1 ! Copyright (C) 2008 Doug Coleman, Michael Judge.
2 ! See http://factorcode.org/license.txt for BSD license.
3 USING: assocs combinators generalizations kernel locals math
4 math.functions math.order math.vectors sequences
5 sequences.private sorting fry arrays grouping sets ;
8 : power-mean ( seq p -- x )
9 [ '[ _ ^ ] map-sum ] [ [ length / ] [ recip ^ ] bi* ] 2bi ; inline
12 [ sum ] [ length ] bi / ; inline
14 : sum-of-squares ( seq -- x )
15 [ sq ] map-sum ; inline
17 : sum-of-squared-errors ( seq -- x )
18 [ mean ] keep [ - sq ] with map-sum ; inline
20 : sum-of-absolute-errors ( seq -- x )
21 [ mean ] keep [ - ] with map-sum ; inline
23 : quadratic-mean ( seq -- x ) ! root-mean-square
24 [ sum-of-squares ] [ length ] bi / sqrt ; inline
26 : geometric-mean ( seq -- x )
27 [ length ] [ product ] bi nth-root ; inline
29 : harmonic-mean ( seq -- x )
30 [ recip ] map-sum recip ; inline
32 : contraharmonic-mean ( seq -- x )
33 [ sum-of-squares ] [ sum ] bi / ; inline
37 :: ((kth-object)) ( seq k nth-quot exchange-quot quot: ( x y -- ? ) -- elt )
38 #! Wirth's method, Algorithm's + Data structues = Programs p. 84
39 k seq bounds-check 2drop
52 [ i seq nth-quot call x quot call ] [ i 1 + i! ] while
53 [ x j seq nth-quot call quot call ] [ j 1 - j! ] while
55 i j seq exchange-quot call
66 : (kth-object) ( seq k nth-quot exchange-quot quot: ( x y -- ? ) -- elt )
67 #! The algorithm modifiers seq, so we clone it
68 [ clone ] 4dip ((kth-object)) ; inline
70 : kth-object-unsafe ( seq k quot: ( x y -- ? ) -- elt )
71 [ [ nth-unsafe ] [ exchange-unsafe ] ] dip (kth-object) ; inline
73 : kth-objects-unsafe ( seq kths quot: ( x y -- ? ) -- elts )
75 '[ [ nth-unsafe ] [ exchange-unsafe ] _ ((kth-object)) ] with map ; inline
79 : kth-object ( seq k quot: ( x y -- ? ) -- elt )
80 [ [ nth ] [ exchange ] ] dip (kth-object) ; inline
82 : kth-objects ( seq kths quot: ( x y -- ? ) -- elts )
84 '[ [ nth ] [ exchange ] _ ((kth-object)) ] with map ; inline
86 : kth-smallests ( seq kths -- elts ) [ < ] kth-objects-unsafe ;
88 : kth-smallest ( seq k -- elt ) [ < ] kth-object-unsafe ;
90 : kth-largests ( seq kths -- elts ) [ > ] kth-objects-unsafe ;
92 : kth-largest ( seq k -- elt ) [ > ] kth-object-unsafe ;
94 : count-relative ( seq k -- lt eq gt )
97 { +lt+ [ [ 1 + ] 2dip ] }
99 { +eq+ [ [ 1 + ] dip ] }
103 : minmax-relative ( seq k -- lt eq gt lt-max gt-min )
104 [ 0 0 0 -1/0. 1/0. ] 2dip '[
106 { +lt+ [ [ 1 + ] 5 ndip '[ _ max ] dip ] }
107 { +gt+ [ [ 1 + ] 3dip min ] }
108 { +eq+ [ [ 1 + ] 4dip drop ] }
112 : lower-median-index ( seq -- n )
114 [ length odd? [ 1 - ] unless ] bi ;
116 : lower-median ( seq -- elt )
117 [ ] [ lower-median-index ] bi kth-smallest ;
119 : upper-median ( seq -- elt )
120 dup midpoint@ kth-smallest ;
122 : medians ( seq -- lower upper )
124 [ [ lower-median-index ] [ midpoint@ ] bi 2array ]
125 bi kth-smallests first2 ;
127 : median ( seq -- x )
129 [ lower-median ] [ medians + 2 / ] if ;
131 ! quantile can be any n-tile. quartile is n = 4, percentile is n = 100
132 ! a,b,c,d parameters, N - number of samples, q is quantile (1/2 for median, 1/4 for 1st quartile)
133 ! http://mathworld.wolfram.com/Quantile.html
135 ! could subtract 1 from a
137 : quantile-x ( a b N q -- x )
138 [ + ] dip * + 1 - ; inline
142 >fraction [ /mod nip ] keep / ; inline
144 :: quantile-indices ( seq qs a b c d -- seq )
145 qs [ [ a b seq length ] dip quantile-x ] map ;
147 :: qabcd ( y-floor y-ceiling x c d -- qabcd )
148 y-floor y-ceiling y-floor - c d x frac * + * + ;
150 :: quantile-abcd ( seq qs a b c d -- quantile )
151 seq qs a b c d quantile-indices :> indices
152 indices [ [ floor ] [ ceiling ] bi 2array ] map
153 concat :> index-pairs
155 seq index-pairs kth-smallests
156 2 group indices [ [ first2 ] dip c d qabcd ] 2map ;
158 : quantile1 ( seq qs -- seq' )
159 0 0 1 0 quantile-abcd ;
161 : quantile3 ( seq qs -- seq' )
162 1/2 0 0 0 quantile-abcd ;
164 : quantile4 ( seq qs -- seq' )
165 0 0 0 1 quantile-abcd ;
167 : quantile5 ( seq qs -- seq' )
168 1/2 0 0 1 quantile-abcd ;
170 : quantile6 ( seq qs -- seq' )
171 0 1 0 1 quantile-abcd ;
173 : quantile7 ( seq qs -- seq' )
174 1 -1 0 1 quantile-abcd ;
176 : quantile8 ( seq qs -- seq' )
177 1/3 1/3 0 1 quantile-abcd ;
179 : quantile9 ( seq qs -- seq' )
180 3/8 1/4 0 1 quantile-abcd ;
182 : quartile ( seq -- seq' )
183 { 1/4 1/2 3/4 } quantile5 ;
187 : (sequence>assoc) ( seq map-quot: ( x -- ..y ) insert-quot: ( ..y assoc -- ) assoc -- assoc )
188 [ swap curry compose each ] keep ; inline
192 : sequence>assoc! ( assoc seq map-quot: ( x -- ..y ) insert-quot: ( ..y assoc -- ) -- assoc )
193 4 nrot (sequence>assoc) ; inline
195 : sequence>assoc ( seq map-quot: ( x -- ..y ) insert-quot: ( ..y assoc -- ) exemplar -- assoc )
196 clone (sequence>assoc) ; inline
198 : sequence>hashtable ( seq map-quot: ( x -- ..y ) insert-quot: ( ..y assoc -- ) -- hashtable )
199 H{ } sequence>assoc ; inline
201 : histogram! ( hashtable seq -- hashtable )
202 [ ] [ inc-at ] sequence>assoc! ;
204 : histogram-by ( seq quot: ( x -- bin ) -- hashtable )
205 [ inc-at ] sequence>hashtable ; inline
207 : histogram ( seq -- hashtable )
210 : sorted-histogram ( seq -- alist )
211 histogram sort-values ;
213 : collect-pairs ( seq quot: ( x -- v k ) -- hashtable )
214 [ push-at ] sequence>hashtable ; inline
216 : collect-by ( seq quot: ( x -- x' ) -- hashtable )
217 [ dup ] prepose collect-pairs ; inline
221 [ ] [ [ [ second ] bi@ > ] most ] map-reduce first ;
223 ERROR: empty-sequence ;
225 : minmax ( seq -- min max )
229 [ first dup ] keep [ [ min ] [ max ] bi-curry bi* ] each
235 : sample-var ( seq -- x )
236 #! normalize by N-1; unbiased
240 [ sum-of-squared-errors ] [ length 1 - ] bi /
243 : full-var ( seq -- x )
247 [ sum-of-squared-errors ] [ length ] bi /
250 ALIAS: var sample-var
252 : sample-std ( seq -- x ) sample-var sqrt ;
254 : full-std ( seq -- x ) full-var sqrt ;
256 ALIAS: std sample-std
258 : mean-dev ( seq -- x ) dup mean v-n vabs mean ;
260 : median-dev ( seq -- x ) dup median v-n vabs mean ;
262 : sample-ste ( seq -- x ) [ sample-std ] [ length ] bi sqrt / ;
264 : full-ste ( seq -- x ) [ full-std ] [ length ] bi sqrt / ;
266 ALIAS: ste sample-ste
268 : ((r)) ( mean(x) mean(y) {x} {y} -- (r) )
269 ! finds sigma((xi-mean(x))(yi-mean(y))
270 0 [ [ [ pick ] dip swap - ] bi@ * + ] 2reduce 2nip ;
272 : (r) ( mean(x) mean(y) {x} {y} sx sy -- r )
273 * recip [ [ ((r)) ] keep length 1 - / ] dip * ;
275 : [r] ( {{x,y}...} -- mean(x) mean(y) {x} {y} sx sy )
276 first2 [ [ [ mean ] bi@ ] 2keep ] 2keep [ std ] bi@ ;
278 : r ( {{x,y}...} -- r )
281 : r^2 ( {{x,y}...} -- r )
284 : least-squares ( {{x,y}...} -- alpha beta )
285 [r] { [ 2dup ] [ ] [ ] [ ] [ ] } spread
286 ! stack is mean(x) mean(y) mean(x) mean(y) {x} {y} sx sy
287 [ (r) ] 2keep ! stack is mean(x) mean(y) r sx sy
288 swap / * ! stack is mean(x) mean(y) beta
291 : cov ( {x} {y} -- cov )
292 [ dup mean v-n ] bi@ v* mean ;
294 : sample-corr ( {x} {y} -- corr )
295 [ cov ] [ [ sample-var ] bi@ * sqrt ] 2bi / ;
297 : full-corr ( {x} {y} -- corr )
298 [ cov ] [ [ full-var ] bi@ * sqrt ] 2bi / ;
300 ALIAS: corr sample-corr
302 : cum-sum ( seq -- seq' )
303 0 swap [ + dup ] map nip ;
305 : cum-product ( seq -- seq' )
306 1 swap [ * dup ] map nip ;
308 : cum-min ( seq -- seq' )
309 [ ?first ] keep [ min dup ] map nip ;
311 : cum-max ( seq -- seq' )
312 [ ?first ] keep [ max dup ] map nip ;
314 : entropy ( seq -- n )
315 dup members [ [ = ] curry count ] with map
316 dup sum v/n dup [ log ] map v* sum neg ;
318 : binary-entropy ( p -- h )
319 [ dup log * ] [ 1 swap - dup log * ] bi + neg 2 log / ;
321 : standardize ( u -- v )
322 [ dup mean v-n ] [ std ] bi v/n ;
324 : differences ( u -- v )
325 [ 1 tail-slice ] keep [ - ] 2map ;
328 dup minmax over - [ v-n ] [ v/n ] bi* ;