Efficient binary embedding of categorical data using BinSketch

In this work, we present a dimensionality reduction algorithm, aka. sketching, for categorical datasets. Our proposed sketching algorithm Cabin constructs low-dimensional binary sketches from high-dimensional vectors, and our distance estimation Cham computes close approximation of the Hamming between any two original vectors only their sketches. The minimum dimension required by to ensure good theoretically depends on sparsity data points—making it useful many real-life scenarios involving sparse We rigorous theoretical analysis approach supplement with extensive experiments several real-world sets, including one over million dimensions. show that duo is significantly fast accurate tasks such as $$\mathrm {RMSE}$$ , all-pair similarity, clustering when compared working full dataset other techniques.