resource:

shared_resource:

resources_cf_stack_name: "BanditsSharedResourceStack" # cloud formation stack

experiment_db:

table_name: "BanditsExperimentTable" # Dynamo table for status of an experiment

model_db:

table_name: "BanditsModelTable" # Dynamo table for status of all models trained

join_db:

table_name: "BanditsJoinTable" # Dynamo table for status of all joining job for reward ingestion

iam_role:

role_name: "BanditsIAMRole"

private_resource:

hosting_fleet:

instance_type: "ml.t2.medium"

instance_count: 1

training_fleet:

instance_type: "ml.c5.4xlarge"

evaluation_fleet:

instance_type: "ml.c5.4xlarge"

image: "462105765813.dkr.ecr.{AWS_REGION}.amazonaws.com/sagemaker-rl-vw-container:vw-8.7.0-cpu" # Vowpal Wabbit container

algor: # Vowpal Wabbit algorithm parameters

algorithms_parameters:

exploration_policy: "cover" # supports "egreedy", "bag", "cover"

epsilon: 0.10 # percent to explore with egreedy exploration policy

num_policies: 3 # number of nested policies to create when using bag or cover exploration policy

num_arms: 2

cfa_type: "dr" # supports "dr", "ips"

local_mode: false # use local mode?

soft_deployment: true # use the same endpoint with updated model using a blue-green deployment?