在 Keras 中，如何在预测期间使用自定义回调

Question

I have a keras model like this-

And for this model, I have a callback function like this-

import tensorflow as tf
import numpy as np

class WriteLayerValCallback(tf.keras.callbacks.Callback):
    def __init__(self):
            self.data = np.random.rand(1,10)

    def on_epoch_end(self, epoch, logs=None):
            #dns_layer = self.model.layers[6]
            dns_layer = self.model.get_layer('activation')
            outputs = dns_layer(self.data)
            tf.print(f'\n input: {self.data}')
            tf.print(f'\n output: {outputs}')

And I am predicting my model like this-

yhat = model.predict(X)

I like to call this callback function during Keras prediction -

Can anyone please help me, with how I can do that?

Answer 1

If I Understand Correctly (IIUC), You need on_predict_batch_end for calling callback in the keras prediction .

(You can read here , We can write call back with different function call : on_predict_begin , on_predict_end , on_predict_batch_end , on_test_end , ...)

import tensorflow as tf
import numpy as np

class CustomCallback(tf.keras.callbacks.Callback):
   def __init__(self, data):
        self.data = data
   def on_predict_batch_end(self, batch, logs=None):
        dns_layer = self.model.layers[6]
        outputs = dns_layer(self.data)
        tf.print(f'\n batch: {batch}')
        tf.print(f'\n input: {self.data}')
        tf.print(f'\n output: {outputs}')


x_train = tf.random.normal((10, 32, 32))
y_train = tf.random.uniform((10, 1), maxval=10)

model = tf.keras.models.Sequential()
model.add(tf.keras.layers.LSTM(256, input_shape=(x_train.shape[1], x_train.shape[2]), return_sequences=True))
model.add(tf.keras.layers.Dropout(0.2))
model.add(tf.keras.layers.LSTM(256))
model.add(tf.keras.layers.Dropout(0.2))
model.add(tf.keras.layers.Dense(10, activation='softmax')) 
model.add(tf.keras.layers.Dropout(0.2))
model.add(tf.keras.layers.Dense(5, activation='softmax'))
model.add(tf.keras.layers.Dropout(0.2))
model.add(tf.keras.layers.Dense(10, activation='softmax')) 
model.compile(optimizer='adam', loss = tf.keras.losses.SparseCategoricalCrossentropy(False))
model.summary()    

for layer in model.layers:
    print(layer)

model.fit(x_train, y_train , epochs=1, batch_size=32)
yhat = model.predict(tf.random.normal((5, 32, 32)), batch_size=1, callbacks=[CustomCallback(np.random.rand(1,10))])

Output:

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 lstm (LSTM)                 (None, 32, 256)           295936    
                                                                 
 dropout (Dropout)           (None, 32, 256)           0         
                                                                 
 lstm_1 (LSTM)               (None, 256)               525312    
                                                                 
 dropout_1 (Dropout)         (None, 256)               0         
                                                                 
 dense (Dense)               (None, 10)                2570      
                                                                 
 dropout_2 (Dropout)         (None, 10)                0         
                                                                 
 dense_1 (Dense)             (None, 5)                 55        
                                                                 
 dropout_3 (Dropout)         (None, 5)                 0         
                                                                 
 dense_2 (Dense)             (None, 10)                60        
                                                                 
=================================================================
Total params: 823,933
Trainable params: 823,933
Non-trainable params: 0
_________________________________________________________________
<keras.layers.recurrent_v2.LSTM object at 0x7fd9bf3a1690>
<keras.layers.core.dropout.Dropout object at 0x7fd9b74f60d0>
<keras.layers.recurrent_v2.LSTM object at 0x7fd9b32cb090>
<keras.layers.core.dropout.Dropout object at 0x7fd9b325c110>
<keras.layers.core.dense.Dense object at 0x7fd9b31f4690>
<keras.layers.core.dropout.Dropout object at 0x7fd9b315a890>
<keras.layers.core.dense.Dense object at 0x7fd9b32a1050>
<keras.layers.core.dropout.Dropout object at 0x7fd9b3239f50>
<keras.layers.core.dense.Dense object at 0x7fd9b317e1d0>
1/1 [==============================] - 12s 12s/step - loss: 2.3411

 batch: 0

 input: [[0.77380147 0.84481026 0.0211125  0.2323317  0.27449231 0.7934265
  0.82050726 0.28287153 0.56995795 0.65609332]]

 output: [[0.07206324 0.29164252 0.41107002 0.10761011 0.11761407]]

 batch: 1

 input: [[0.77380147 0.84481026 0.0211125  0.2323317  0.27449231 0.7934265
  0.82050726 0.28287153 0.56995795 0.65609332]]

 output: [[0.07206324 0.29164252 0.41107002 0.10761011 0.11761407]]

 batch: 2

 input: [[0.77380147 0.84481026 0.0211125  0.2323317  0.27449231 0.7934265
  0.82050726 0.28287153 0.56995795 0.65609332]]

 output: [[0.07206324 0.29164252 0.41107002 0.10761011 0.11761407]]

 batch: 3

 input: [[0.77380147 0.84481026 0.0211125  0.2323317  0.27449231 0.7934265
  0.82050726 0.28287153 0.56995795 0.65609332]]

 output: [[0.07206324 0.29164252 0.41107002 0.10761011 0.11761407]]

 batch: 4

 input: [[0.77380147 0.84481026 0.0211125  0.2323317  0.27449231 0.7934265
  0.82050726 0.28287153 0.56995795 0.65609332]]

 output: [[0.07206324 0.29164252 0.41107002 0.10761011 0.11761407]]