Instance Space
(Redirected from item space)
Jump to navigation
Jump to search
An Instance Space is a Space of the all possible Instances for a Machine Learning Task.
- AKA: Object Space, Item Space, Example Space.
- Example(s):
- a geometric space where each dimension corresponds to each attribute in a attribute-value learning task.
- a metric space of measurable features of an instance in a Traveling Salesman Problem.
- …
- Counter-Example(s):
See: Algorihtm Footprint, Geometric Space, Euclidean space, Supervised Machine Learning Task, Unsupervised Machine Learning Task, Concept Learning Task.
References
2017
- (Sammut & Webb, 2017) ⇒ Claude Sammut, and Geoffrey I. Webb. (2017). "Instance Space." In: (Sammut & Webb, 2017).
- An instance space is the space of all possible instances for some learning task. In attribute-value learning, the instance space is often depicted as a geometric space, one dimension corresponding to each attribute.
2012
- (Smith-Miles & Tan, 2012) ⇒ Kate Smith-Miles, and Thomas T. Tan (2012, June). "Measuring algorithm footprints in instance space" (PDF). In Evolutionary Computation (CEC), 2012 IEEE Congress on (pp. 1-8). IEEE.
- QUOTE: But how can we create a metric space of instances so that we can readily visualize where they lie and their properties? We define the instance space to be the highdimensional space that summarizes a set of instances by a feature vector. The measurable features or properties of an instance of the TSP, for example, might include the cluster structure of the cities, and statistical properties of the distance matrix. The region of instance space where an algorithm can be expected to perform well has recently been described as an algorithm footprint [1], but the methodologies for defining, visualizing, and characterizing this footprint across the highdimensional instance space do not currently exist.
(...) Figure 2 shows the regions in instance space where each algorithm’s performance is “good” or “bad” (with “bad”meaning the tour length is greater than 1% of the optimal tour length).
.
- QUOTE: But how can we create a metric space of instances so that we can readily visualize where they lie and their properties? We define the instance space to be the highdimensional space that summarizes a set of instances by a feature vector. The measurable features or properties of an instance of the TSP, for example, might include the cluster structure of the cities, and statistical properties of the distance matrix. The region of instance space where an algorithm can be expected to perform well has recently been described as an algorithm footprint [1], but the methodologies for defining, visualizing, and characterizing this footprint across the highdimensional instance space do not currently exist.
- ↑ D. Corne and A. Reynolds, “Optimisation and Generalisation: Footprints in Instance Space,” Parallel Problem Solving from Nature, PPSN XI pp. 22-31, 2010.