Tensorflow saving weights and indexes

General Discussion
#1

When saving weights or different types of models in an output channel of a trainer, how do I differentiate between these in a resolver to prevent picking up the wrong model in the next run of a TFX pipeline run?

#3

@Robert_Crowe grateful for any pointers

#5

Hi James,

See this example to use a Resolver: tfx/penguin_pipeline_local.py at v1.12.0 · tensorflow/tfx · GitHub

There are certain resolve policies we have defined, eg. takes in the latest artifact.

However, we don’t have a resolve policy to resolve based on the model weights .

you can:

  1. implement the custom resolving policy.
  2. alter the way to write the pipeline and use conditional expression to control the pipeline logic to be decided by parameters related to your models.

I think 2 might be easier to pursue.