The VGG16_model name for a bottleneck model is dangerous

[kitsnet]

If someone reuses the same naming pattern for their ResNet50 bottleneck model, they are likely to overwrite the existing (non-bottleneck) ResNet50_model, breaking the code of dog_detector().

It's better to replace the name with something more telling (and less confusing), like VGG16_dog_model or VGG16_bottleneck_model.

[nblumoe]

Thanks for reporting this (also on the AIND Slack)! I had issues with that naming convention, too: I also used ResNet50 for the final classifier but named my bottleneck model Resnet50_model. Using auto-completion in a few places then mixed up the two different models (existing ResNet50_model and my Resnet50_model) and caused a hard to find bug in using the trained bottleneck model.

This isn't really a bug in the dog-project but indeed the naming convention could be improved to avoid issues we ran into.