Support for TF 2.0

[SivamPillai]

Hi,

Are there any plan to upgrade the library to support TF 2.0? I feel some of the new features can help simplify the code library and may also improve performance.

Thanks,
Sivam.

[Romaixn]

+1 That would be cool! And the library could work like this, for the moment tf.contrib doesn't exist

[xavierad]

I solved it by using:

    # vocab_words = tf.contrib.lookup.index_table_from_file(
    #     params['words'], num_oov_buckets=params['num_oov_buckets'])
    num_oov_buckets = params['num_oov_buckets']
    vocab_words = tf.lookup.StaticVocabularyTable(
        tf.lookup.TextFileInitializer(
            params['words'],
            key_dtype=tf.string, key_index=tf.lookup.TextFileIndex.WHOLE_LINE,
            value_dtype=tf.int64, value_index=tf.lookup.TextFileIndex.LINE_NUMBER,
            delimiter="\t"),
        num_oov_buckets)

    # vocab_chars = tf.contrib.lookup.index_table_from_file(
    #     params['chars'], num_oov_buckets=params['num_oov_buckets'])
    vocab_chars = tf.lookup.StaticVocabularyTable(
        tf.lookup.TextFileInitializer(
            params['chars'],
            key_dtype=tf.string, key_index=tf.lookup.TextFileIndex.WHOLE_LINE,
            value_dtype=tf.int64, value_index=tf.lookup.TextFileIndex.LINE_NUMBER,
            delimiter="\t"),
        num_oov_buckets)

[JamesPiggott]

Dear xavierad, I cannot imagine that that is the only alteration you made. I get error messages that LSTMBlockFusedCell is no longer supported with TF 2.0 which is confirmed online.

[xavierad]

Yes. I just copied the class code to local. Tensorflow sucks

[JamesPiggott]

Interesting, could you push those changes to your fork?

[iprovalo]

Here is my fork with TF2 implementation of one of the models:

https://github.com/iprovalo/tf_ner/tree/master/models/chars_conv_lstm_crf_tf2