SPARTAAS (version 1.2.7)

cerardat_estim_nf: Number of axes to keep

Description

Estimate the correct number of axes to keep in the regression model (see cerardat). We use a resampling procedure see train.

Usage

cerardat_estim_nf(df, row.sup, date, option=list(method = "LOOCV"))

Value

nf

Number of axes to keep (default minimal PRESS value)

PRESS

plot of the Predicted Residual Error Sum of Squares.

RMSE

plot of the Root Mean Squared Error.

MAE

plot of the Mean Absolute Error.

adj.R_sq

plot of the Coefficient of determination R².

Pvalue

plot of p-values from statistical tests on the hypotheses.

data

data frame of MSE, PRESS and R_sq values.

data_stat

data frame of the p-values.

Arguments

df

The data (data.frame) is a contingency table with the observations in rows and the technical groups in columns.

row.sup

Index of supplementary rows in df (vector).

date

The dates of each observation or NA (vector).

option

A list of values that define how the resampling method acts. See detail.

Author

A. COULON

L. BELLANGER

P. HUSI

Details

We implement a resampling procedure (see train) and calculate several metrics in order to determine the number of axes to retain (nf).

The metrics are: PRESS (always use LOOCV for calculations), RMSE, MAE, and adjusted R². In addition, hypothesis tests for normality (shapiro.test), heteroscedasticity (bptest) and autocorrelation (dwtest) are carried out for each value of nf.

You can give a list of options, there are 3 diffrents way of resampling implemented :

Leave-One-Out: `option=list(method="LOOCV")` (Default)

K-fold cross-validation: `option=list(method="cv", number=5)`

repeated K-fold cross-validation: `option=list(method="repeatedcv", number=5, repeats=10)`

Warning: For the CV and repeated CV methods, ensure you have sufficient data per fold. These approaches are particularly useful when the computation time for LOOCV becomes too long (For small datasets, they may generate warnings).

Note: when using the LOOCV method, the values for the Root Mean Square Error (RMSE) are essentially the same as those for PRESS, with the difference that RMSE is the mean rather than the sum.

References

Bellanger L. and Husi P. (2012) Statistical tool for dating and interpreting archaeological contexts using pottery. Journal of Archaeological Science, Elsevier, 39 (4), pp.777-790. doi:10.1016/j.jas.2011.06.031.

Examples

# \donttest{
data(datacerardat)
res = cerardat_estim_nf(datacerardat$df, datacerardat$row.sup, datacerardat$date)

#Number of axes to keep (minimal PRESS value)
res$nf

#the plots
res$PRESS
res$MAE
res$adj.R_sq

#p-value
res$Pvalue
# }