6  R Interop & Plot

(ns assignment.r-interop
  (:require
    [assignment.generate-data :refer [data]]
    [assignment.lda :refer [lda-pipe-fn models]]
    [clojisr.v1.applications.plotting
     :refer [plot->svg]]
    [clojisr.v1.r :refer [r+]]
    [clojisr.v1.require :refer [require-r]]
    [scicloj.kindly.v4.kind :as kind]
    [scicloj.ml.core :as ml]
    [scicloj.ml.dataset :as ds]
    [scicloj.ml.dataset :as ds]))

6.1 Load the required R libraries

(require-r '[base :refer [summary]]
           '[ggplot2 :refer [ggplot aes geom_point
                             geom_contour theme_bw]])

nil

(summary data)

       x1                 x2              group          
 Min.   :-5.96223   Min.   :-30.5006   Length:600        
 1st Qu.:-0.03666   1st Qu.: -5.0922   Class :character  
 Median : 3.25954   Median : -0.9309   Mode  :character  
 Mean   : 3.07651   Mean   :  0.2922                     
 3rd Qu.: 6.04087   3rd Qu.:  5.5367                     
 Max.   :12.26393   Max.   : 19.8487                     

6.2 Setup dataset

Clojure does a nice job constructing cartesian products with the for function. As an illustration:

(def letters ["A" "B"])

(def numbers [1 2 3])

(for [x letters
      y numbers]
  [x y])

(["A" 1] ["A" 2] ["A" 3] ["B" 1] ["B" 2] ["B" 3])

In other words, cartesian product takes each element of sequence x (in illustration, letters) and pairs them with each element of sequence y (in illustration, numbers) to create a new set of ordered pairs.

(def dat (for [x (range -5 12 0.1)
               y (range -14 20 0.2)]
           {:x1 x :x2 y :group nil}))

Using the cartesian product, I can create a nil prediction dataset with :x1 and :x2 values from x in [-5, 12] to y in [-14, 20].

(def contour-data
  (ds/dataset dat))

(def predictions
  (-> contour-data
      (ml/transform-pipe
        lda-pipe-fn
        (-> models second :fit-ctx))
      :metamorph/data
      :group
      vec))

(def lda-predict
  (ds/add-or-replace-column contour-data :group predictions))

Predictions variable is collecting the estimated :group based on the lda-pipe-fn plus the best model described in the last chapter. LDA-predict is putting those predictions in the contour-data set.

6.3 Plot LDA

^kind/hiccup
(-> (ggplot :data data (aes :x 'x1 :y 'x2 :color 'group))
    (r+ (geom_point))
    (r+ (geom_contour :data lda-predict (aes :x 'x1 :y 'x2 :z 'group) :col "black"))
    (r+ (theme_bw))
    plot->svg)

(comment                                                    ;reverse process
  (def dat
    (for [minx (Math/floor (apply min (:x1 data)))
          maxx (Math/floor (apply max (:x1 data)))
          miny (Math/floor (apply min (:x2 data)))
          maxy (Math/floor (apply max (:x2 data)))
          x (range (int minx) (int maxx) 0.1)
          y (range (int miny) (int maxy) 0.2)]
      {:x1 x :x2 y :group nil}))

  (def lookup-table
    (-> models second :fit-ctx :model
        :target-categorical-maps :group :lookup-table))

  (def lookup-table-invert
    (clojure.set/map-invert lookup-table))

  (def lda-predict
    (-> lda-pred-pre-transform
        (ds/add-or-replace-column
          :group (map #(get lookup-table-invert %)
                      (map int (:group lda-predict)))))))

source: src/assignment/r_interop.clj