Dimensionality Reduction for Categorical Data

Categorical attributes are those that can take a discrete set of values, e.g., colours. This work is about compressing vectors over categorical to low-dimension vectors. The current hash-based methods do not provide any guarantee on the Hamming distances between compressed representations. Here we present FSketch create sketches for sparse data and an estimator estimate pairwise among uncompressed only from their sketches. We claim these be used in usual mining tasks place original without compromising quality task. For that, ensure also categorical, sparse, distance estimates reasonably precise. Both sketch construction estimation algorithms require just single-pass; furthermore, changes point incorporated into its efficient manner. compressibility depends upon how independent dimension -- making our algorithm attractive many real-life scenarios. Our claims backed by rigorous theoretical analysis properties supplemented extensive comparative evaluations with related some real-world datasets. show significantly faster, accuracy obtained using top standard unsupervised RMSE, clustering similarity search.