Few issues with running gpu-notebook on a jupyter-tensorflow-cuda-full:v1.10.0-rc.1 based jupyter notebook

While running this test on a newly deployed kubeflow environment, I encountered few issues.
I am listing down the issues and changes used to workaround the issues.
Note: However it is possible that these workarounds are not the best options to overcome the issues.

1. There is a error at this cell.

# One hot encoding labels
y_train_encoded = keras.utils.to_categorical(y_train, num_classes = 10, dtype = 'float32')
y_test_encoded = keras.utils.to_categorical(y_test, num_classes = 10, dtype = 'float32')

This is the error.

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
Cell In[6], line 2
      1 # One hot encoding labels
----> 2 y_train_encoded = keras.utils.to_categorical(y_train, num_classes = 10, dtype = 'float32')
      3 y_test_encoded = keras.utils.to_categorical(y_test, num_classes = 10, dtype = 'float32')

TypeError: to_categorical() got an unexpected keyword argument 'dtype'

To overcome this we can simply drop the dtype attribute.

# One hot encoding labels
y_train_encoded = keras.utils.to_categorical(y_train, num_classes = 10)
y_test_encoded = keras.utils.to_categorical(y_test, num_classes = 10)

2. There is a error at this cell.

# MultiGPU prediction. Run nvidia-smi command in another terminal tab to check GPU ussage
model_gpu = get_model_cnn()
model_gpu.fit(X_train_scaled, y_train_encoded, epochs=10, batch_size=64)

This is the error.

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[10], line 2
      1 # MultiGPU prediction. Run nvidia-smi command in another terminal tab to check GPU ussage
----> 2 model_gpu = get_model_cnn()
      3 model_gpu.fit(X_train_scaled, y_train_encoded, epochs=10, batch_size=64)

Cell In[9], line 17, in get_model_cnn()
     15 model.add(Dense(10, activation='softmax'))
     16 # compile model
---> 17 opt = keras.optimizers.SGD(lr=0.001, momentum=0.9)
     18 model.compile(optimizer=opt, loss='categorical_crossentropy', metrics=['accuracy'])
     19 return model

File /opt/conda/lib/python3.11/site-packages/keras/src/optimizers/sgd.py:60, in SGD.__init__(self, learning_rate, momentum, nesterov, weight_decay, clipnorm, clipvalue, global_clipnorm, use_ema, ema_momentum, ema_overwrite_frequency, loss_scale_factor, gradient_accumulation_steps, name, **kwargs)
     43 def __init__(
     44     self,
     45     learning_rate=0.01,
   (...)
     58     **kwargs,
     59 ):
---> 60     super().__init__(
     61         learning_rate=learning_rate,
     62         name=name,
     63         weight_decay=weight_decay,
     64         clipnorm=clipnorm,
     65         clipvalue=clipvalue,
     66         global_clipnorm=global_clipnorm,
     67         use_ema=use_ema,
     68         ema_momentum=ema_momentum,
     69         ema_overwrite_frequency=ema_overwrite_frequency,
     70         loss_scale_factor=loss_scale_factor,
     71         gradient_accumulation_steps=gradient_accumulation_steps,
     72         **kwargs,
     73     )
     74     if not isinstance(momentum, float) or momentum < 0 or momentum > 1:
     75         raise ValueError("`momentum` must be a float between [0, 1].")

File /opt/conda/lib/python3.11/site-packages/keras/src/backend/tensorflow/optimizer.py:21, in TFOptimizer.__init__(self, *args, **kwargs)
     20 def __init__(self, *args, **kwargs):
---> 21     super().__init__(*args, **kwargs)
     22     self._distribution_strategy = tf.distribute.get_strategy()

File /opt/conda/lib/python3.11/site-packages/keras/src/optimizers/base_optimizer.py:90, in BaseOptimizer.__init__(self, learning_rate, weight_decay, clipnorm, clipvalue, global_clipnorm, use_ema, ema_momentum, ema_overwrite_frequency, loss_scale_factor, gradient_accumulation_steps, name, **kwargs)
     86     warnings.warn(
     87         "Argument `decay` is no longer supported and will be ignored."
     88     )
     89 if kwargs:
---> 90     raise ValueError(f"Argument(s) not recognized: {kwargs}")
     92 if name is None:
     93     name = auto_name(self.__class__.__name__)

ValueError: Argument(s) not recognized: {'lr': 0.001}

To overcome this we need to replace lr with learning_rate in the get_model_cnn() function.

opt = keras.optimizers.SGD(learning_rate=0.001, momentum=0.9)

3. There is a error at this cell.

# CPU only prediction
%%timeit -n1 -r1
# CPU
with tf.device('/CPU:0'):
    model_cpu = get_model()
    model_cpu.fit(X_train_scaled, y_train_encoded, epochs = 1)

This is the error.

UsageError: Line magic function `%%timeit` not found.

Need to move magic function above the comment line.

%%timeit -n1 -r1
# CPU only prediction

Locate the gpu-notebook cells described in the issue and run them in the jupyter-tensorflow-cuda-full:v1.10.0-rc.1 environment. Check the to_categorical, SGD, and %%timeit cells against the reported errors and apply the documented compatibility changes. Done means the notebook runs through these cells without the three reported errors.