def _hijack_and_recover(self, parameterization, **kwargs):
module = tm
_, original_layers = ops.hijack_keras_module(parameterization, module,
**kwargs)
out = tm.build_model_with_all_configurable_types(self.inputs)
ops.recover_module_functions(original_layers, module)
return out
def testHijackingLocalAliases(self):
parameterization = {'conv/Conv2D': 1}
module = tm
_, original_layers = ops.hijack_keras_module(parameterization, module)
out = tm.build_simple_keras_model_from_local_aliases(self.inputs)
ops.recover_module_functions(original_layers, module)
self.assertEqual(out.shape.as_list()[-1], 1)
def testHijackingImportedLayerLib(self):
parameterization = {'conv/Conv2D': 1}
module = tm.layers
_, original_layers = ops.hijack_keras_module(parameterization, module)
out = tm.build_simple_keras_model(self.inputs)
ops.recover_module_functions(original_layers, module)
self.assertEqual(out.shape.as_list()[-1], 1)
def testConstructedOps(self):
parameterization = {
'conv/Conv2D': 1,
'sep_conv/separable_conv2d': 2,
'dense/Tensordot/MatMul': 3,
}
module = tm
constructed_ops, original_layers = ops.hijack_keras_module(
parameterization, module)
out = tm.build_model_with_all_configurable_types(self.inputs)
ops.recover_module_functions(original_layers, module)
self.assertEqual(out.shape.as_list()[-1], 3)
self.assertDictEqual(constructed_ops, parameterization)
def testRecover(self):
# If this test does not work well, then it might have some bizarre effect on
# other tests as it changes the functions in layers
decorator = ops.ConfigurableOps()
true_separable_conv2d = layers.separable_conv2d
original_dict = ops.hijack_module_functions(decorator, layers)
self.assertEqual(true_separable_conv2d,
original_dict['separable_conv2d'])
# Makes sure hijacking worked.
self.assertNotEqual(true_separable_conv2d, layers.separable_conv2d)
# Recovers original ops
ops.recover_module_functions(original_dict, layers)
self.assertEqual(true_separable_conv2d, layers.separable_conv2d)
def testPassThroughHijacking(self):
parameterization = {
'conv1/Conv2D':
0, # followed by BatchNorm, Activation, Add and Concat
'conv2/Conv2D': 1,
}
module = tm.layers
_, original_layers = ops.hijack_keras_module(
parameterization, module, keep_first_channel_alive=False)
# output = Concat([branch1, branch2, branch1 + branch2]).
# if branch1 vanishes, output should have only 2 channels.
output, branch1, branch2 = tm.build_two_branch_model(self.inputs)
ops.recover_module_functions(original_layers, module)
self.assertEqual(branch1, ops.VANISHED)
self.assertEqual(branch2.shape.as_list(), [1, 8, 8, 1])
self.assertEqual(output.shape.as_list(), [1, 8, 8, 2])