Hard vs. soft voting
The problem, from first principles
Once several classifiers are trained, you still have to turn their separate answers into one. Hard voting counts class votes. Soft voting averages predicted probabilities, so a confident model can outweigh two lukewarm ones.
From theory to code
Implement hard_vote(predictions) and soft_vote(probabilities, weights=None). Both return one predicted class per sample.
Constraints
predictionshas shape(n_models, n_samples)and holds integer class labels.probabilitieshas shape(n_models, n_samples, n_classes);weightsis an optional length-n_modelsarray.- On a hard-vote tie, return the smallest label. On a soft-vote tie, return the smallest class index.
- Return integer NumPy arrays of length
n_samples.
Hints
Open one at a time. Each gives away a little more than the last.
Hint 1
For hard voting, np.unique(column, return_counts=True) lists labels in sorted order, so np.argmax over the counts already breaks ties toward the smallest label.
Hint 2
For soft voting, average over the model axis first (weighted if weights are given, divided by their sum), then take argmax over the class axis.
Keywords
No results yet
Run your code to check it against the examples, or submit to run every test.