In Blueoil, we can train with GCS input and save training results to GCS, but other operations (e.g. convert, evaluate...) are not supported with GCS input/output.
Issues
There are many files and operations that depend on input and output directly.
We can replace them to support GCS with tensorflow.io.gfile, but some operations need to rewrite several lines. So it is better to add utils of operations of I/O and manage them in a single module.
Targets
There are operations with files/directories directly.
Since it is difficult to manage all of these operations in individual files and manage them separately, it is better to provide a module that aggregates the operations that perform file I/O and change it to use this module for all except special cases.
In Blueoil, we can train with GCS input and save training results to GCS, but other operations (e.g. convert, evaluate...) are not supported with GCS input/output.
Issues
There are many files and operations that depend on input and output directly.
and so on...
We can replace them to support GCS with tensorflow.io.gfile, but some operations need to rewrite several lines. So it is better to add utils of operations of I/O and manage them in a single module.
Targets
There are operations with files/directories directly.
Since it is difficult to manage all of these operations in individual files and manage them separately, it is better to provide a module that aggregates the operations that perform file I/O and change it to use this module for all except special cases.