Where to read the learned model

In model_fn in model_fn_builder python

if init_checkpoint:
  (assignment_map, initialized_variable_names
   ) = modeling.get_assignment_map_from_checkpoint(tvars, init_checkpoint)

I’m doing it.

• tvars = tf.trainable_variables()
• init_vars = tf.train.list_variables(init_checkpoint)
• Writing name→name mapping to assignment_map for each element in init_vars that is included in tvars
  • The comment says """Compute the union of the current variables and checkpoint variables."" Isn’t that Intersect, not Union?
• Why are we doing this?
  • This allows you to read both Pretrained and Fine-tuned models without worrying about

Then there are some branches where you have to put an adapter for the TPU, but basically python

tf.train.init_from_checkpoint(init_checkpoint, assignment_map)

• tf.compat.v1.train.init_from_checkpoint | TensorFlow Core r2.0

  • The first argument is a file name, etc.
  • assignment_map describes which variables map to which variables

• initialized_variable_names seems to be used only for log display

---

This page is auto-translated from /nishio/学習済みモデルはどこで読むか using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.