Exception with Callback in Keras - Tensorflow 2.0 - Python

Question

The following code runs a Sequential Keras model, pretty straight forward , on the MNIST data that are packaged with Keras.

In running the following piece of code I get an exception.

The code is readily reproducible.

import tensorflow as tf

class myCallback(tf.keras.callbacks.Callback):
      def on_epoch_end(self, epoch, logs={}):
        if(logs.get('acc')>0.99):
          print("\nReached 99% accuracy so cancelling training!")
          self.model.stop_training = True

mnist = tf.keras.datasets.mnist

(x_train, y_train),(x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0

callbacks = myCallback()

model = tf.keras.models.Sequential([
  tf.keras.layers.Flatten(input_shape=(28, 28)),
  tf.keras.layers.Dense(512, activation=tf.nn.relu),
  tf.keras.layers.Dense(10, activation=tf.nn.softmax)
])
model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

model.fit(x_train, y_train, epochs=10, callbacks=[callbacks])

The exception is:

Epoch 1/10
59296/60000 [============================>.] - ETA: 0s - loss: 0.2005 - accuracy: 0.9400

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-26-f5e673b24d24> in <module>()
     23               metrics=['accuracy'])
     24 
---> 25 model.fit(x_train, y_train, epochs=10, callbacks=[callbacks])

C:\Program Files (x86)\Microsoft Visual Studio\Shared\Anaconda3_64\lib\site-packages\tensorflow\python\keras\engine\training.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_freq, max_queue_size, workers, use_multiprocessing, **kwargs)
    871           validation_steps=validation_steps,
    872           validation_freq=validation_freq,
--> 873           steps_name='steps_per_epoch')
    874 
    875   def evaluate(self,

C:\Program Files (x86)\Microsoft Visual Studio\Shared\Anaconda3_64\lib\site-packages\tensorflow\python\keras\engine\training_arrays.py in model_iteration(model, inputs, targets, sample_weights, batch_size, epochs, verbose, callbacks, val_inputs, val_targets, val_sample_weights, shuffle, initial_epoch, steps_per_epoch, validation_steps, validation_freq, mode, validation_in_fit, prepared_feed_values_from_dataset, steps_name, **kwargs)
    406     if mode == ModeKeys.TRAIN:
    407       # Epochs only apply to `fit`.
--> 408       callbacks.on_epoch_end(epoch, epoch_logs)
    409     progbar.on_epoch_end(epoch, epoch_logs)
    410 

C:\Program Files (x86)\Microsoft Visual Studio\Shared\Anaconda3_64\lib\site-packages\tensorflow\python\keras\callbacks.py in on_epoch_end(self, epoch, logs)
    288     logs = logs or {}
    289     for callback in self.callbacks:
--> 290       callback.on_epoch_end(epoch, logs)
    291 
    292   def on_train_batch_begin(self, batch, logs=None):

<ipython-input-26-f5e673b24d24> in on_epoch_end(self, epoch, logs)
      3 class myCallback(tf.keras.callbacks.Callback):
      4       def on_epoch_end(self, epoch, logs={}):
----> 5         if(logs.get('acc')>0.99):
      6           print("\nReached 99% accuracy so cancelling training!")
      7           self.model.stop_training = True

TypeError: '>' not supported between instances of 'NoneType' and 'float'

Answer 1

In model.compile function you defined metrics=['accuracy']. You need to use 'accuracy' in logs.get ie logs.get('accuracy').

Answer 2

It's just that with the upgrade of tensorflow to version 2.x the dictionary tag 'acc' has been changed to 'accuracy' therefore replacing the line 5 as follows should do the trick!

if(logs.get('accuracy')>0.99):

Answer 3

Just change logs.get('accuracy') --> logs.get('acc'). It should work fine!

Answer 4

I had the same problem. I changed it to "acc" and it worked like a charm. I made the following changes.

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['acc'])

And in the callback,

class myCallback(tf.keras.callbacks.Callback):
    def on_epoch_end(self, epoch, logs={}):
        if(logs.get("acc") >= 0.99):
            print("Reached 99% accuracy so cancelling training!")
            self.model.stop_training = True

Answer 5

In Jupyter notebook I had to use "acc", but in google Colab "accuracy" instead. I guess it depends on the tensorflow version installed.

Answer 6

I think it can come from the way you call your function :

If your function is

class myCallback(tf.keras.callbacks.Callback):
...

It should be called like that :

model.fit(x_train, y_train, epochs=10, callbacks=[myCallback()])

Answer 7

可能您正在使用 tensorflow 1.，所以您可以尝试： if(logs.get('acc')>0.998) and metrics=['acc']

Answer 8

The problem is logs.get('acc')>0.99 . On your end logs.get('acc') is None from some reason.

Just execute:

None>0.99 and you will get the same error. You probable migrated your code from Python 2 where this would actually work :).

You can simple tweak this with

if(logs.get('acc') is None): # in this case you cannot compare...

Or you can use try : ... except : blocks.

BTW, the same code works just fine on my end.

Answer 9

For some reason I did ['acc'] in the callback class, with ['accuracy'] in the metrics and it worked.