ggcorrplot 0.2.0

New features

• New argument sig.stars to append significance stars (***, **, *) to
  the coefficient labels when lab = TRUE and a p.mat is supplied, e.g.
  "-0.85**". Defaults to FALSE (#26, #41, #50; inspired by the ggcorrplot2
  package by @caijun).

• New argument circle.scale to scale the circle sizes when method = "circle",
  useful when the output device size makes the default circles too small or too
  large. Defaults to 1 (contributed by @jdeut, #8).

• New argument nsmall to set a minimum number of decimals in the coefficient
  labels (e.g. nsmall = 2 keeps trailing zeros such as 0.70). Defaults to 0,
  the current behavior (#43; label-formatting idiom suggested by @PawelKulawiak in #15).

• New argument legend.limit to control the limits of the fill color scale.
  Defaults to c(-1, 1); set legend.limit = NULL to use the data range, e.g.
  to display a covariance matrix (#54).

• The colors argument now accepts a vector of any length >= 2, not only 3.
  A length-3 vector still maps to low/mid/high via scale_fill_gradient2 (default
  output unchanged); any other length is spread across the scale with
  scale_fill_gradientn, so an n-color palette such as
  RColorBrewer::brewer.pal(11, "RdBu") can be passed straight to colors =
  without adding a second fill scale (and without the "Scale for fill is already
  present" message) (#52). Requested by @glocke-senda.

• New argument coord.fixed (default TRUE) to optionally drop the fixed 1:1
  aspect ratio. Set coord.fixed = FALSE to let the cells fill the plotting
  area, which can look better with many long variable names (#40).

• New argument lab_fontface to set the font face ("plain", "bold",
  "italic", "bold.italic") of the correlation coefficient labels. Defaults
  to "plain", the current behavior (#15).

• New argument leading.zero to drop the leading zero of the coefficient labels
  (e.g. .23, -.67 instead of 0.23, -0.67), common in correlation tables.
  Defaults to TRUE (leading zero kept, current behavior); set
  leading.zero = FALSE to remove it (#15; idiom from @PawelKulawiak's comment).

• New arguments tl.vjust and tl.hjust to control the vertical and horizontal
  justification of the x-axis text labels. Both default to 1, the current
  behavior (#56).

• New argument use in cor_pmat() to align the p-value matrix's NA pattern
  with a correlation matrix. The default "pairwise.complete.obs" keeps the
  current behavior; use = "everything" sets a pair to NA as soon as either
  variable has a missing value, matching cor()'s default so the two matrices
  line up (@elizabethwe, #51).

Minor changes

• Replaced the deprecated ggplot2::aes_string() with tidy-evaluation aes()
  internally, silencing the ggplot2 deprecation warnings on recent ggplot2
  versions. Default output is unchanged (#57, #58, #59, #60, #61). Based on the
  contribution by @jeherschberger (#62).

• Added a CITATION file so citation("ggcorrplot") returns a proper
  reference (#42, #47).

• Added an internal structural regression test suite that asserts on the built
  plot (layer composition, built data, fill-scale semantics, coordinate system,
  ordering and significance handling), so the plot's structure is checked on CI
  and CRAN and not only by the local visual snapshots (#81).

Bug fixes

• cor_pmat() no longer aborts when a pair of variables has fewer than three
  overlapping non-missing observations to correlate (e.g. two variables that
  never co-occur). Such a pair now returns NA for that cell instead of erroring
  out for the whole matrix; pairs that can be tested are computed as before
  (@elizabethwe, #51).

• A non-square (m x n) correlation matrix now gives a clear error when combined
  with hc.order = TRUE or type = "lower"/"upper" (which require a square
  matrix), instead of silently producing an incorrect plot. type = "full"
  still works for non-square matrices (#5, #10).

• The significance markers no longer error or misalign when the correlation
  matrix and the p-value matrix have different missing-value patterns. P-values
  are now matched to each cell by name instead of by row position.

• When hc.order = TRUE, the hierarchical clustering is now computed on the
  unrounded correlation matrix. Previously the matrix was rounded to digits
  before clustering, so the internal rounding could introduce ties that changed
  the ordering (@buddha2490, #14).

• The tl.col argument (color of the axis text labels) is now applied; it was
  previously ignored. It defaults to NULL, inheriting the color from the theme,
  so the default appearance is unchanged (@LafontRapnouilTristan, #44, #45).

• The significance test is no longer affected by hc.order. Previously, when
  hc.order = TRUE, the p-value matrix was rounded to digits before being
  compared with sig.level, so a p-value just above the threshold (e.g. 0.054)
  could be shown as significant while the same data with hc.order = FALSE
  showed it as non-significant (@worden-lee, #25).

• The significance markers now stay aligned with the tiles when the matrix has
  numeric-looking names. The p-value matrix is now reshaped with the same
  as.is setting as the correlation matrix, so as.is = TRUE no longer places
  the markers off-plot (@cabaez, #37).