I feel like I have followed all the provided instructions, but I’m still unable to make a successful submission.
I named the package in the Config class
and I have imported it in the appropriate place.
I have attempted uploading a zip file and including the packages in the requirements.txt file to no avail. I also attempted installing and loading the packages within each function as suggested in another discussion post. Am I missing an extra step or have I overlooked some instruction?
[EDIT]:
Here is the most recent error:
"DockerBuildError: The command '/bin/sh -c apt -qq update && apt -qq install -y xgboost && rm -rf /var/lib/apt/*' returned a non-zero code: 100"
Based on the error, it looks like you have APT_PACKAGES = ["xgboost"] in your Config class. Please make sure that you define your pip/python packages only in ADDITIONAL_PACKAGES and not APT_PACKAGES. (this happened with your last two submissions)
For your submission with ID #110256, the model is failing load with the error xgboost.core.XGBoostError: [18:40:46] ../src/learner.cc:922: Check failed: header == serialisation_header_. Can you confirm that your load_model function works for you on your colab notebook?
PS: When posting on public forums, please be careful about your API_KEY. This key will give others access to your account.
Thanks for getting back to me. I modified my Config class to include my python packages only in the ADDITIONAL_PACKAGES list and I ensured the load model function works (no errors):
Based on the logs it looks like the load_model part is still failing and this is due to different xgboost version being used during evaluation. Can you try using the save_model and load_model methods offered by xgboost instead of pickle?