Some improvements of plot.AccurateGLM()

[kkondo1981]

Fixed to use plot.AccurateGLM() with non-Gaussian models.

[kkondo1981]

todo

• fix resids
• restore par
• draw setting

[kkondo1981]

Fixed the last two issues in my comment before this.

[kkondo1981]

Modify the last issue (partial residual plot for non-Gaussian models) as below.

1. plot.AccurateGLM(resid=TRUE) uses working residuals for partial residuals as default like residuals.glm(type="partial")
2. However, in some cases (e.g., Poisson regression with small mu), distributions of working residuals could have large variances and skewness. So it provides several options as below. a. Showing smoothed partial residuals as dotted lines with smooth_resid=TRUE, smooth_resid="both", or smooth_resid="smooth_only". b. Adding options to use other kinds of residuals (Pearson residuals, deviance residuals) with resid="pearson" or resid="deviance". c. Adding an option to use user-calculated residuals with resid=r (here, r is asuumed as a numeric vector).

Moreover, I added an example markdown file as below. https://github.com/kkondo1981/aglm/blob/fix/plot-smallfix-20190706/examples/freMTPL2freq.rmd

[kkondo1981]

Further improvement to select smoother and to get clearer partial residual plots in the example.

[kkondo1981]

Further improvement to add rug plots with add_rug=TRUE in plot.AccurateGLM().

[kkondo1981]

Finished tests and debugs.