One of ways to solve linearly inseparable problems is to allow some wrong classification, that way reducing precision, but at the same time allowing to separate most values. Slack variable idea can also be applied to other machine learning algorithms. Neural network could be called one huge slack variable, it just separates classes any way it likes and that is why it is so good with noisy data. In the diagram below the depicted dataset is similar to Iris, it has linearly separable groups as well as inseparable.
 |
| Figure 3. Slack variable (Source: Ubaby, 2016) |
No comments:
Post a Comment