Skip to content

QuantumScope v0.9.13

Choose a tag to compare

@Onimee58 Onimee58 released this 15 Nov 06:56

QuantumScope v0.9.13: Choosing features by annealing

Quantum annealing, put to a real use. A new problem, Feature selection (from data), uses annealing to choose which features of a dataset a classifier keeps, one per qubit. The choice favors features that separate the classes well on their own, and avoids features that repeat each other.

  • Any data: a built-in dataset, or your own CSV through the Datasets tab. Only the training samples are used to choose, so test accuracy stays honest.
  • Your settings: how many features to keep, how much relevance counts against redundancy, and up to 10 candidate features.
  • Many reads: the anneal is run 100 times and the best choice is kept, as real annealers are used.
  • Chosen features tab: shows how relevant each candidate is, which candidates repeat each other, and how the annealed choice compares with the most relevant features alone, PCA and all candidates on test data.
  • Works both ways: send the chosen features to the Datasets tab, where they replace PCA for both classifiers, or start from the Datasets tab with Choose by annealing…. Saved models remember which features they use.

Learn: a new lecture, Choosing features by annealing, and a lab that keeps 3 of the 30 breast-cancer measurements and trains a quantum classifier on them. Both explain honestly what annealing adds at this size: with 10 candidates, a classical computer can check every choice directly.