Saving a model defined by model subclassing can not be saved about tensorflow HOT 5 CLOSED

Hi @PowerToThePeople111 ,

I don't think the problem in saving is actually coming from model subclassing. Most probably, it comes from passing df as argument to the call method.

By slightly changing your code to pass a list instead of a Pandas dataframe, the model can be saved.

class DataPrep2(tf.keras.Model):
    def __init__(self):
        super().__init__()
        self.layer1 = tf.keras.layers.StringLookup(output_mode='int', vocabulary=["a", "b"])
        self.layer2 = tf.keras.layers.Discretization(bin_boundaries=[0,1,2,3,4], output_mode='int')


    def call(self, x):
        return {
            "conv1": self.layer1(x[0]),
            "conv2": self.layer2(x[1])
        }
dp2 = DataPrep2()
dp2([df["var1"], df["var2"]])
dp2.save("delteme")

from tensorflow.

@PowerToThePeople111 Please refer to the TensorFlow documentation for the latest recommendations on saving subclassed models: https://www.tensorflow.org/tutorials/keras/save_and_load and kindly provide the access to your notebook.
Thank you!

from tensorflow.

Hi sushreebarsa,

thank you for your reply. You should now be able to access the notebook. Also I put the whole code neccessary below the link to it.

I read the documentation again and was lead by the following parts:

You can switch to the SavedModel format by:

Passing save_format='tf' to save()
Passing a filename without an extension

...

*Custom objects (for example, subclassed models or layers) require special attention when saving and loading. Refer to the Saving custom objects section below.

...

So for me it seemed that the above is basically everything i need to do since later on in the "Custom objects"-section you can read at the very top:

If you are using the SavedModel format, you can skip this section.

So actually I should be able to store the model by using save() with save_format 'tf' and providing a filename without ending, which is what I did:
Neither dp2.save("delteme", save_format="tf") nor dp2.save("delteme") did the trick for me and result in the error message i posted in my initial statement.

Did I miss something here?

from tensorflow.

Thank you @karimsr4,

that solution works. :)

from tensorflow.

Are you satisfied with the resolution of your issue?
Yes
No

from tensorflow.