def test_prune_simple_model_spec_validation_no_tags(self):
model_spec = [
schema.OneOf(['a', 'b', 'c'], basic_specs.OP_TAG),
schema.OneOf(['z', 'w'], basic_specs.FILTERS_TAG),
]
with self.assertRaisesWithLiteralMatch(
ValueError,
'Genotype contains 1 oneofs but model_spec contains 2'):
search_space_utils.prune_model_spec(model_spec=model_spec,
genotype=[0],
prune_filters_by_value=True)
def test_prune_model_spec_prune_filters_by_value_with_invalid_value(self):
model_spec = {
'op': schema.OneOf(['a', 'b', 'c'], basic_specs.OP_TAG),
'filters': schema.OneOf([128, 256, 512], basic_specs.FILTERS_TAG),
}
genotype = {
basic_specs.OP_TAG: [1],
basic_specs.FILTERS_TAG: [1024], # Use values instead of indices
}
with self.assertRaises(ValueError):
search_space_utils.prune_model_spec(model_spec,
genotype,
prune_filters_by_value=True)
def mnasnet_b1():
model_spec = _proxylessnas_search_base(
mobile_model_archive.MNASNET_FILTERS)
model_spec = search_space_utils.prune_model_spec(
model_spec,
{basic_specs.OP_TAG: mobile_model_archive.MNASNET_OPERATIONS})
return model_spec
def test_prune_model_spec_with_path_dropout_rate_tensor(self):
model_spec = {
'op1':
schema.OneOf([
mobile_search_space_v3.ConvSpec(kernel_size=2, strides=2),
basic_specs.ZeroSpec(),
], basic_specs.OP_TAG),
'op2':
schema.OneOf([
mobile_search_space_v3.ConvSpec(kernel_size=3, strides=4),
], basic_specs.OP_TAG),
'filter':
schema.OneOf([32], basic_specs.FILTERS_TAG),
}
model_spec = search_space_utils.prune_model_spec(
model_spec, {basic_specs.OP_TAG: [0, 0]},
path_dropout_rate=tf.constant(2.0) / tf.constant(10.0),
training=True)
self.assertCountEqual(model_spec.keys(), ['op1', 'op2', 'filter'])
self.assertEqual(model_spec['op1'].mask.shape, tf.TensorShape([1]))
self.assertIsNone(model_spec['op2'].mask)
self.assertIsNone(model_spec['filter'].mask)
# The value should either be 0 or 1 / (1 - path_dropout_rate) = 1.25
op_mask_value = self.evaluate(model_spec['op1'].mask)
self.assertTrue(
abs(op_mask_value - 0) < 1e-6 or abs(op_mask_value - 1.25) < 1e-6,
msg='Unexpected op_mask_value: {}'.format(op_mask_value))
def proxylessnas_mobile():
model_spec = _proxylessnas_search_base(
mobile_model_archive.PROXYLESSNAS_MOBILE_FILTERS)
model_spec = search_space_utils.prune_model_spec(
model_spec,
{basic_specs.OP_TAG: mobile_model_archive.PROXYLESSNAS_MOBILE_OPERATIONS})
return model_spec
def mobilenet_v2():
"""Specification for MobileNet V2 w/ relative expansion filters."""
model_spec = _proxylessnas_search_base(
mobile_model_archive.MOBILENET_V2_FILTERS)
model_spec = search_space_utils.prune_model_spec(
model_spec,
{basic_specs.OP_TAG: mobile_model_archive.MOBILENET_V2_OPERATIONS})
return model_spec
def test_prune_model_spec_with_path_dropout_training(self):
model_spec = {
'op1':
schema.OneOf([
mobile_search_space_v3.ConvSpec(kernel_size=2, strides=2),
basic_specs.ZeroSpec(),
], basic_specs.OP_TAG),
'op2':
schema.OneOf([
mobile_search_space_v3.ConvSpec(kernel_size=3, strides=4),
], basic_specs.OP_TAG),
'filter':
schema.OneOf([32], basic_specs.FILTERS_TAG),
}
model_spec = search_space_utils.prune_model_spec(
model_spec, {basic_specs.OP_TAG: [0, 0]},
path_dropout_rate=0.2,
training=True)
self.assertCountEqual(model_spec.keys(), ['op1', 'op2', 'filter'])
self.assertEqual(model_spec['op1'].mask.shape, tf.TensorShape([1]))
self.assertIsNone(model_spec['op2'].mask)
self.assertIsNone(model_spec['filter'].mask)
self.assertEqual(
model_spec['op1'].choices,
[mobile_search_space_v3.ConvSpec(kernel_size=2, strides=2)])
self.assertEqual(
model_spec['op2'].choices,
[mobile_search_space_v3.ConvSpec(kernel_size=3, strides=4)])
self.assertEqual(model_spec['filter'].choices, [32])
self.assertEqual(model_spec['op1'].tag, basic_specs.OP_TAG)
self.assertEqual(model_spec['op2'].tag, basic_specs.OP_TAG)
self.assertEqual(model_spec['filter'].tag, basic_specs.FILTERS_TAG)
op_mask_sum = 0
for _ in range(100):
# The value should either be 0 or 1 / (1 - path_dropout_rate) = 1.25
op_mask_value = self.evaluate(model_spec['op1'].mask)
self.assertTrue(
abs(op_mask_value - 0) < 1e-6
or abs(op_mask_value - 1.25) < 1e-6,
msg='Unexpected op_mask_value: {}'.format(op_mask_value))
op_mask_sum += op_mask_value[0]
# The probability of this test failing by random chance is roughly 0.002%.
# Our random number generators are deterministically seeded, so the test
# shouldn't be flakey.
self.assertGreaterEqual(op_mask_sum, 75)
self.assertLessEqual(op_mask_sum, 113)
def test_prune_simple_model_spec_no_tags(self):
pruned_spec = search_space_utils.prune_model_spec(
model_spec=[
schema.OneOf(['a', 'b', 'c'], basic_specs.OP_TAG),
schema.OneOf(['z', 'w'], basic_specs.FILTERS_TAG),
],
genotype=[2, 1],
prune_filters_by_value=True)
self.assertEqual(pruned_spec, [
schema.OneOf(['c'], basic_specs.OP_TAG),
schema.OneOf(['w'], basic_specs.FILTERS_TAG),
])
def get_model_spec(ssd,
filters=None,
op_indices=None,
indices=None,
filters_multipliers=1.0,
path_dropout_rate=0.0,
training=None):
"""Get a model spec namedtuple for the given search space definition.
Args:
ssd: Search space definition to use.
filters: List of filter sizes to use. Required for V2 search spaces.
op_indices: List of integers specifying the operations to select.
indices: List of integers specifying the values to use for all the
operations in the search space. If specified, this will override
op_indices and filters.
filters_multipliers: Single value or a list of possible values, used to
scale up/down the number of filters in each layer of the network.
path_dropout_rate: Rate of path dropout to use during stand-alone training.
training: Boolean. Only needs to be specified if path_dropout_rate > 0.
Returns:
A basic_specs.ConvTowerSpec namedtuple.
"""
if ssd in mobile_search_space_v3.ALL_SSDS:
model_spec = mobile_search_space_v3.get_search_space_spec(ssd)
else:
raise ValueError('Unsupported SSD: {}'.format(ssd))
if indices:
genotype = indices
else:
genotype = dict()
if op_indices:
genotype[basic_specs.OP_TAG] = op_indices
if filters and ssd in mobile_search_space_v3.ALL_SSDS:
genotype[basic_specs.FILTERS_TAG] = filters
model_spec = search_space_utils.prune_model_spec(
model_spec,
genotype,
prune_filters_by_value=True,
path_dropout_rate=path_dropout_rate,
training=training)
model_spec = search_space_utils.scale_conv_tower_spec(
model_spec, filters_multipliers)
return model_spec
def test_prune_model_spec_prune_filters_by_value(self):
model_spec = {
'op': schema.OneOf(['a', 'b', 'c'], basic_specs.OP_TAG),
'filters': schema.OneOf([128, 256, 512], basic_specs.FILTERS_TAG),
}
genotype = {
basic_specs.OP_TAG: [1],
basic_specs.FILTERS_TAG: [512], # Use values instead of indices
}
pruned_spec = search_space_utils.prune_model_spec(
model_spec, genotype, prune_filters_by_value=True)
expected_spec = {
'op': schema.OneOf(['b'], basic_specs.OP_TAG),
'filters': schema.OneOf([512], basic_specs.FILTERS_TAG),
}
self.assertEqual(pruned_spec, expected_spec)
def test_prune_model_spec_with_path_dropout_eval(self):
model_spec = {
'op1':
schema.OneOf([
mobile_search_space_v3.ConvSpec(kernel_size=2, strides=2),
basic_specs.ZeroSpec(),
], basic_specs.OP_TAG),
'op2':
schema.OneOf([
mobile_search_space_v3.ConvSpec(kernel_size=3, strides=4),
], basic_specs.OP_TAG),
'filter':
schema.OneOf([32], basic_specs.FILTERS_TAG),
}
model_spec = search_space_utils.prune_model_spec(
model_spec, {basic_specs.OP_TAG: [0, 0]},
path_dropout_rate=0.2,
training=False)
self.assertCountEqual(model_spec.keys(), ['op1', 'op2', 'filter'])
# Even though path_dropout_rate=0.2, the controller should not populate
# the mask for op1 because we called prune_model_spec() with training=False.
# In other words, path_dropout_rate should only affect the behavior during
# training, not during evaluation.
self.assertIsNone(model_spec['op1'].mask)
self.assertIsNone(model_spec['op2'].mask)
self.assertIsNone(model_spec['filter'].mask)
self.assertEqual(model_spec['op1'].tag, basic_specs.OP_TAG)
self.assertEqual(model_spec['op2'].tag, basic_specs.OP_TAG)
self.assertEqual(model_spec['filter'].tag, basic_specs.FILTERS_TAG)
self.assertEqual(
model_spec['op1'].choices,
[mobile_search_space_v3.ConvSpec(kernel_size=2, strides=2)])
self.assertEqual(
model_spec['op2'].choices,
[mobile_search_space_v3.ConvSpec(kernel_size=3, strides=4)])
self.assertEqual(model_spec['filter'].choices, [32])
