AML Wrapper for ddm_predictor.py

[akzaidi]

Soundtrack

• EABS - Raise Your Hearts, Drop Your Guns

Description

Updates the ddm_predictor.py script to utilize Ray for training your policy rather than Bonsai.

Usage

• Run using Ray-on-AML with: python ddm_predictor.py simulator.policy=ray-aml.
• Run locally with Ray with: python ddm_predictor.py simulator.policy=ray-local.

Fixes # (issue): NA

Type of change

Please delete options that are not relevant.

• New feature (non-breaking change which adds functionality)
• This change requires a documentation update

How Has This Been Tested?

Please describe the tests that you ran to verify your changes. Provide instructions so we can reproduce. Please also list any relevant details for your test configuration

• Test A
• Test B

Checklist:

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• Any dependent changes have been merged and published in downstream modules

[akzaidi]

@juanvergaramunoz - I'd like to make one more edit today that will make things a bit easier, which is to convert the DDM repo into a package that we can pip install. That will make things easier for ray-on-aml with installing the ddm scripts and ensures they are in the correct location when running the remote jobs. I still wanted to publish this so you can test this, but perhaps you can focus on the test local path before trying the longer running tests on AML?

[thisisvergara]

@akzaidi, did you see the pytest failed? I think we need to setup numpy version to be >=1.16 & <1.20 -- and it seems currently we ask for 1.22.0

On a different note, if you are close to pushing the commit with the change you indicated above, I will hold my review. Otherwise, I will review what is already there.

[thisisvergara]

Numpy error identified that causes verbose logging to the point of storage overflow. Adding the object type to the creation of the numpy input data array in L462 (X = np.array(input_array, dtype=object).reshape(1, -1)), removed the following verbose error:

ddm_predictor.py:462: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray.
   X = np.array(input_array).reshape(1, -1)

Removal of that message causes an output file size reduction of x9. Reducing the original output log file from 12GB to only 1.3GB

Additionally, removal of a print that indicated "No signal builder used" further reduced the output log file by another factor of x3.