Store result files to s3

[katxiao]

If the user passes cache_dir as an s3://<bucket-name>/path/to/dir the scores, synthetic data, and error files are stored in the corresponding path of the given bucket. If aws_key and aws_secret are provided, they will be used to authenticate the s3 requests.

✅ Ran with aws credentials and a private s3 bucket, and verified that the results files were uploaded.

Resolve #81

[csala]

I'm thinking about changing this approach and, instead of writing a method that handles all the different situations, just write a write_file(contents, path, aws_key, aws_secret) function that knows whether to write in S3 or in a local file depending on the path prefix.
Additionally, a write_csv could be added that just dumps the DataFrame as a CSV string and then calls write_file.

With that, we can keep the code almost as it was before, doing:

base_path = str(cache_dir / f'{name}_{dataset_name}_{iteration}_{run_id}')
if scores is not None:
    write_csv(scores, base_path + '_scores.csv')
if 'synthetic_data' in output:
    synthetic_data = compress_pickle.dumps(output['synthetic_data'])
    write_file(synthetic_data, base_path + '.data.gz')
if 'exception' in output:
    write_file(output['exception'],base_path + '_error.txt')

[csala]

I would rename this module to s3.py, as there is no other place where S3 is mentioned, so there is no ambiguity

[katxiao]

@csala I think you're right about writing text files with write mode w instead of wb. I updated it to special-case the gzip files to write mode wb.

[csala]

All looks good! Let's wait for the rest of the PRs to be merged to this, and then this can go in