To print dataset after each layer? possible

General Discussion
, ,
#1

I have following model defined (below) and using fit to train. But during training, I would like to print out the dataset or at least its shape/dimensions for troubleshooting purpose. Is it possble?

model=keras.models.Sequential([\
#    keras.layers.Conv2D(64, 7, activation="relu", padding="same", input_shape=[28, 28, 1]),\
    keras.layers.Conv2D(64, 7, activation="relu", padding="same", input_shape=[28, 28, 1]),\
    keras.layers.MaxPooling2D(2), \
    keras.layers.Conv2D(128, 3, activation="relu", padding="same"), \
    keras.layers.Conv2D(128, 3, activation="relu", padding="same"), \
    keras.layers.MaxPooling2D(2), \
    keras.layers.Conv2D(256, 3, activation="relu", padding="same"), \
    keras.layers.Conv2D(256, 3, activation="relu", padding="same"), \
    keras.layers.MaxPooling2D(2), \
    keras.layers.Flatten(), \
    keras.layers.Dense(128, activation="relu"), \
    keras.layers.Dropout(0.5), \
    keras.layers.Dense(64, activation="relu"), \
    keras.layers.Dropout(0.5), \
    keras.layers.Dense(10, activation="softmax") \
])
model.compile(loss="sparse_categorical_crossentropy", optimizer="sgd", metrics=["accuracy"])
history=model.fit(X_train, y_train, epochs=CONFIG_EPOCHS, batch_size=CONFIG_BATCH_SIZE, validation_data=(X_valid, y_valid))

and I am doigng exactly same thing in pytorch and this is how pytorch conveniently allows printing of x between layers:

class MLP(Module):

prepare the dataset

# define model elements

def init(self):
super(MLP, self).init()
self.conv1 = Conv2d(1, 64, 7, padding=“same”)
self.act1 = ReLU()
self.maxpool1 = MaxPool2d(2)
…
…

def forward(self, X):
if DEBUG:
print("forward entered: X: ", X.size())

printdbg("X: " + str(X.size()))

X = self.conv1(X)
X = self.act1(X)
printdbg("X, conv1/act1: " + str(X.size()))
X = self.maxpool1(X)
printdbg("X, maxpool1: " + str(X.size()))

X = self.conv2a(X)
X = self.act2a(X)
printdbg("X, conv2a/act2a: " + str(X.size()))
X = self.conv2b(X)
X = self.act2b(X)

def printdbg(msg):
if DEBUG_PRT:
print(msg)

#3

You can visualize the whole graph with the shapes of each layer:
tf.keras.utils.plot_model(model, show_shapes=True, show_dtype=True)

#5

Thanks, that could be possibility but I really want to print the dataset dimension just like it in pytorch. Provided if it is possible at all?

#7

For datasets dimensions printout, following worked pretty good right after training (fit)

for i in model.layers:
        print("------")
#       print(i, "\n", i.input_shape, "\n", i.output_shape, "\n", i.get_weights())
        print(i, "\n", i.input_shape, "\n", i.output_shape, "\n")



860/860 [==============================] - 6s 7ms/step - loss: 0.6708 - accuracy: 0.7530 - val_loss: 0.4993 - val_accuracy: 0.8162
------
<keras.layers.convolutional.Conv2D object at 0x7f87bcdf11d0>
 (None, 28, 28, 1)
 (None, 28, 28, 64)

------
<keras.layers.pooling.MaxPooling2D object at 0x7f869ff17400>
 (None, 28, 28, 64)
 (None, 14, 14, 64)

------
<keras.layers.convolutional.Conv2D object at 0x7f869ff17630>
 (None, 14, 14, 64)
 (None, 14, 14, 128)

------
<keras.layers.convolutional.Conv2D object at 0x7f869ff17b38>
 (None, 14, 14, 128)
 (None, 14, 14, 128)

------
<keras.layers.pooling.MaxPooling2D object at 0x7f869ff45080>
 (None, 14, 14, 128)
 (None, 7, 7, 128)

------
<keras.layers.convolutional.Conv2D object at 0x7f869ff452b0>
 (None, 7, 7, 128)
 (None, 7, 7, 256)

------
<keras.layers.convolutional.Conv2D object at 0x7f869ff457b8>
 (None, 7, 7, 256)
 (None, 7, 7, 256)

------
<keras.layers.pooling.MaxPooling2D object at 0x7f869ff45cc0>
 (None, 7, 7, 256)
 (None, 3, 3, 256)

------
<keras.layers.core.Flatten object at 0x7f869ff45ef0>
 (None, 3, 3, 256)
 (None, 2304)

------
<keras.layers.core.Dense object at 0x7f869fecb0b8>
 (None, 2304)
 (None, 128)

------
<keras.layers.core.Dropout object at 0x7f869fecb3c8>
 (None, 128)
 (None, 128)

------
<keras.layers.core.Dense object at 0x7f869fecb4e0>
 (None, 128)
 (None, 64)

------
<keras.layers.core.Dropout object at 0x7f869fecb828>
 (None, 64)
 (None, 64)

------
<keras.layers.core.Dense object at 0x7f869fecb908>
 (None, 64)
 (None, 10)
#9

In Tensorflow model.summary() gives the output shape/dimension at each layer along with Parameters.

_________________________________________________________________
Model: "sequential_6"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d (Conv2D)              (None, 123, 123, 32)      2432      
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 121, 121, 32)      9248      
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 40, 40, 32)        0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 38, 38, 32)        9248      
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 36, 36, 32)        9248      
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 12, 12, 32)        0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 10, 10, 32)        9248      
_________________________________________________________________
conv2d_5 (Conv2D)            (None, 8, 8, 32)          9248      
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 4, 4, 32)          0         
=================================================================
Total params: 48,672
Trainable params: 48,672
Non-trainable params: 0