In tidyclust: A Common API to Clustering

#| child: aaa.Rmd
#| include: false

r descr_models("mean_shift", "LPCM")

Tuning Parameters

#| label: LPCM-param-info
#| echo: false
defaults <-
  tibble::tibble(
    tidyclust = c("bandwidth"),
    default = c("no default")
  )

param <-
  mean_shift() |>
  set_engine("LPCM") |>
  set_mode("partition") |>
  make_parameter_list(defaults)

This model has r nrow(param) tuning parameters:

#| label: LPCM-param-list
#| echo: false
#| results: asis
param$item

Translation from tidyclust to the original package (partition)

#| label: LPCM-cls
mean_shift(bandwidth = 0.5) |>
  set_engine("LPCM") |>
  set_mode("partition") |>
  translate_tidyclust()

Preprocessing requirements

#| child: template-makes-dummies.Rmd

LPCM::ms() scales each variable internally to the unit range before applying the Gaussian kernel, so the bandwidth value lives on the scaled scale rather than the raw data scale. Bandwidths between roughly 0.05 and 1 are typical; smaller values find more clusters and larger values merge them.

What does it mean to predict?

To predict the cluster assignment for a new observation, the mean shift procedure is run from the new point until it converges to a mode. The observation is then assigned to the cluster of the nearest discovered training mode by Euclidean distance.

References

Cheng, Y. (1995). Mean shift, mode seeking, and clustering. IEEE Transactions on Pattern Analysis and Machine Intelligence, 17(8), 790–799. doi: 10.1109/34.400568
Comaniciu, D., & Meer, P. (2002). Mean shift: A robust approach toward feature space analysis. IEEE Transactions on Pattern Analysis and Machine Intelligence, 24(5), 603–619. doi: 10.1109/34.1000236
Einbeck, J., Evers, L., & Hinchliff, K. (2010). Data compression and regression based on local principal curves. In A. Fink, B. Lausen, W. Seidel, & A. Ultsch (Eds.), Advances in Data Analysis, Data Handling and Business Intelligence (pp. 701–712). Springer.