15 Takeaways

This document explains associations in detail. We finish it with a list of key points:

• Association Discovery (or associations) finds meaningful relationships among fields and their values in high-dimensional datasets, whereas statistical techniques focus on controlling the risk of making false discoveries.

• Associations output is easily expressed as rules that can be understood by non-experts.

• You can create associations from datasets that have been created in BigML, and then create a new dataset from the association rules that you discover. (See Figure 15.1 .)

• Associations require the data to be structured in a specific way, using the items field type.

• You can create an association with just 1-click or configure it as you wish.

• There is no single measure (Support, Coverage, Confidence (Associations), Leverage, or Lift) that is always more important than others. This will depend on your main goals.

• You can set minimum levels for a number of association measures that let you focus on more interesting association rules, while filtering out potentially spurious ones.

• You can control multiple interestingness measures, yet easy to tune without having to configure difficult to comprehend parameters.

• You can easily discretize your numeric fields to transform them into categorical fields.

• BigML lets you create associations for a sample of your dataset.

• After associations are created, you will get a table that summarizes all the rules discovered, and you can visualize these rules in a network chart.

• You can download your association rules in a CSV file, and export the network chart as an image.

• You can programmatically create, list, delete, and use your associations through the BigML API and the BigML bindings.

• You can furnish your associations with descriptive information (name, description, tags, and category).

• You can stop an association creation before the Task is finished.

• You can permanently delete an association.