Data for benchmarking machine learning models, taken from Better than classical? The subtle art of benchmarking quantum machine learning models. The Linearly Separable task can be seen as a ”fruit-fly'' example for classification. It is straightforward and well-understood in the field. Even in the early days of artificial intelligence research, investigators already knew that linearly separable classification tasks can be learned by a simple perceptron model.
Description of the dataset
The data collection consists of 19 individual datasets with 300 samples each that vary in dimension from d=2d=2 to d=20d=20. The samples are points in dd-dimensional space and can be separated into their classes by a hyperplane.
Each dataset is generated by sampling inputs uniformly from a dd-dimensional hypercube. The inputs are divided into two classes by the hyperplane orthogonal to the (1,…,1)T(1,…,1)T vector.
There is a data-free margin ΔΔ around the hyperplane which guarantees that all datapoints xx fulfill ∣xw∣>δ∣xw∣>δ. The size of the margin grows with the dimension as Δ=0.02dΔ=0.02d.
Additional details
Source code
tab to check how the data was generated.Example usage
[ds] = qml.data.load("other", name="linearly-separable")
ds.train['4']['inputs'] # points in 4-dimensional space
ds.train['4']['labels'] # labels for the points above
Joseph Bowles, Shahnawaz Ahmed, Maria Schuld
version 0.1 : initial public release
Maria Schuld
Dedicated to making quantum machine learning a reality one day.
Joseph Bowles
Quantum Machine Learning researcher at Xanadu
Shahnawaz Ahmed
Code. Quantum. ML