Project 3 Summary

While the grades were not as quite high as the previous two projects, overall the class still did a great job on Project 3!

Grades

20 grades were 20/25 or higher!

  • 25-25+: (27, 27),(25,25,25,25,25,25)

  • 23-24: (24,24,24,24,24,24) (23,23)

  • 21-22: (22,22,22,21)

Leader Board

  • 1st place, +2 bonus: 1.0 accuracy – one group

  • 2nd place, +1 bonus: 0.982 accuracy – one group

  • 3rd place: 0.9649 accuracy – one group

  • 4th place: 0.9509 accuracy – three groups

Common Issues

Analysis:

  • Incorrect Architectures (e.g., wrong number of convultion/pooling layers in the Alternate Lenet 5 architecture)

  • Train/test split – trying to use the sklearn function (impressive, but keep in mind memory issues)

  • Wrong number of images in the directory

Inference servers: By far, the most issues were with the inference servers.

  • Docker image not pushed to Docker Hub

  • Docker images did not build.

  • HTTP inference server got low accuracy

Report:

  • Readme not provided with instructions to start inference server.

  • Missing report