Project 3 Summary

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

Grades

19 grades were 21/25 or higher!

  • 25+: (27, 27), (25.5, 25.5), (25, 25), 25

  • 23-24.5: (24.5, 24.5), (24, 24), 23.5,

  • 21-22.5: 22.5, (22.5, 22.5), (22, 22), (21.5, 21.5),

Leader Board

  • 1st place, +2 bonus: 0.97076 accuracy – two groups!

  • 2nd place, +1 bonus: 0.95906 accuracy

  • 3rd place: 0.953216 accuracy – two groups!

  • 4th place: 0.935672 accuracy

  • 5th place: 0.812865 accuracy

  • 6th place: 0.619883 accuracy

Common Issues

Analysis:

  • Incorrect Architectures (e.g., wrong number of convultion/pooling layers)

  • Insufficient number of epochs (e.g., 5); usually you will want to do at least 10.

  • 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 POST endpoint under-specified, got various issues (wrong shape, etc) when trying to call it

  • HTTP inference server got low accuracy

Report:

  • Didn’t explain the choices made for the model. Looking for things like “average pooling vs max pooling”, “number of epochs”, etc.