`geom_smooth` and `scale_x/y_date`

Vindaar commented 3 years ago

Finally adds support for geom_smooth using two different smoothers at the moment:

Savitzky-Golay filter: often called "LOESS" or local regression in other areas. Performs a convolution of the Savitzky-Golay coefficients (essentially a polynomial) with the input data within the window defined by span. The larger the window, the smoother the result.
Polynomial fits. A simple polynomial fit of order N to the full input data. Yields the smoothest (and least precise) result, but can give oscillating predictions (Runge's phenomenon).

Levenberg-Marquardt fits will be added soon. A simple linear fit can just be done using a polynomial of order 1.

Also adds scale_x/y_date that finally allow to define sensible date time ticks and labels based on either string input or timestamps according to some format string and a duration between ticks.

Things left to do:

[x] add recipe showing usage of geom_smooth
[x] add recipe showing usage of geom_smooth for a linear fit (penguins?)
[x] add recipe showing usage of scale_x/y_date

@pietroppeter, @haxscramper

Vindaar commented 3 years ago

Note: the tests will only pass once https://github.com/Vindaar/ginger/pull/31 is merged and the new tag created.

Vindaar commented 3 years ago

This one is for @dom96 :sunglasses:

smooth_test

(red and blue are 2 different smoothing methods)

Vindaar commented 3 years ago

CI is still broken, due to ginger PR not merged (and tagged). Done now, but merge will wait a few days until I have time to deal with possible CI issues.

Vindaar commented 3 years ago

Ah, need to push my latest fix to the least squares wrapper...

Vindaar / ggplotnim

`geom_smooth` and `scale_x/y_date` #127