This is a follow up to Tensorflow input of varying shapes.

In my particular situation, for each training sample, I have three different inputs. Two are scalars and the third is a sparse tensor.

The two scalars are each used to build tensors to use as state for LSTM. For instance:

hidden_state = tf.tile(self.state_init_values, [scalar_1, 1])
cell_state = tf.zeros([scalar_1, 100])

My confusion right now is mainly on how to get the value of these scalars. The tf.print fuction is capable of that, but I’m not sure how. Here is a simplified example:

class DataGenerator(tf.keras.utils.PyDataset):
    def __init__(self):
        pass
        
    def __getitem__(self, index):
        return tf.constant([0.1, 0.2, 0.3]), tf.constant([1, 1, 1])

    def __len__(self):
        return 1
        
class InfoModel(tf.keras.Model):
    def __init__(self):
        super(InfoModel, self).__init__()

    def call(self, inputs, training=False):
        #Prints: Tensor("info_model_9_1/strided_slice:0", shape=(), dtype=float32)
        #scalar_1 = inputs[0]

        #TypeError: int() argument must be a string, a bytes-like object or a real number, not 'SymbolicTensor'
        #scalar_1 = int(inputs[0])

        #AttributeError: 'SymbolicTensor' object has no attribute 'numpy'
        #scalar_1 = inputs[0].numpy()
        
        #print(scalar_1)

        #This works...
        tf.print(inputs[0])
        
        return tf.convert_to_tensor([1,1,1])

gen = DataGenerator()

model = InfoModel()
model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])
model.fit(gen, epochs=1, verbose=1)
model.summary()

Is there a way to get at this value? Or am I going about this the wrong way to accomplish what I want?