Endpoints test inference gives error 415

Hi and thanks for reaching out,

Today, the endpoint test UI in SageMaker Studio only supports application/json content, so can't be used with endpoints like yours that don't support this content type: You'll need to use code, for example from a notebook or terminal.

From Python in particular you would have two options:

1. The low-level boto3 SDK as used here, which maps directly to the underlying InvokeEndpoint API and requires you to serialize your content (e.g. to CSV string) yourself, or
2. The high-level sagemaker SDK, which might be more familiar if you've followed tutorials with e.g. the Estimator / Predictor syntax.

Either is fine really... The high-level SDK just packages the process of stringifying and de-stringifying objects for the HTTP(S) request using the serializer and deserializer classes - so you would set your predictor up with the appropriate de/serializer and then just pass in the data as your usual e.g. DataFrame or nested list format:

import pandas as pd
import sagemaker

predictor = sagemaker.Predictor(
    "your-endpoint-name",
    serializer=sagemaker.serializers.CSVSerializer(),
    deserializer=sagemaker.deserializers.CSVDeserializer(),
)

df = pd.read_csv("small-dataframe.csv")
result = predictor.predict(df)

Alternatively if having the test UI is important for your use-case, you could explore other SageMaker algorithms that support JSON I/O? If you're using XGBoost you could consider using it in "framework mode" instead of "algorithm mode", letting you provide a custom inference script to parse JSON or whatever other request formats you'd like. If you're using AutoGluon-Tabular or one of the other "JumpStart-based" algorithms, you may be able to download, modify, and re-upload the inference deploy_source_uri (see this example) to likewise accept custom request/response types.

Thanks @Alex for update. It will be good if AWS test endpoint UI also support another types as well. As every new developer/learner starts with Abalone example, so they are expecting result on UI interface as well. This is the very basic and initial requirement of AWS test endpoint UI.

Thanks again.