_output = MaskedVariableDescription('unmasked_output', shape=10, tags=['mphys_coordinates'])
unmasker = UnmaskedConverter(input=unmask_input, output=unmask_output, mask=mask, distributed=True,
default_values=1.0)

self.prob.model.add_subsystem('unmasker', unmasker)

self.prob.model.connect('inputs.unmasked_input', 'masker.unmasked_input')
self.prob.model.connect('inputs.masked_input_1', 'unmasker.masked_input_1')
self.prob.model.connect('inputs.masked_input_2', 'unmasker.masked_input_2')

self.prob.setup(force_alloc_complex=True)

def test_run_model(self):
self.common.test_run_model(self, write_n2=False)

def test_check_partials(self):
partials = self.prob.check_partials(compact_print=True, method='cs')
tol = 1e-9

rel_error = partials['masker'][('masked_output_1', 'unmasked_input')]['rel error']
assert_near_equal(rel_error.reverse, 0.0, tolerance=tol)
assert_near_equal(rel_error.forward, 0.0, tolerance=tol)
assert_near_equal(rel_error.forward_reverse, 0.0, tolerance=tol)
rel_error = partials['masker'][('masked_output_2', 'unmasked_input')]['rel error']
assert_near_equal(rel_error.reverse, 0.0, tolerance=tol)
assert_near_equal(rel_error.forward, 0.0, tolerance=tol)
assert_near_equal(rel_error.forward_reverse, 0.0, tolerance=tol)

rel_error = partials['unmasker'][('unmasked_output', 'masked_input_1')]['rel error']
assert_near_equal(rel_error.reverse, 0.0, tolerance=tol)
assert_near_equal(rel_error.forward, 0.0, tolerance=tol)
assert_near_equal(rel_error.forward_reverse, 0.0, tolerance=tol)
rel_error = partials['unmasker'][('unmasked_output', 'masked_input_2')]['rel error']
assert_near_equal(rel_error.reverse, 0.0, tolerance=tol)
assert_near_equal(rel_error.forward, 0.0, tolerance=tol)
assert_near_equal(rel_error.forward_reverse, 0.0, tolerance=tol)

if __name__ == '__main__':
unittest.main()