[SPARK-8998] Collect enough frequent prefixes before projection in PrefixSpan

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.5.0
• Fix Version/s: 1.5.0
• Component/s: MLlib
• Target Version/s: 1.5.0
• Assignee: Zhang JiaJin
• Reporter: Xiangrui Meng
• Shepherd: Xiangrui Meng
• Created: 11/Jul/15 04:17
• Updated: 30/Jul/15 15:14
• Resolved: 30/Jul/15 15:14

Description

The implementation in SPARK-6487 might have scalability issues when the number of frequent items is very small. In this case, we can generate candidate sets of higher orders using Apriori-like algorithms and count them, until we collect enough prefixes.

Issue Links

• blocks: SPARK-8999 Support non-temporal sequence in PrefixSpan
• depends upon: SPARK-6487 Add sequential pattern mining algorithm PrefixSpan to Spark MLlib
• links to: [Github] Pull Request #7383 (zhangjiajin)
• links to: [Github] Pull Request #7412 (zhangjiajin)
• links to: [Github] Pull Request #7783 (feynmanliang)