2

Note: This is not an issue related to scikit-learn package.

Hi,

I tried deploying a toy model using scikit-learn on Clipper cluster using DockerContainerManager.

It's successfully able to build the docker image but gets stuck on

18-08-05:18:37:04 INFO [docker_container_manager.py:353] [default-cluster] Found 0 replicas for tree-model:1. Adding 1

On inspecting the docker container, first it exits and on checking the logs of the respective container I get:

Starting Python Closure container Connecting to Clipper with default port: 7000 Encountered an ImportError when running container. You can use the pkgs_to_install argument when calling clipper_admin.build_model() to supply any needed Python packages.

I think this is a common error, even others encountered it: https://github.com/ShuaiW/serve-ml-with-clipper/issues/3

Any help would be really appreciated!

Aniruddh Khera
  • 111
  • 1
  • 13
  • 1
    The error was due to the scikit dependencies, numpy, scipy, which is required to be specified explicitly in pkgs_to_install before 'scikit-learn' . – Aniruddh Khera Aug 06 '18 at 15:07
  • @AniruddhKera I am having the same issue, although I have specified both `numpy` and `scipy` in `pkgs_to_install`. – Simone Dec 04 '20 at 19:58
  • 1
    @Simone did you try installing numpy, scipy before scikit-learn? Also, check their versions are compatible with the scikit-learn version you are using. – Aniruddh Khera Dec 07 '20 at 17:10
  • @AniruddhKera Thanks. I have tried with the most recent versions of `numpy`, `scipy`, `scikit-learn`, and `pandas`, also I have update `pip` (in the `Docker` image), but I am getting the same error. – Simone Dec 13 '20 at 18:01

0 Answers0