In QML_Tutorial.ipynb, I found two errors.

1. Same training and test data sets are used.
2. In exercise 4, the representations (X_test) do not correspond to the test atomization energies (Y_test), which yields a large mean-absolute-error (MAE).

To solve these issues, the following two lines in the fourth cell can be removed:
# Run on only a subset of the first 1000 (for speed)
X = X[:1000]