randomness of audit data set

[rickardbrannvall]

Issue

Problem Description

I noticed that indices included in audit_dataset differ between runs even if random seeds are kept fixed.

Expected Behavior

Same audit dataset should be generated for fixed random seeds (and sizes etc).

What Needs to be Done

Perhaps random number generators (streams) need to be global and used explicitly in all random functions.

How Can It Be Tested or Reproduced

Log audit dataset indices between two runs and check for agreement.

[rickardbrannvall]

Problem may be related to code in data_preparation.py in dev_utils function prepare_train_test_dataset.