search

hackingmaterials / matminer_examples

A repo of examples for the matminer (https://github.com/hackingmaterials/matminer) code
Other
103 stars 60 forks source link

Predicting bulk modulus - r^2 values #53

Open computron opened 5 years ago

computron commented 5 years ago

The bulk modulus notebook takes the absolute value of r^2 when computing the scoring metric. But, it should not do that - if r^2 is less than zero, it is worse than predicting the mean.

Probably it won't affect this specific notebook, but better not to have that code in there. Very simple fix - just remove np.abs() around any r2 computation.

computron commented 5 years ago

(note that negative r^2 would really not happen if you are fitting a model to a provided data set, but extrapolation of a prior model to a new data set can certainly be worse than just predicting the average of values in the new data set.)