Why does shape of the input is empty?

def build_classifier_model():
    text_input = tf.keras.layers.Input(shape=(), dtype=tf.string, name='text')
    preprocessing_layer = hub.KerasLayer(tfhub_handle_preprocess, name='preprocessing')
    encoder_inputs = preprocessing_layer(text_input)
    encoder = hub.KerasLayer(tfhub_handle_encoder, trainable=True, name='BERT_encoder')
    outputs = encoder(encoder_inputs)
    net = outputs['pooled_output']
    net = tf.keras.layers.Dropout(0.1)(net)
    net = tf.keras.layers.Dense(1, activation=None, name='classifier')(net)
    return tf.keras.Model(text_input, net)

This is taken from Classifique o texto com BERT  |  Text  |  TensorFlow

What I think is because the string data is of varying length of tokens, and it will automatically made uniform by preprocessing layer + converted to numeric vector form, we just need to define an entry point (input layer) for the model then passthrough the data (data passed from input layer to first hidden layer is always unchanged).

1 Like

Hi @tbhaxor. I would think you have it right. text_input here is simple string. shape=() generally expects a scalar value or a 0-dimensional tensor. And as you said, by setting shape to shape=() the model accepts input strings of different lengths, as the BERT model can handle variable-length sequences of texts.

1 Like

Just to be clear, it is equivalent of shape=(None, ), and if we had lets say 2d text data, then shape=(None, None, ) would have been used.

Yes, I again guessed it right

1 Like