archetypal-package {archetypal}

Finds the Archetypal Analysis of a Data Frame

Description

Performs archetypal analysis by using Principal Convex Hull Analysis (PCHA) under a full control of all algorithmic parameters. It contains a set of functions for determining the initial solution, the optimal algorithmic parameters and the optimal number of archetypes. Post run tools are also available for the assessment of the derived solution.

Compute Archetypal Analysis (AA)

The main function is archetypal which is a variant of PCHA algorithm, see [1], [2], suitable for R language. It provides control to the entire set of involved parameters and has two main options:

1. initialrows = NULL, then a method from "projected_convexhull", "convexhull",
   "partitioned_convexhul", "furthestsum", "outmost", "random" is used

2. initialrows = (a vector of kappas rows), then given rows form the initial solution for AA

This is the main function of the package, but extensive trials has shown that:

• AA may be very difficult to run if a random initial solution has been chosen

• for the same data set the final Sum of Squared Errors (SSE) may be much smaller if initial solution is close to the final one

• even the quality of AA done is affected from the starting point

This is the reason why we have developed a whole set of methods for choosing initial solution for the PCHA algorithm.

Find a time efficient initial approximation for AA

There are three functions that work with the Convex Hull (CH) of data set.

1. find_outmost_convexhull_points computes the CH of all points

2. find_outmost_projected_convexhull_points computes the CH for all possible combinations of variables taken by npr (default=2)

3. find_outmost_partitioned_convexhull_points makes np partitions of data frame (defualt=10), then computes CH for each partition and finally gives the CH of overall union

The most simple method for estimating an initial solution is find_outmost_points where we just compute the outermost points, i.e. those that are the most frequent outermost for all available points.

The default method "FurthestSum" (FS) of PCHA (see [1], [2]) is used by find_furthestsum_points which applies FS for nfurthest times (default=10) and then finds the most frequent points.

Of course "random" method is available for comparison reasons and that gives a random set of kappas points as initial solution.

All methods give the number of rows for the input data frame as integers. Attention needed if your data frame has row names which are integers but not identical to 1:dim(df)[1].

Find the optimal number of archetypes

For that task find_optimal_kappas is available which runs for each kappas from 1 to maxkappas (default=15) ntrials (default=10) times AA, stores SSE, VarianceExplained from each run and then computes knee or elbow point by using UIK method, see [3].

Determining the optimal updating parameters

Extensive trials have shown us that choosing the proper values for algorithmic updating parameters (muAup, muAdown, muBup, muBdown) can speed up remarkably the process. That is the task of find_pcha_optimal_parameters which conducts a grid search with different values of these parameters and returns the values which minimize the SSE after a fixed number of iterations (testing_iters, default=10).

Evaluate the quality of Archetypal Analysis

By using function check_Bmatrix we can evaluate the overall quality of applied method and algorithm. Quality can be considered high:

1. if every archetype is being created by a small number of data points

2. if relevant weights are not numerically insignificant

Of course we must take into account the SSE and VarianceExplained, but if we have to compare two solutions with similar termination status, then we must choose that of the simplest B matrix form.

Resampling

The package includes a function for resampling (grouped_resample) which may be used for standard bootstrapping or for subsampling. This function allows samples to be drawn with or without replacement, by groups and with or without Dirichlet weights. This provides a variety of options for researchers who wish to correct sample biases, estimate empirical confidence intervals, and/or subsample large data sets.

Post-run tools

Except from check_Bmatrix there exist next functions for checking the convergence process itself and for examining the local neighborhoud of archetypes:

1. The function study_AAconvergence analyzes the history of iterations done and produces a multi-panel plot showing the steps and quality of the convergence to the final archetypes.

2. By setting the desired number npoints as argument in function find_closer_points we can then find the data points that are in the local neighborhood of each archetype. This allows us to study the properties of the solution or manually choose an initial approximation to search for a better fit.

Note

Bug reports and feature requests can be sent to [e-mail] or
[e-mail].

Author(s)

Maintainer: Demetris Christopoulos [e-mail]

Other contributors:

• David Midgley [e-mail] [contributor]

• Sunil Venaik [e-mail] [contributor]

• INSEAD Fontainebleau France [funder]

References

[1] M Morup and LK Hansen, "Archetypal analysis for machine learning and data mining", Neurocomputing (Elsevier, 2012). https://doi.org/10.1016/j.neucom.2011.06.033.

[2] Source: https://mortenmorup.dk/?page_id=2 , last accessed 2024-03-09

[3] Christopoulos, Demetris T., Introducing Unit Invariant Knee (UIK) As an Objective Choice for Elbow Point in Multivariate Data Analysis Techniques (March 1, 2016). Available at SSRN: https://ssrn.com/abstract=3043076 or http://dx.doi.org/10.2139/ssrn.3043076

See Also

archetypal