single item inference request fails

[aleksandersumowski]

This project is marked as deprecated, yet still seems to be used for sklearn model hosting in Sagemaker. The default_input_fn when parsing csv input calls squeeze on the parsed result:

sagemaker-containers/src/sagemaker_containers/_encoders.py

Line 117 in 526dda9

array = np.array([row for row in reader]).squeeze()

That causes inferences with just single item to fail as the input array becomes 1D instead of 2D

[laurenyu]

we are working with the team that owns the scikit-learn images to have them migrate to https://github.com/aws/sagemaker-inference-toolkit

The updated implementation looks to not make this squeeze() call: https://github.com/aws/sagemaker-inference-toolkit/blob/master/src/sagemaker_inference/decoder.py#L43-L57

[aleksandersumowski]

Thanks for the response! Is there any timeline for the migration?

I guess it's still useful to keep the issue open for the time being in case someone has a similar issue in the meantime

[laurenyu]

Is there any timeline for the migration?

Sorry, I've reached out to the team, but don't have any timeline to share at this time.