text_classification_model_simple.py 文件源码

def remove_padding(self, input_text):
        # calculate max length of the input_text
        mask = tf.greater_equal(input_text, 0)  # true for words false for padding
        sequence_length = tf.reduce_sum(tf.cast(mask, tf.int32), 1)

        # truncate the input text to max length
        max_sequence_length = tf.reduce_max(sequence_length)
        input_text_length = tf.shape(input_text)[1]
        empty_padding_lenght = input_text_length - max_sequence_length
        input_text, _ = tf.split(input_text, [max_sequence_length, empty_padding_lenght], axis=1)

        return input_text, sequence_length