Adaptive Sampling Methods for Scaling Up Knowledge Discovery Algorithms

Abstract

Scalability is a key requirement for any KDD and data mining algorithm, and one of the biggest research challenges is to develop methods that allow to use large amounts of data. One possible approach for dealing with huge amounts of data is to take a random sample and do data mining on it, since for many data mining applications approximate answers are acceptable. However, as argued by several researchers, random sampling is difficult to use due to the difficulty of determining an appropriate sample size. In this paper, we take a sequential sampling approach for solving this difficulty, and propose an adaptive sampling algorithm that solves a general problem covering many problems arising in applications of discovery science. The algorithm obtains examples sequentially in an on-line fashion, and it determines from the obtained examples whether it has already seen a large enough number of examples. Thus, sample size is not fixed a priori; instead, it adaptively depends on the situation. Due to this adaptiveness, if we are not in a worst case situation as fortunately happens in many practical applications, then we can solve the problem with a number of examples much smaller than the required in the worst case. For illustrating the generality of our approach, we also describe how different instantiations of it can be applied to scale up knowledge discovery problems that appear in several areas.

Parallel mining of maximal sequential patterns using multiple samples

“…However, they tested this method with only one candidate itemset. Thus, as suggested in [8,9], more works need to be done to combine this adaptive sampling method with some mining algorithms, such as Apriori, and more experiments are required to test the cost of on-line sampling.…”

“…MSPX can avoid or alleviate some problems inherent in the traditional single-sample methods [5,8,9,12,17,20,21]: (1) The performance of single-sample methods usually varies considerably from one run to another for the same mining task, because a bad sample can degrade the overall performance of the mining. On the other hand, by using multiple samples, MSPX effectively prevents the candidate generation from the overestimates made by a bad sample, so its performance is much more stable.…”

“…In [8,9], they proposed how to dynamically determine the sample size based on the estimated support of the candidate itemsets, which is expected to be tighter than the sample size based on the Chernoff boundary. The basic idea is that a relatively small sample can be used if the support of a candidate itemset is far from the minimum support.…”

“…Sampling has been used for mining frequent itemsets or sequences [5,8,9,12,17,20,21]. In [5], the FAST algorithm progressively refines the initial sample to obtain a small final sample and reports the set of frequent itemsets in the final sample as the result.…”

Adaptive data reduction for large-scale transaction data

“…A variety of procedures for selecting subsets from a large dataset are studied in [6]; and the results of using different techniques are empirically compared. A sequential sampling method for determining appropriate sample sizes for data reduction is proposed in [7]. The afore-mentioned data reduction approaches are mainly based on statistical sampling techniques, such as simple random sampling, stratified sampling or cluster sampling.…”