diff --git a/src/model_modules/flatten.py b/src/model_modules/flatten.py
index e2dde4481cb405078dffdff324f5d8157388cd83..39d61f251eea165fd427cb36d95dd5acc712dd03 100644
--- a/src/model_modules/flatten.py
+++ b/src/model_modules/flatten.py
@@ -18,6 +18,8 @@ def get_activation(input_to_activate: keras.layers, activation: Union[Callable,
:return:
"""
if isinstance(activation, str):
+ name = kwargs.pop('name', None)
+ kwargs['name'] = f'{name}_{activation}'
act = keras.layers.Activation(activation, **kwargs)(input_to_activate)
else:
act = activation(**kwargs)(input_to_activate)
@@ -63,12 +65,9 @@ def flatten_tail(input_x: keras.layers, inner_neurons: int, activation: Union[Ca
if reduction_filter is None:
x_in = input_x
else:
- x_in = keras.layers.Conv2D(reduction_filter, (1, 1), name=f'{name}_Conv_1x1')(input_x)
+ x_in = keras.layers.Conv2D(reduction_filter, (1, 1), name=f'{name}_Conv_1x1',
+ kernel_regularizer=kernel_regularizer)(input_x)
x_in = get_activation(x_in, activation, name=f'{name}_conv_act')
- # if isinstance(activation, str):
- # x_in = keras.layers.Activation(activation, )
- # else:
- # x_in = activation(name='{}_conv_act'.format(name))(x_in)
x_in = keras.layers.Flatten(name='{}'.format(name))(x_in)
@@ -80,10 +79,6 @@ def flatten_tail(input_x: keras.layers, inner_neurons: int, activation: Union[Ca
x_in = keras.layers.Activation('tanh')(x_in)
else:
x_in = get_activation(x_in, activation, name=f'{name}_act')
- # try:
- # x_in = activation(name='{}_act'.format(name))(x_in)
- # except:
- # x_in = activation()(x_in)
if dropout_rate is not None:
x_in = keras.layers.Dropout(dropout_rate, name='{}_Dropout_2'.format(name))(x_in)
diff --git a/test/test_model_modules/test_flatten_tail.py b/test/test_model_modules/test_flatten_tail.py
new file mode 100644
index 0000000000000000000000000000000000000000..0de138ec2323aea3409d5deadfb26c9741b89f50
--- /dev/null
+++ b/test/test_model_modules/test_flatten_tail.py
@@ -0,0 +1,119 @@
+import keras
+import pytest
+from src.model_modules.flatten import flatten_tail, get_activation
+
+
+class TestGetActivation:
+
+ @pytest.fixture()
+ def model_input(self):
+ input_x = keras.layers.Input(shape=(7, 1, 2))
+ return input_x
+
+ def test_string_act(self, model_input):
+ x_in = get_activation(model_input, activation='relu', name='String')
+ act = x_in._keras_history[0]
+ assert act.name == 'String_relu'
+
+ def test_sting_act_unknown(self, model_input):
+ with pytest.raises(ValueError) as einfo:
+ get_activation(model_input, activation='invalid_activation', name='String')
+ assert 'Unknown activation function:invalid_activation' in str(einfo.value)
+
+ def test_layer_act(self, model_input):
+ x_in = get_activation(model_input, activation=keras.layers.advanced_activations.ELU, name='adv_layer')
+ act = x_in._keras_history[0]
+ assert act.name == 'adv_layer'
+
+ def test_layer_act_invalid(self, model_input):
+ with pytest.raises(TypeError) as einfo:
+ get_activation(model_input, activation=keras.layers.Conv2D, name='adv_layer')
+
+
+class TestFlattenTail:
+
+ @pytest.fixture()
+ def model_input(self):
+ input_x = keras.layers.Input(shape=(7, 1, 2))
+ return input_x
+
+ @staticmethod
+ def step_in(element, depth=1):
+ for _ in range(depth):
+ element = element.input._keras_history[0]
+ return element
+
+ def test_flatten_tail_no_bound_no_regul_no_drop(self, model_input):
+ tail = flatten_tail(input_x=model_input, inner_neurons=64, activation=keras.layers.advanced_activations.ELU,
+ output_neurons=2, output_activation='linear',
+ reduction_filter=None,
+ name='Main_tail',
+ bound_weight=False,
+ dropout_rate=None,
+ kernel_regularizer=None)
+ final_act = tail._keras_history[0]
+ assert final_act.name == 'Main_tail_final_act_linear'
+ final_dense = self.step_in(final_act)
+ assert final_act.name == 'Main_tail_final_act_linear'
+ assert final_dense.units == 2
+ assert final_dense.kernel_regularizer is None
+ inner_act = self.step_in(final_dense)
+ assert inner_act.name == 'Main_tail_act'
+ assert inner_act.__class__.__name__ == 'ELU'
+ inner_dense = self.step_in(inner_act)
+ assert inner_dense.name == 'Main_tail_inner_Dense'
+ assert inner_dense.units == 64
+ assert inner_dense.kernel_regularizer is None
+ flatten = self.step_in(inner_dense)
+ assert flatten.name == 'Main_tail'
+ input_layer = self.step_in(flatten)
+ assert input_layer.input_shape == (None, 7, 1, 2)
+
+ def test_flatten_tail_all_settings(self, model_input):
+ tail = flatten_tail(input_x=model_input, inner_neurons=64, activation=keras.layers.advanced_activations.ELU,
+ output_neurons=3, output_activation='linear',
+ reduction_filter=32,
+ name='Main_tail_all',
+ bound_weight=True,
+ dropout_rate=.35,
+ kernel_regularizer=keras.regularizers.l2())
+
+ final_act = tail._keras_history[0]
+ assert final_act.name == 'Main_tail_all_final_act_linear'
+
+ final_dense = self.step_in(final_act)
+ assert final_dense.name == 'Main_tail_all_out_Dense'
+ assert final_dense.units == 3
+ assert isinstance(final_dense.kernel_regularizer, keras.regularizers.L1L2)
+
+ final_dropout = self.step_in(final_dense)
+ assert final_dropout.name == 'Main_tail_all_Dropout_2'
+ assert final_dropout.rate == 0.35
+
+ inner_act = self.step_in(final_dropout)
+ assert inner_act.get_config() == {'name': 'activation_1', 'trainable': True, 'activation': 'tanh'}
+
+ inner_dense = self.step_in(inner_act)
+ assert inner_dense.units == 64
+ assert isinstance(inner_dense.kernel_regularizer, keras.regularizers.L1L2)
+
+ inner_dropout = self.step_in(inner_dense)
+ assert inner_dropout.get_config() == {'name': 'Main_tail_all_Dropout_1', 'trainable': True, 'rate': 0.35,
+ 'noise_shape': None, 'seed': None}
+
+ flatten = self.step_in(inner_dropout)
+ assert flatten.get_config() == {'name': 'Main_tail_all', 'trainable': True, 'data_format': 'channels_last'}
+
+ reduc_act = self.step_in(flatten)
+ assert reduc_act.get_config() == {'name': 'Main_tail_all_conv_act', 'trainable': True, 'alpha': 1.0}
+
+ reduc_conv = self.step_in(reduc_act)
+
+ assert reduc_conv.kernel_size == (1, 1)
+ assert reduc_conv.name == 'Main_tail_all_Conv_1x1'
+ assert reduc_conv.filters == 32
+ assert isinstance(reduc_conv.kernel_regularizer, keras.regularizers.L1L2)
+
+ input_layer = self.step_in(reduc_conv)
+ assert input_layer.input_shape == (None, 7, 1, 2)
+