how to provide predictor the endpoint name?

[yolanda0202]

Hi
I have the model in the existing endpoint, and I know if I want to apply this model to the new data, I need to create a Predictor object and provide it the endpoint name.
So here is the question...
How to provide the endpoint name? Could you show me the query I can use?
I tried
kmeans_predictor = predictor("kmeans-2018-05-02-16-48-42-652")
but doesn't work.

Thank you!

[laurenyu]

You can create a provide the endpoint name through the Predictor's constructor, e.g.

from sagemaker.tensorflow import TensorFlowPredictor

predictor = TensorFlowPredictor('myexistingendpoint')
result = predictor.predict(['my request body'])

For more, see https://github.com/aws/sagemaker-python-sdk#how-do-i-make-predictions-against-an-existing-endpoint

[yolanda0202]

@laurenyu Thanks for your reply. But I wonder how to do it without tensorflow, like the first method metioned in the Link
Create a Predictor object and provide it your endpoint name. Then, simply call its predict() method with your input.

[laurenyu]

I see. In that case, you should be able to use the RealTimePredictor class: https://sagemaker.readthedocs.io/en/stable/predictors.html

[yolanda0202]

Got the error. Anything else I need to set up?@laurenyu sorry I'm new to Python

[laurenyu]

the class keyword in Python is for defining a new class. The error you're seeing indicates that there's an issue with how the words are being used.

Instead, you want to import the class, and then initialize the class:

from sagemaker.predictor import RealTimePredictor

predictor = RealTimePredictor('kmeans-2018-05-02-16-48-42-652')

does that make sense?

[yolanda0202]

@laurenyu Thank you so much for your explanation! Now I got another bug.

When I'm in the notebook of training the model, it works for me to predict with the pred_set dataset, whose format is exactly as same as the training dataset

kmeans_predictor = kmeans.deploy(initial_instance_count=1,
instance_type='ml.m4.xlarge')

result = kmeans_predictor.predict(pred_set)

But when I called the endpoint in another notebook and trying to predict with the same pred_set dataset

from sagemaker.predictor import RealTimePredictor
predictor = RealTimePredictor('kmeans-2018-05-02-16-48-42-652')
predictor.predict(pred_set)

I got
ParamValidationError: Parameter validation failed:
Invalid type for parameter Body, value: [[ 6. 0. 1. 1. 20.]
[ 12. 0. 7. 1. 13.]
[ 1. 0. 1. 1. 0.]
...
[ 9. 0. 1. 1. 26.]
[ 10. 0. 2. 1. 19.]
[283. 0. 112. 93. 0.]], type: <class 'numpy.ndarray'>, valid types: <class 'bytes'>, <class 'bytearray'>, file-like object

[laurenyu]

That is because the RealTimePredictor doesn't default to a specific content type, while the dataset you're using is a numpy array. You can pass a serializer and deserializer to allow the Predictor to handle numpy arrays.

I probably should've suggested this first instead of the more general case, but you probably can also just use the KMeansPredictor class: https://sagemaker.readthedocs.io/en/stable/kmeans.html#sagemaker.KMeansPredictor

[yolanda0202]

problem solved! really appreciate your help! @laurenyu

[laurenyu]

good to hear, @yolanda0202 - happy to help!

[githubhnw]

Hello. I have a similar issue. I'm using a HyperParameterTuner for a Linear Learner model from Sagemaker. And I train it on data stored as record set in S3. When I call predict on a numpy ndarray of test data it gives me the same error as @yolanda0202 . I tried setting the content_type of the RealTimePredictor to 'application/x-recordio-protobuf' and use the test set stored as record set in S3 but it has an error related to batch_size from the logs. My issue is: what serialize and/or deserializer, content_type and accept arguments should be when training and wanting to predict on record sets ? Regards