Skip to content

Commit

Permalink
remove no longer working nested models
Browse files Browse the repository at this point in the history
  • Loading branch information
@Dobiasd
Dobiasd committed Apr 9, 2024
1 parent 18fb8e0 commit e023db1
Showing 1 changed file with 50 additions and 44 deletions.
94 changes: 50 additions & 44 deletions keras_export/generate_test_models.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,7 @@
from tensorflow.keras.layers import MaxPooling2D, AveragePooling2D, UpSampling2D
from tensorflow.keras.layers import MaxPooling3D, AveragePooling3D
from tensorflow.keras.layers import MultiHeadAttention
from tensorflow.keras.layers import Multiply, Add, Subtract, Average, Maximum, Minimum, Dot
from tensorflow.keras.layers import Multiply, Add, Subtract, Average, Maximum, Minimum
from tensorflow.keras.layers import Permute, Reshape, RepeatVector
from tensorflow.keras.layers import SeparableConv2D, DepthwiseConv2D
from tensorflow.keras.layers import ZeroPadding3D, Cropping3D
Expand Down Expand Up @@ -170,11 +170,11 @@ def get_test_model_exhaustive():
outputs.append(GlobalAveragePooling1D(keepdims=True)(inputs[6]))

outputs.append(Normalization(axis=None, mean=2.1, variance=2.2)(inputs[4]))
#outputs.append(Normalization(axis=-1, mean=2.1, variance=2.2)(inputs[6])) # No longer supported in TensorFlow 2.16
# outputs.append(Normalization(axis=-1, mean=2.1, variance=2.2)(inputs[6])) # No longer supported in TensorFlow 2.16
outputs.append(Normalization(axis=-1, mean=2.1, variance=2.2)(inputs[46]))
outputs.append(Normalization(axis=1, mean=2.1, variance=2.2)(inputs[46]))
outputs.append(Normalization(axis=-1, mean=2.1, variance=2.2)(inputs[47]))
#outputs.append(Normalization(axis=1, mean=2.1, variance=2.2)(inputs[47])) # No longer supported in TensorFlow 2.16
# outputs.append(Normalization(axis=1, mean=2.1, variance=2.2)(inputs[47])) # No longer supported in TensorFlow 2.16
outputs.append(Normalization(axis=2, mean=2.1, variance=2.2)(inputs[47]))
for axis in range(1, 6):
shape = input_shapes[0][axis - 1]
Expand Down Expand Up @@ -225,8 +225,8 @@ def get_test_model_exhaustive():
outputs.append(Resizing(5, 6)(inputs[4]))
outputs.append(Resizing(19, 53, interpolation="bilinear")(inputs[23]))
outputs.append(Resizing(19, 53, interpolation="nearest")(inputs[23]))
#outputs.append(Resizing(7, 9, interpolation="area")(inputs[22])) # No longer supported in TensorFlow 2.16
#outputs.append(Resizing(19, 53, interpolation="area")(inputs[23])) # No longer supported in TensorFlow 2.16
# outputs.append(Resizing(7, 9, interpolation="area")(inputs[22])) # No longer supported in TensorFlow 2.16
# outputs.append(Resizing(19, 53, interpolation="area")(inputs[23])) # No longer supported in TensorFlow 2.16
outputs.append(Resizing(19, 53, crop_to_aspect_ratio=True)(inputs[23]))

outputs.append(Permute((3, 4, 1, 5, 2))(inputs[0]))
Expand Down Expand Up @@ -364,14 +364,14 @@ def get_test_model_exhaustive():
outputs.append(Minimum()([inputs[8], inputs[9]]))

# No longer works in TensorFlow 2.16, see: https://github.com/tensorflow/tensorflow/issues/65056
#for normalize in [True, False]:
#outputs.append(Dot(axes=(1, 1), normalize=normalize)([inputs[8], inputs[9]]))
#outputs.append(Dot(axes=(1, 1), normalize=normalize)([inputs[0], inputs[10]]))
#outputs.append(Dot(axes=1, normalize=normalize)([inputs[0], inputs[10]]))
#outputs.append(Dot(axes=(3, 1), normalize=normalize)([inputs[31], inputs[32]]))
#outputs.append(Dot(axes=(2, 3), normalize=normalize)([inputs[31], inputs[32]]))
#outputs.append(Dot(axes=(2, 3), normalize=normalize)([inputs[14], inputs[16]]))
#outputs.append(Dot(axes=(3, 2), normalize=normalize)([inputs[24], inputs[26]]))
# for normalize in [True, False]:
# outputs.append(Dot(axes=(1, 1), normalize=normalize)([inputs[8], inputs[9]]))
# outputs.append(Dot(axes=(1, 1), normalize=normalize)([inputs[0], inputs[10]]))
# outputs.append(Dot(axes=1, normalize=normalize)([inputs[0], inputs[10]]))
# outputs.append(Dot(axes=(3, 1), normalize=normalize)([inputs[31], inputs[32]]))
# outputs.append(Dot(axes=(2, 3), normalize=normalize)([inputs[31], inputs[32]]))
# outputs.append(Dot(axes=(2, 3), normalize=normalize)([inputs[14], inputs[16]]))
# outputs.append(Dot(axes=(3, 2), normalize=normalize)([inputs[24], inputs[26]]))

outputs.append(Reshape((16,))(inputs[8]))
outputs.append(Reshape((2, 8))(inputs[8]))
Expand All @@ -397,10 +397,10 @@ def get_test_model_exhaustive():
outputs.append(Concatenate(axis=axis)([inputs[14], inputs[15]]))
for axis in [-1, 3]:
outputs.append(Concatenate(axis=axis)([inputs[16], inputs[17]]))
#for axis in [-1, 4]:
#outputs.append(Concatenate(axis=axis)([inputs[18], inputs[19]])) # no longer supported in TensorFlow 2.16
#for axis in [-1, 5]:
#outputs.append(Concatenate(axis=axis)([inputs[20], inputs[21]])) # no longer supported in TensorFlow 2.16
# for axis in [-1, 4]:
# outputs.append(Concatenate(axis=axis)([inputs[18], inputs[19]])) # no longer supported in TensorFlow 2.16
# for axis in [-1, 5]:
# outputs.append(Concatenate(axis=axis)([inputs[20], inputs[21]])) # no longer supported in TensorFlow 2.16

outputs.append(UpSampling1D(size=2)(inputs[6]))
# outputs.append(UpSampling1D(size=2)(inputs[8])) # ValueError: Input 0 of layer up_sampling1d_1 is incompatible with the layer: expected ndim=3, found ndim=2. Full shape received: [None, 16]
Expand Down Expand Up @@ -503,39 +503,45 @@ def get_test_model_exhaustive():
outputs.append(Maximum()([inputs[26], inputs[30], inputs[30]]))
outputs.append(Concatenate()([inputs[26], inputs[30], inputs[30]]))

intermediate_input_shape = (3,)
intermediate_in = Input(intermediate_input_shape)
intermediate_x = intermediate_in
intermediate_x = Dense(8)(intermediate_x)
intermediate_x = Dense(5, name='duplicate_layer_name')(intermediate_x)
intermediate_model = Model(
inputs=[intermediate_in], outputs=[intermediate_x],
name='intermediate_model')
intermediate_model.compile(loss='mse', optimizer='nadam')
# TensorFlow 2.16 no longer puts
# "inbound_nodes": []
# for such nested models.
# todo: Check if the situation resolved with later versions.
if False:
intermediate_input_shape = (3,)
intermediate_in = Input(intermediate_input_shape)
intermediate_x = intermediate_in
intermediate_x = Dense(8)(intermediate_x)
intermediate_x = Dense(5, name='duplicate_layer_name')(intermediate_x)
intermediate_model = Model(
inputs=[intermediate_in], outputs=[intermediate_x],
name='intermediate_model')
intermediate_model.compile(loss='mse', optimizer='nadam')

x = intermediate_model(x) # (1, 1, 5)
x = intermediate_model(x)[0] # (1, 1, 5)

intermediate_model_2 = Sequential()
intermediate_model_2.add(Dense(7, input_shape=(5,)))
intermediate_model_2.add(Dense(5, name='duplicate_layer_name'))
intermediate_model_2.compile(optimizer='rmsprop',
loss='categorical_crossentropy')

x = intermediate_model_2(x) # (1, 1, 5)
intermediate_model_2 = Sequential(name="intermediate_model_2")
intermediate_model_2.add(Dense(7, input_shape=(5,)))
intermediate_model_2.add(Dense(5, name='duplicate_layer_name'))
intermediate_model_2.compile(optimizer='rmsprop',
loss='categorical_crossentropy')

intermediate_model_3_nested = Sequential()
intermediate_model_3_nested.add(Dense(7, input_shape=(6,)))
intermediate_model_3_nested.compile(optimizer='rmsprop', loss='categorical_crossentropy')
x = intermediate_model_2(x) # (1, 1, 5)

intermediate_model_3 = Sequential()
intermediate_model_3.add(Dense(6, input_shape=(5,)))
intermediate_model_3.add(intermediate_model_3_nested)
intermediate_model_3.add(Dense(8))
intermediate_model_3.compile(optimizer='rmsprop', loss='categorical_crossentropy')
intermediate_model_3_nested = Sequential(name="intermediate_model_3_nested")
intermediate_model_3_nested.add(Dense(7, input_shape=(6,)))
intermediate_model_3_nested.compile(optimizer='rmsprop', loss='categorical_crossentropy')

x = intermediate_model_3(x) # (1, 1, 8)
intermediate_model_3 = Sequential(name="intermediate_model_3")
intermediate_model_3.add(Dense(6, input_shape=(5,)))
intermediate_model_3.add(intermediate_model_3_nested)
intermediate_model_3.add(Dense(8))
intermediate_model_3.compile(optimizer='rmsprop', loss='categorical_crossentropy')

x = Dense(3)(x) # (1, 1, 3)
x = intermediate_model_3(x) # (1, 1, 8)

x = Dense(3)(x) # (1, 1, 3)

shared_activation = Activation('tanh')

Expand Down Expand Up @@ -690,7 +696,7 @@ def get_test_model_recurrent():
nested_model.compile(loss='categorical_crossentropy', optimizer='nadam')
outputs.append(TimeDistributed(nested_model)(inputs[0]))

nested_sequential_model = Sequential()
nested_sequential_model = Sequential(name="nested_sequential_model")
nested_sequential_model.add(Flatten(input_shape=input_shapes[0][1:]))
nested_sequential_model.compile(optimizer='rmsprop',
loss='categorical_crossentropy')
Expand Down

0 comments on commit e023db1

Please sign in to comment.