Reorganize core abstract function calls and simplify APIs

[AvinashBukkittu]

Reorganize core abstract function calls and simplify APIs:

• For processors:

  • Currently, few processors have default HParams and a few don't. I suggest the following to maintain consistency. This function should appear in base_processor but can have a default empty implementation.
    1. All processors will have the default_config method. If a processor has no parameters, this method can return {}
    2. If the user wishes to use default hyperparameters, the user should not be forced to supply it during the add_processor call to the pipeline. The pipeline should automatically use the default HParams during initialize() of that processor
    3. If the user wishes to use a non-default hyperparameter, then it can be passed through add_processor method
  • initialize function can have a default implementation and only be overwritten when needed

• For readers:

  • cache_key_function can have a default implementation of returning None (or set the do_caching flag to false), so at this time we don't do caching.

• For Hparams:

  • We should accept user passing configs as plain Python Dict and auto-box them to HParams.

[gpengzhi]

We might also allow dict as input instead of Hparams when setting config.

[gpengzhi]

I will work on the hparams initialization part. @AvinashBukkittu @hunterhector

[gpengzhi]

Reopen this issue because only part of the issues is resolved.

[hunterhector]

The issue regarding default_config is solved. default_config now is a class method that can be overwritten. initialize is also to be implemented when needed. The remaining problem is cache_key_function