Merged PR from Dirk Eddelbuettel for defining STRICT_R_HEADERS in Rcpp
rcpp_wt_bases_paul.cpp to fix array
out-of-bounds reading issue
plot.biwavelet help file with correct phase arrow interpretation
Updated package by removing benchmarks in vignettes for CRAN submission
yaxis tickmarks should now be accurate (no more rounding issues)
Fixed documentation for wtc function
Fixed return NULL issues with Windows platforms
plot.biwavelet so that the COI extends all
the way to the bottom of the plot (max of periods)
plot.biwavelet so that the
argument applies to rsq values for wtc and pwtc objects
plot.biwavelet so that the
argument applies to rsq values
phase.biwavelet now plots the regions whose significance
arrow.cutoff. If the object being plotted does not have
a significance field, regions whose zvalues exceed the
quantile will be plotted.
Fixed C++ warning about unsupported dynamically sized arrays
Fixed handling of
lag1 coefficients in
Fixed handling of axis preferences in
Fixed handling of time in
Fixed x-axis in
Fixed displacement of COI, contours and phase arrows in
plot.biwavelet when adding a color bar.
check.datum; hopefully for the last time.
Faster wt bases and row quantile (Rcpp implementations):
param parameter for all
rcpp_wt_bases_* must be
within interval (0..10).
rcpp_row_quantile function requires
a matrix as a parameter (use
as.matrix() for vectors).
Fixed Rcpp implementation of the
int type with
double type for
which caused small scales to be rendered incorrectly.
Fixed interpretation of phase differences in
plot.biwavelet help file
Added unit tests for approx 78% of the code.
Implemented a parallelized Monte Carlo simulation function
wtc_sig_parallel which is 2 to 4 times faster on a 4-core CPU
than the original
wtc.sig. The speedup is noticeable on:
nrads >= 800,
multi-core systems with 4+ cores.
However, parallelization involves a significant heat-up phase because all the workers need to be started and they need to load all the required packages. This will be addresses in future versions of biwavelet.
Added a speed-optimized version of
arima.sim function with a pair of functions
ar1_ma0_sim. These functions are still
implemented in R. We can reimplement them later in C.
wt.bases morlet, paul and dog in C.
Removed unused function
close all progress bars after use
wtc can now handle non-finite values when computing
the quantiles of the rsq values from the Monte Carlo simulations
Added ability to handle custom color palettes in
Users can now specify any color scheme using the
Fixed limited padding issue, which could lead to weird edge effects. Current padding level is identical to that of Torrence & Compo (1998).
Changed the default
tol value from 0.95 to 1 in the
Added semi-transparent COI.
check.datum function so that it does not assume a sampling frequency of 1.
Added ability to set
Improved the implementation of
phase.plot to allow for much better looking phase arrows (thanks Huidong Tang).
wt faster by avoiding excessive padding (thanks Huidong Tang).
check.datum tolerate slight inconsistencies in the size of timesteps.
Added arguments in
phase.plot to control the length
of the phase arrows and the size of the arrow heads independently.
Fixed code in
check.data to test for constant step size in the data.
pwtc can be used to perform partial wavelet coherence between two time series
x1 by controlling for (or partialling-out) a third time series
Users can now specify the density of the phase arrows using the
Fixed bug in
wt affecting the significance region (thanks Patrick Kilduff and Flora Cordoleani).
Users can now specify the color, line width and line type for
the COI, significance contours and phase arrows using the
Removed misleading examples showing how to compute the 'bias-corrected' wavelet coherence.
There is no bias for the wavelet coherence function, so using the default
type argument in the
plot function is recommended.
Fixed typos in the documentation of plot.biwavelet and xwt (thanks Lei Cheng).
As of biwavelet version 0.14, the bias-corrected wavelet and cross-wavelet spectra are automatically computed and plotted by default using the methods described by Liu et al. (2007) and Veleda et al. (2012). This correction is needed because the traditional approach for computing the power spectrum (e.g., Torrence and Compo 1998) leads to an artificial and systematic reduction in power at lower periods.
Plotting function now accepts traditional plotting flags such as xaxt and yaxt to control x and y tickmarks.