DeepChem Common Errors

When using DeepChem, there are some classes of common errors I've observed that don't have great error messages. I'm starting this thread up for us to start documenting some patterns which hopefully can lead to improved error messages:

- **Using the wrong featurization:** When copy/pasting example code, you'll often do things like using ECFP featurizations with graph convs which will yield a very confusing TensorFlow error. This ideally should have a natural error message suggesting the user check their featurization.
- **Padding:** The generators in KerasModel pad by default. When using custom models, this often confuses newcomers since they enter in a dataset of say 100 elements with batch size 64 and are very surprised to receive back 128 predictions. Some warning should be emitted along with a suggested fix

I'll add on more errors here as I come across them. Please add on your own observed confusing DeepChem behavior!