remove AWS_REGION from kiosk-training definition

[willgraf]

The tensorflow file-system can natively use the cloud bucket filesystems with the protocol prefix (e.g. s3://bucket/key). For AWS, the S3 buckets are globally accessible (names must be unique across all regions).

The kiosk is deployed in AWS_REGION=us-west-2. This environment variable is accessible for the training jobs to use. However, this value of AWS_REGION causes the following error:

Encountered Unknown AWSError
PermanentRedirect
The bucket you are attempting to access must be addressed using the specified endpoint. Please send all future requests to this endpoint.:

This error is resolved by changing AWS_REGION=us-east-1, the default region value. Using us-east-1, training works as expected. I expect this is caused by the global nature of buckets, and using the "global" protocol value of s3://bucket/key clashes with the non-default AWS_REGION value.

I'm unsure of the ultimate root cause, but setting the AWS_REGION for the kiosk-training to the default region resolves the AWS 301 error seen kiosk-training: fixes vanvalenlab/kiosk-training#5.

[willgraf]

I've just removed AWS_REGION entirely, as the tf default is us-east-1.

[dylanbannon]

I saw a demonstration that this change leads to successful training.