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'

When creating a callback , if we need an accuracy threshold for training , previous TF versions have logs.get('acc') but in this version we need to use logs.get('accuracy') for it to work. There wasn't any documentation regarding this change — amalik2205
– amalik2205, Commented Jun 21, 2020 at 16:26
The error is because logs.get('acc') must match metric value in model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy']) — Timothy L.J. Stewart
– Timothy L.J. Stewart, Commented Aug 18, 2020 at 1:55
@amalik2205 you're comment is very on point! +1. This also happened to me! — An old man in the sea.
– An old man in the sea., Commented Oct 6, 2021 at 22:00

Alif Jamaluddin · Accepted Answer · 2019-08-13 13:25:06Z

7

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

answered Aug 13, 2019 at 13:25

Alif Jamaluddin

5184 silver badges18 bronze badges

Sign up to request clarification or add additional context in comments.

Comments

blade · Accepted Answer · 2020-09-03 15:51:36Z

2

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

answered Sep 3, 2020 at 15:51

blade

8631 gold badge10 silver badges21 bronze badges

1 Comment

codingInMyBasement Over a year ago

same here, not sure why that is the case but this solution worked for me. Maybe there are different versions installed in Colab compared to the notebook.

retro-inferno · Accepted Answer · 2020-03-23 12:03:13Z

1

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):

answered Mar 23, 2020 at 12:03

retro-inferno

111 bronze badge

Comments

Aditya Saurabh · Accepted Answer · 2020-05-21 13:47:33Z

1

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

answered May 21, 2020 at 13:47

Aditya Saurabh

313 bronze badges

Comments

Akash B · Accepted Answer · 2020-06-15 05:39:46Z

1

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

answered Jun 15, 2020 at 5:39

Akash B

111 bronze badge

Comments

Thibault Bacqueyrisses · Accepted Answer · 2019-06-03 08:27:45Z

0

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()])

answered Jun 3, 2019 at 8:27

Thibault Bacqueyrisses

2,3511 gold badge9 silver badges18 bronze badges

Comments

Nguyễn Quốc Tấn · Accepted Answer · 2020-09-24 02:52:08Z

0

probably you are using tensorflow 1., so you may try: if(logs.get('acc')>0.998) andmetrics=['acc']

answered Sep 24, 2020 at 2:52

Nguyễn Quốc Tấn

1

Comments

prosti · Accepted Answer · 2019-06-03 08:38:10Z

-2

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.

edited Jun 3, 2019 at 8:38

answered Jun 3, 2019 at 8:31

prosti

46.9k19 gold badges199 silver badges161 bronze badges

Comments

Jan Černý · Accepted Answer · 2020-06-14 18:24:50Z

-2

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

enter image description here

edited Jun 14, 2020 at 18:24

Jan Černý

1,4162 gold badges22 silver badges39 bronze badges

answered Jun 14, 2020 at 13:18

Neil

1

Collectives™ on Stack Overflow

Exception with Callback in Keras - Tensorflow 2.0 - Python

9 Answers 9

Comments

1 Comment

Comments

Comments

Comments

Comments

Comments

Comments

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

9 Answers 9

Comments

1 Comment

Comments

Comments

Comments

Comments

Comments

Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related