Chapter 14 worked answers: Preparing for Machine Learning and Deep Learning

Compare your reasoning after an attempt. Different valid implementations are welcome when they preserve the observation set and the stated meaning.

B2-14-E01 · Evaluate a numeric baseline

Calculate MAE and RMSE in minutes and verify one error by hand.

Absolute errors are 2,4,7; squared errors are 4,16,49. Their means give the reported measures.

actual=np.array([9.,15.,18.]);prediction=np.array([11.,11.,11.])
mae=np.abs(actual-prediction).mean();rmse=np.sqrt(((actual-prediction)**2).mean())
assert np.isclose(mae,13/3) and np.isclose(rmse,np.sqrt(23))
print(mae,rmse)

4.333333333333333 4.795831523312719

B2-14-E02 · Reject preprocessing leakage

Learn the centering value from training only and transform evaluation with it.

The evaluation distribution must not determine the learned center used to assess the procedure.

train=np.array([10.,20.]);evaluation=np.array([100.,200.])
center=train.mean();transformed=evaluation-center
assert center==15 and transformed.tolist()==[85,185]
print(center,transformed)

15.0 [ 85. 185.]

B2-14-E03 · Align targets by ID

Construct y in the feature-row order and verify [2,1].

Equal lengths are insufficient; IDs establish the correspondence.

features=pd.DataFrame({"id":["b","a"],"x":[20,10]}).set_index("id")
target=pd.Series([1,2],index=["a","b"])
y=target.reindex(features.index).to_numpy()
assert y.tolist()==[2,1]
print(y)

[2 1]

Return to practice

Change one supplied input, predict the effect and rerun your own implementation. Return to the exercise prompts to identify a task you can now explain independently.

Chapter exercisesWorked answers after an attempt