Apache Spark Deep Learning Cookbook

There's more...

As with everything, there are pros and cons to using transfer learning. As we discussed earlier on in the section, transfer learning is ideal when you are limited in resources to perform your own modeling on a large dataset. There is always the chance that the source data at hand does not exhibit many of the features unique to it in the pre-trained models leading to poor model performance. There is always the option to switch from one pre-trained model to another and evaluate model performance. Again, transfer learning is a fail fast approach that can be taken when other options are not available.