1 ! Copyright (C) 2008 Doug Coleman, Michael Judge, Loryn Jenkins.
2 ! See http://factorcode.org/license.txt for BSD license.
3 USING: accessors arrays assocs combinators fry generalizations
4 grouping kernel locals math math.functions math.order
5 math.vectors sequences sequences.private sorting ;
6 FROM: math.ranges => [a,b] ;
9 : power-mean ( seq p -- x )
10 [ '[ _ ^ ] map-sum ] [ [ length / ] [ recip ^ ] bi* ] 2bi ; inline
12 ! Delta in degrees-of-freedom
13 : mean-ddof ( seq ddof -- x )
14 [ [ sum ] [ length ] bi ] dip -
15 [ drop 0 ] [ / ] if-zero ; inline
20 : meanest ( seq -- x )
21 [ mean ] keep [ - abs ] with infimum-by ;
23 GENERIC: sum-of-squares ( seq -- x )
24 M: object sum-of-squares [ sq ] map-sum ;
25 M: iota sum-of-squares
26 n>> 1 - [ ] [ 1 + ] [ 1/2 + ] tri * * 3 / ;
28 GENERIC: sum-of-cubes ( seq -- x )
29 M: object sum-of-cubes [ 3 ^ ] map-sum ;
30 M: iota sum-of-cubes sum sq ;
32 GENERIC: sum-of-quads ( seq -- x )
33 M: object sum-of-quads [ 4 ^ ] map-sum ;
45 : sum-of-squared-errors ( seq -- x )
46 [ mean ] keep [ - sq ] with map-sum ; inline
48 : sum-of-absolute-errors ( seq -- x )
49 [ mean ] keep [ - ] with map-sum ; inline
51 : quadratic-mean ( seq -- x ) ! root-mean-square
52 [ sum-of-squares ] [ length ] bi / sqrt ; inline
54 : geometric-mean ( seq -- x )
55 [ [ log ] map-sum ] [ length ] bi /f e^ ; inline
57 : harmonic-mean ( seq -- x )
58 [ [ recip ] map-sum ] [ length swap / ] bi ; inline
60 : contraharmonic-mean ( seq -- x )
61 [ sum-of-squares ] [ sum ] bi / ; inline
65 : trim-points ( p seq -- from to seq )
66 [ length [ * >integer ] keep over - ] keep ;
70 : trimmed-mean ( seq p -- x )
71 swap natural-sort trim-points <slice> mean ;
73 : winsorized-mean ( seq p -- x )
74 swap natural-sort trim-points
76 [ nip dupd nth <array> ]
77 [ [ 1 - ] dip nth <array> ] 3tri
82 :: kth-object-impl ( seq k nth-quot exchange-quot quot: ( x y -- ? ) -- elt )
83 ! Wirth's method, Algorithm's + Data structues = Programs p. 84
84 k seq bounds-check 2drop
97 [ i seq nth-quot call x quot call ] [ i 1 + i! ] while
98 [ x j seq nth-quot call quot call ] [ j 1 - j! ] while
100 i j seq exchange-quot call
109 k seq nth-unsafe ; inline
111 : (kth-object) ( seq k nth-quot exchange-quot quot: ( x y -- ? ) -- elt )
112 ! The algorithm modifiers seq, so we clone it
113 [ >array ] 4dip kth-object-impl ; inline
115 : kth-object-unsafe ( seq k quot: ( x y -- ? ) -- elt )
116 [ [ nth-unsafe ] [ exchange-unsafe ] ] dip (kth-object) ; inline
118 : kth-objects-unsafe ( seq kths quot: ( x y -- ? ) -- elts )
119 '[ _ kth-object-unsafe ] with map ; inline
123 : kth-object ( seq k quot: ( x y -- ? ) -- elt )
124 [ [ nth ] [ exchange ] ] dip (kth-object) ; inline
126 : kth-objects ( seq kths quot: ( x y -- ? ) -- elts )
127 '[ _ kth-object ] with map ; inline
129 : kth-smallests ( seq kths -- elts ) [ < ] kth-objects-unsafe ;
131 : kth-smallest ( seq k -- elt ) [ < ] kth-object-unsafe ;
133 : kth-largests ( seq kths -- elts ) [ > ] kth-objects-unsafe ;
135 : kth-largest ( seq k -- elt ) [ > ] kth-object-unsafe ;
137 : count-relative ( seq k -- lt eq gt )
140 { +lt+ [ [ 1 + ] 2dip ] }
142 { +eq+ [ [ 1 + ] dip ] }
146 : minmax-relative ( seq k -- lt eq gt lt-max gt-min )
147 [ 0 0 0 -1/0. 1/0. ] 2dip '[
149 { +lt+ [ [ 1 + ] 5 ndip '[ _ max ] dip ] }
150 { +gt+ [ [ 1 + ] 3dip min ] }
151 { +eq+ [ [ 1 + ] 4dip drop ] }
155 : lower-median-index ( seq -- n )
157 [ length odd? [ 1 - ] unless ] bi ;
159 : lower-median ( seq -- elt )
160 [ ] [ lower-median-index ] bi kth-smallest ;
162 : upper-median ( seq -- elt )
163 dup midpoint@ kth-smallest ;
165 : medians ( seq -- lower upper )
167 [ [ lower-median-index ] [ midpoint@ ] bi 2array ]
168 bi kth-smallests first2 ;
170 : median ( seq -- x )
171 dup length odd? [ lower-median ] [ medians + 2 / ] if ;
173 ! quantile can be any n-tile. quartile is n = 4, percentile is n = 100
174 ! a,b,c,d parameters, N - number of samples, q is quantile (1/2 for median, 1/4 for 1st quartile)
175 ! http://mathworld.wolfram.com/Quantile.html
177 ! could subtract 1 from a
179 : quantile-x ( a b N q -- x )
180 [ + ] dip * + 1 - ; inline
184 >fraction [ /mod nip ] keep / ; inline
186 :: quantile-indices ( seq qs a b -- seq )
187 qs [ [ a b seq length ] dip quantile-x ] map ;
189 :: qabcd ( y-floor y-ceiling x c d -- qabcd )
190 y-floor y-ceiling y-floor - c d x frac * + * + ;
192 :: quantile-abcd ( seq qs a b c d -- quantile )
193 seq qs a b quantile-indices :> indices
194 indices [ [ floor 0 max ] [ ceiling seq length 1 - min ] bi 2array ] map
195 concat :> index-pairs
197 seq index-pairs kth-smallests
198 2 group indices [ [ first2 ] dip c d qabcd ] 2map ;
200 : quantile1 ( seq qs -- seq' )
201 0 0 1 0 quantile-abcd ;
203 : quantile3 ( seq qs -- seq' )
204 1/2 0 0 0 quantile-abcd ;
206 : quantile4 ( seq qs -- seq' )
207 0 0 0 1 quantile-abcd ;
209 : quantile5 ( seq qs -- seq' )
210 1/2 0 0 1 quantile-abcd ;
212 : quantile6 ( seq qs -- seq' )
213 0 1 0 1 quantile-abcd ;
215 : quantile7 ( seq qs -- seq' )
216 1 -1 0 1 quantile-abcd ;
218 : quantile8 ( seq qs -- seq' )
219 1/3 1/3 0 1 quantile-abcd ;
221 : quantile9 ( seq qs -- seq' )
222 3/8 1/4 0 1 quantile-abcd ;
224 : quartile ( seq -- seq' )
225 { 1/4 1/2 3/4 } quantile5 ;
227 : interquartile ( seq -- q1 q3 )
228 quartile [ first ] [ last ] bi ;
230 : interquartile-range ( seq -- n )
233 : midhinge ( seq -- n )
234 interquartile + 2 / ;
236 : trimean ( seq -- x )
237 quartile first3 [ 2 * ] dip + + 4 / ;
239 : histogram! ( hashtable seq -- hashtable )
240 over '[ _ inc-at ] each ;
242 : histogram-by ( seq quot: ( x -- bin ) -- hashtable )
243 H{ } clone [ '[ @ _ inc-at ] each ] keep ; inline
245 : histogram ( seq -- hashtable )
248 : sorted-histogram ( seq -- alist )
249 histogram sort-values ;
251 : normalized-histogram ( seq -- alist )
252 [ histogram ] [ length ] bi '[ _ / ] assoc-map ;
254 : equal-probabilities ( n -- array )
255 dup recip <array> ; inline
258 histogram >alist [ second ] supremum-by first ;
260 : minmax ( seq -- min max )
261 [ first dup ] keep [ [ min ] [ max ] bi-curry bi* ] 1 each-from ;
266 : fivenum ( seq -- seq' )
267 [ quartile ] [ minmax ] bi [ prefix ] [ suffix ] bi* ;
269 : var-ddof ( seq n -- x )
270 2dup [ length ] dip - 0 <= [
273 [ [ sum-of-squared-errors ] [ length ] bi ] dip - /
276 : population-var ( seq -- x ) 0 var-ddof ; inline
278 : sample-var ( seq -- x ) 1 var-ddof ; inline
280 : std-ddof ( seq n -- x ) var-ddof sqrt ; inline
282 : population-std ( seq -- x ) 0 std-ddof ; inline
284 : sample-std ( seq -- x ) 1 std-ddof ; inline
286 ALIAS: std sample-std
288 : signal-to-noise ( seq -- x ) [ mean ] [ population-std ] bi / ;
290 : demean ( seq -- seq' ) dup mean v-n ;
292 : mean-dev ( seq -- x ) demean vabs mean ;
294 : demedian ( seq -- seq' ) dup median v-n ;
296 : median-dev ( seq -- x ) demedian vabs mean ;
298 : ste-ddof ( seq n -- x ) '[ _ std-ddof ] [ length ] bi sqrt / ;
300 : population-ste ( seq -- x ) 0 ste-ddof ;
302 : sample-ste ( seq -- x ) 1 ste-ddof ;
305 : r-sum-diffs ( x-mean y-mean x-seq y-seq -- (r) )
306 ! finds sigma((xi-mean(x))(yi-mean(y))
307 0 [ [ reach - ] bi@ * + ] 2reduce 2nip ;
309 : (r) ( x-mean y-mean x-seq y-seq x-std y-std -- r )
310 * recip [ [ r-sum-diffs ] keep length 1 - / ] dip * ;
312 : r-stats ( xy-pairs -- x-mean y-mean x-seq y-seq x-std y-std )
313 first2 [ [ [ mean ] bi@ ] 2keep ] 2keep [ population-std ] bi@ ;
316 : pearson-r ( xy-pairs -- r ) r-stats (r) ;
318 : least-squares ( xy-pairs -- alpha beta )
319 r-stats [ 2dup ] 4dip
320 ! stack is x-mean y-mean x-mean y-mean x-seq y-seq x-std y-std
321 [ (r) ] 2keep ! stack is mean(x) mean(y) r sx sy
322 swap / * ! stack is mean(x) mean(y) beta
325 : cov-ddof ( x-seq y-seq ddof -- cov )
326 [ [ demean ] bi@ v* ] dip mean-ddof ;
328 : population-cov ( x-seq y-seq -- cov ) 0 cov-ddof ; inline
330 : sample-cov ( x-seq y-seq -- cov ) 1 cov-ddof ; inline
332 : corr-ddof ( x-seq y-seq n -- corr )
333 [ [ population-cov ] ] dip
334 '[ [ _ var-ddof ] bi@ * sqrt ] 2bi / ;
336 : population-corr ( x-seq y-seq -- corr ) 0 corr-ddof ; inline
338 : sample-corr ( x-seq y-seq -- corr ) 1 corr-ddof ; inline
340 : cum-sum ( seq -- seq' )
341 0 [ + ] accumulate* ;
343 : cum-sum0 ( seq -- seq' )
344 0 [ + ] accumulate nip ;
346 : cum-product ( seq -- seq' )
347 1 [ * ] accumulate* ;
349 : cum-product1 ( seq -- seq' )
350 1 [ * ] accumulate nip ;
352 : cum-mean ( seq -- seq' )
353 0 swap [ [ + dup ] dip 1 + / ] map-index nip ;
355 : cum-count ( seq quot: ( elt -- ? ) -- seq' )
356 [ 0 ] dip '[ @ [ 1 + ] when ] accumulate* ; inline
358 : cum-min ( seq -- seq' )
359 dup ?first [ min ] accumulate* ;
361 : cum-max ( seq -- seq' )
362 dup ?first [ max ] accumulate* ;
364 : entropy ( probabilities -- n )
365 dup sum '[ _ / dup log * ] map-sum neg ;
367 : maximum-entropy ( probabilities -- n )
370 : normalized-entropy ( probabilities -- n )
371 [ entropy ] [ maximum-entropy ] bi / ;
373 : binary-entropy ( p -- h )
374 [ dup log * ] [ 1 swap - dup log * ] bi + neg 2 log / ;
376 : standardize ( u -- v )
377 [ demean ] [ sample-std ] bi [ v/n ] unless-zero ;
379 : standardize-2d ( u -- v )
380 flip [ standardize ] map flip ;
382 : differences ( u -- v )
383 [ rest-slice ] keep v- ;
386 dup minmax over - [ v-n ] [ v/n ] bi* ;
388 : rankings ( histogram -- assoc )
389 sort-keys 0 swap [ rot [ + ] keep swapd ] H{ } assoc-map-as nip ;
391 : rank-values ( seq -- seq' )
392 dup histogram rankings '[ _ at ] map ;
394 : z-score ( seq -- n )
395 [ demean ] [ sample-std ] bi v/n ;
397 : dcg ( scores -- dcg )
398 dup length 1 + 2 swap [a,b] [ log 2 log /f ] map v/ sum ;
400 : ndcg ( scores -- ndcg )
405 swap natural-sort <reversed> dcg /f