Confusion Matrix Mod

A confusion matrix is a square matrix visualization which allows us to judge the performance of a classification model in supervised tasks where the ground truth label is known.

For example, if we train a model to classify 100 image instances into 10 classes, we can construct a confusion matrix to see how well our model is performing. The confusion matrix would be a 10x10 matrix where the cell in the j'th column and i'th row shows the relative number of times the classifier predicted an instance of class j as label i. "Relative" is used here because it is common to normalize the confusion matrix.

All source code for the mod example can be found in the src folder.

Data Requirements

1. Use a unique tire identifier such as rowid for the row is axis
2. Use the predicted and actual axes to select a categorical column

For example, say you have a machine learning model that it's trained to recognize written digit numbers from 0-9 from different tests. The actual are the written numbers by a person and the predicted is the output of the machine learning model algorithm. A small data set sample would look like this:

TestID,Actual, predicted
 01,0, 0
 02,0,8
 03,1,1
 04,2,2
 05,3,3
 06,4,9
 07,4,4
 08,5,2
 09,8,3
 10,9,6
 11,9,9
 12,9,9

Prerequisites

These instructions assume that you have Node.js (which includes npm) installed.

How to get started (with development server)

• Open a terminal at the location of this example.
• Run npm install. This will install necessary tools. Run this command only the first time you are building the mod and skip this step for any subsequent builds.
• Run npm run server. This will start a development server.
• Start editing, for example src/main.js.
• In Spotfire, follow the steps of creating a new mod and connecting to the development server.

Working without a development server

• In Spotfire, follow the steps of creating a new mod and then browse for, and point to, the manifest in the src folder.