AWS H2O Inference Server - Empty Results

Question

I am attempting to deploy my H2O AutoML model to AWS following the instructions here.

I keep getting the following empty result when I curl my endpoint:

curl http://<myIP>:8080/model?type=1\&row=value1,value2,value3

{"result":""}

If I don't submit any value for row, I get results:

curl http://<myIP>:8080/model?type=1\&row=

{"result":"Prediction = 0.723482072353363, 0.17580199241638184, 0.04721980169415474, 0.053496140986680984"}

If I submit a single value for row, I get the same results:

curl http://<myIP>:8080/model?type=1\&row=value1

{"result":"Prediction = 0.723482072353363, 0.17580199241638184, 0.04721980169415474, 0.053496140986680984"}

I have connected to the instance via SSH and confirmed my model has been downloaded to the correct location on boot.

Can anyone help me figure out how I'm supposed to be submitting row data to my endpoint? Thanks!

Answer 1

After looking through the EC2 instance system log I discovered an error was being logged indicating an invalid value being given for a categorical column.

Upon further investigation I discovered the columns in my row query parameter were NOT being interpreted in the same order as they appeared when I trained my model.

After opening my MOJO model's ZIP file and looking at the model.ini file within, I was able to determine the order of the columns as the model was expecting.

Supplying columns in the expected order yields valid prediction results.