$ make create-container
$ make
$ make jupyter
$ make create-image
$ make create-container
$ make
$ make jupyter
according to [https://dev.classmethod.jp/machine-learning/bert-text-embedding/]
- access to [http://projector.tensorflow.org/]
- upload; push load button and choose output.txt an input.txt
- validate with t-SNE.
I almost write down work log to Dockerfile.
Some fixies to bert.
tf_upgrade_v2
- adapt tf2;
- tf.flags -> tf.compat.v1.flags
- tf.contrib.layers.layer_norm -> tf.keras.layers.LayerNormalization
- change tokenizer to use Juman++