Multidimensional Detective Alfred Inselberg Presented By Rajiv Gandhi and Girish Kumar Motivation • Discovering relations among variables • Displaying these relations Cartesian vs. Parallel Coordinates • Cartesian Coordinates: – All axes are mutually perpendicular • Parallel Coordinates: – All axes are parallel to one another – Equally spaced An Example Cartesian Parallel Representation of a 2-D line Why Parallel Coordinates ? • Help represent lines and planes in > 3 D Representation of (-5, 3, 4, -2, 0, 1) Why Parallel Coordinates ? (contd..) • Easily extend to higher dimensions (1,1,0) Why Parallel Coordinates ? (contd..) Cartesian Parallel Representation of a 4-D HyperCube Why Parallel Coordinates ? (contd..) X9 Representation of a 9-D HyperCube Why Parallel Coordinates ? (contd..) Representation of a Circle and a sphere Multidimensional Detective Our Favorite Sentence “The display of multivariate datasets in parallel coordinates transforms the search for relations among the variables into a 2D pattern recognition problem” Discovery Process • Multivariate datasets • Discover relevant relations among variables An Example • Production data of 473 batches of a VLSI chip • Measurements of 16 parameters - X1,..,X16 • Objective – Raise the yield X1 – Maintain high quality X2 • Belief: Defects hindered yield and quality. Is it true? The Full Dataset X1 is normal about its median X2 is bipolar Example (contd..) • Batches high in yield, X1 and quality, X2 • Batches with low X3 values not included in selected subset Example (contd..) • Batches with zero defect in 9 out of 10 defect types • All have poor yields and low quality Example (contd..) • Batches with zero defect in 8 out of 10 defect types • Process is more sensitive to variations in X6 than other defects Example (contd..) • Isolate batch with the highest yield • X3 and X6 are non-zero • Defects of types X3 and X6 are essential for high yield and quality Critique • Strengths – Low representational complexity – Discovery process well explained – Use of parallel coordinates is very effective • Weaknesses – Does not explain how axes permutation affects the discovery process – Requires considerable ingenuity – Display of relations not well explained – References not properly cited Related Work • InfoCrystal [Anslem Spoerri] – Visualizes all possible relationships among N concepts – Example: Get documents related to visual query languages for retrieving information concerning human factors Example Automated Multidimensional Detective • Automates discovery process • details not very clear