al_egal: Implementation of Exploration Guided Active Learning (EGAL)
In e-sensing/activelearning: Active Learning Satellite Image Time Series

Description Usage Arguments Value Author(s) References

Active Learning improves the results of a classification by feeding the classifier with informative samples. The Exploration Guided Active Learning (EGAL) ranks a set of samples based on their density and diversity; those samples with a larger EGAL metric should be submitted first to a human expert (the oracle) for classification.

This function receives a sits tibble with time series samples and it computes the EGAL metric on it. However, this function doesn't guarantee the order of the returned samples.

al_egal(
  samples_tb,
  sim_method = "correlation",
  alpha = NULL,
  beta = alpha,
  w = 0.5
)

`samples_tb`	A sits tibble with both labelled and unlabelled samples (i.e. NA).
`sim_method`	A character. A method for computing the similarity among samples. See proxy::simil for details.
`alpha`	A double. It controls the radius of the neighborhood used in the estimation of sample density.
`beta`	A double. It controls the radius of the neighborhood used in the estimation of the sample candidate set. A bigger beta gives a bigger set. By default is set to be equal to alpha. If NULL, it is computed internally.
`w`	A numeric (between 0 and 1) only used when beta is NULL. This proportion parameter balances the influence of diversity and density in the selection strategy. When w is 0, EGAL becomes a pure-diversity and when w is 1, EGAL becomes a pure density-based sampling algorithm.

           A sits tibble with the EGAL metric. This metric
                   ranks samples based on their density and diversity.
                   Those samples with highest EGAL should be selected
                   first for labeling.

Alber Sanchez, alber.ipia@inpe.br

Hu, R., Jane Delany, S., & Mac Namee, B. (2010). EGAL: Exploration Guided Active Learning for TCBR. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics): Vol. 6176 LNAI (pp. 156–170). doi: 10.1007/978-3-642-14274-1_13

e-sensing/activelearning documentation built on Dec. 20, 2021, 2:21 a.m.