1 ! Copyright (C) 2008 Doug Coleman, Michael Judge, Loryn Jenkins.
2 ! See http://factorcode.org/license.txt for BSD license.
3 USING: arrays assocs combinators fry generalizations grouping
4 kernel locals math math.functions math.order math.vectors
5 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 : sum-of-squares ( seq -- x )
21 [ sq ] map-sum ; inline
23 : sum-of-squared-errors ( seq -- x )
24 [ mean ] keep [ - sq ] with map-sum ; inline
26 : sum-of-absolute-errors ( seq -- x )
27 [ mean ] keep [ - ] with map-sum ; inline
29 : quadratic-mean ( seq -- x ) ! root-mean-square
30 [ sum-of-squares ] [ length ] bi / sqrt ; inline
32 : geometric-mean ( seq -- x )
33 [ [ log ] map-sum ] [ length ] bi /f e^ ; inline
35 : harmonic-mean ( seq -- x )
36 [ [ recip ] map-sum ] [ length swap / ] bi ; inline
38 : contraharmonic-mean ( seq -- x )
39 [ sum-of-squares ] [ sum ] bi / ; inline
43 : trim-points ( p seq -- from to seq )
44 [ length [ * >integer ] keep over - ] keep ;
48 : trimmed-mean ( seq p -- x )
49 swap natural-sort trim-points <slice> mean ;
51 : winsorized-mean ( seq p -- x )
52 swap natural-sort trim-points
54 [ nip dupd nth <array> ]
55 [ [ 1 - ] dip nth <array> ] 3tri
60 :: kth-object-impl ( seq k nth-quot exchange-quot quot: ( x y -- ? ) -- elt )
61 ! Wirth's method, Algorithm's + Data structues = Programs p. 84
62 k seq bounds-check 2drop
75 [ i seq nth-quot call x quot call ] [ i 1 + i! ] while
76 [ x j seq nth-quot call quot call ] [ j 1 - j! ] while
78 i j seq exchange-quot call
87 k seq nth-unsafe ; inline
89 : (kth-object) ( seq k nth-quot exchange-quot quot: ( x y -- ? ) -- elt )
90 ! The algorithm modifiers seq, so we clone it
91 [ >array ] 4dip kth-object-impl ; inline
93 : kth-object-unsafe ( seq k quot: ( x y -- ? ) -- elt )
94 [ [ nth-unsafe ] [ exchange-unsafe ] ] dip (kth-object) ; inline
96 : kth-objects-unsafe ( seq kths quot: ( x y -- ? ) -- elts )
97 '[ _ kth-object-unsafe ] with map ; inline
101 : kth-object ( seq k quot: ( x y -- ? ) -- elt )
102 [ [ nth ] [ exchange ] ] dip (kth-object) ; inline
104 : kth-objects ( seq kths quot: ( x y -- ? ) -- elts )
105 '[ _ kth-object ] with map ; inline
107 : kth-smallests ( seq kths -- elts ) [ < ] kth-objects-unsafe ;
109 : kth-smallest ( seq k -- elt ) [ < ] kth-object-unsafe ;
111 : kth-largests ( seq kths -- elts ) [ > ] kth-objects-unsafe ;
113 : kth-largest ( seq k -- elt ) [ > ] kth-object-unsafe ;
115 : count-relative ( seq k -- lt eq gt )
118 { +lt+ [ [ 1 + ] 2dip ] }
120 { +eq+ [ [ 1 + ] dip ] }
124 : minmax-relative ( seq k -- lt eq gt lt-max gt-min )
125 [ 0 0 0 -1/0. 1/0. ] 2dip '[
127 { +lt+ [ [ 1 + ] 5 ndip '[ _ max ] dip ] }
128 { +gt+ [ [ 1 + ] 3dip min ] }
129 { +eq+ [ [ 1 + ] 4dip drop ] }
133 : lower-median-index ( seq -- n )
135 [ length odd? [ 1 - ] unless ] bi ;
137 : lower-median ( seq -- elt )
138 [ ] [ lower-median-index ] bi kth-smallest ;
140 : upper-median ( seq -- elt )
141 dup midpoint@ kth-smallest ;
143 : medians ( seq -- lower upper )
145 [ [ lower-median-index ] [ midpoint@ ] bi 2array ]
146 bi kth-smallests first2 ;
148 : median ( seq -- x )
149 dup length odd? [ lower-median ] [ medians + 2 / ] if ;
151 ! quantile can be any n-tile. quartile is n = 4, percentile is n = 100
152 ! a,b,c,d parameters, N - number of samples, q is quantile (1/2 for median, 1/4 for 1st quartile)
153 ! http://mathworld.wolfram.com/Quantile.html
155 ! could subtract 1 from a
157 : quantile-x ( a b N q -- x )
158 [ + ] dip * + 1 - ; inline
162 >fraction [ /mod nip ] keep / ; inline
164 :: quantile-indices ( seq qs a b -- seq )
165 qs [ [ a b seq length ] dip quantile-x ] map ;
167 :: qabcd ( y-floor y-ceiling x c d -- qabcd )
168 y-floor y-ceiling y-floor - c d x frac * + * + ;
170 :: quantile-abcd ( seq qs a b c d -- quantile )
171 seq qs a b quantile-indices :> indices
172 indices [ [ floor 0 max ] [ ceiling seq length 1 - min ] bi 2array ] map
173 concat :> index-pairs
175 seq index-pairs kth-smallests
176 2 group indices [ [ first2 ] dip c d qabcd ] 2map ;
178 : quantile1 ( seq qs -- seq' )
179 0 0 1 0 quantile-abcd ;
181 : quantile3 ( seq qs -- seq' )
182 1/2 0 0 0 quantile-abcd ;
184 : quantile4 ( seq qs -- seq' )
185 0 0 0 1 quantile-abcd ;
187 : quantile5 ( seq qs -- seq' )
188 1/2 0 0 1 quantile-abcd ;
190 : quantile6 ( seq qs -- seq' )
191 0 1 0 1 quantile-abcd ;
193 : quantile7 ( seq qs -- seq' )
194 1 -1 0 1 quantile-abcd ;
196 : quantile8 ( seq qs -- seq' )
197 1/3 1/3 0 1 quantile-abcd ;
199 : quantile9 ( seq qs -- seq' )
200 3/8 1/4 0 1 quantile-abcd ;
202 : quartile ( seq -- seq' )
203 { 1/4 1/2 3/4 } quantile5 ;
205 : interquartile ( seq -- q1 q3 )
206 quartile [ first ] [ last ] bi ;
208 : interquartile-range ( seq -- n )
211 : midhinge ( seq -- n )
212 interquartile + 2 / ;
214 : trimean ( seq -- x )
215 quartile first3 [ 2 * ] dip + + 4 / ;
217 : histogram! ( hashtable seq -- hashtable )
218 over '[ _ inc-at ] each ;
220 : histogram-by ( seq quot: ( x -- bin ) -- hashtable )
221 H{ } clone [ '[ @ _ inc-at ] each ] keep ; inline
223 : histogram ( seq -- hashtable )
226 : sorted-histogram ( seq -- alist )
227 histogram sort-values ;
229 : normalized-histogram ( seq -- alist )
230 [ histogram ] [ length ] bi '[ _ / ] assoc-map ;
232 : equal-probabilities ( n -- array )
233 dup recip <array> ; inline
236 histogram >alist [ second ] supremum-by first ;
238 : minmax ( seq -- min max )
239 [ first dup ] keep [ [ min ] [ max ] bi-curry bi* ] 1 each-from ;
244 : fivenum ( seq -- seq' )
245 [ quartile ] [ minmax ] bi [ prefix ] [ suffix ] bi* ;
247 : var-ddof ( seq n -- x )
248 2dup [ length ] dip - 0 <= [
251 [ [ sum-of-squared-errors ] [ length ] bi ] dip - /
254 : population-var ( seq -- x ) 0 var-ddof ; inline
256 : sample-var ( seq -- x ) 1 var-ddof ; inline
258 : std-ddof ( seq n -- x ) var-ddof sqrt ; inline
260 : population-std ( seq -- x ) 0 std-ddof ; inline
262 : sample-std ( seq -- x ) 1 std-ddof ; inline
264 ALIAS: std sample-std
266 : signal-to-noise ( seq -- x ) [ mean ] [ population-std ] bi / ;
268 : demean ( seq -- seq' ) dup mean v-n ;
270 : mean-dev ( seq -- x ) demean vabs mean ;
272 : demedian ( seq -- seq' ) dup median v-n ;
274 : median-dev ( seq -- x ) demedian vabs mean ;
276 : ste-ddof ( seq n -- x ) '[ _ std-ddof ] [ length ] bi sqrt / ;
278 : population-ste ( seq -- x ) 0 ste-ddof ;
280 : sample-ste ( seq -- x ) 1 ste-ddof ;
283 : r-sum-diffs ( x-mean y-mean x-seq y-seq -- (r) )
284 ! finds sigma((xi-mean(x))(yi-mean(y))
285 0 [ [ reach - ] bi@ * + ] 2reduce 2nip ;
287 : (r) ( x-mean y-mean x-seq y-seq x-std y-std -- r )
288 * recip [ [ r-sum-diffs ] keep length 1 - / ] dip * ;
290 : r-stats ( xy-pairs -- x-mean y-mean x-seq y-seq x-std y-std )
291 first2 [ [ [ mean ] bi@ ] 2keep ] 2keep [ population-std ] bi@ ;
294 : pearson-r ( xy-pairs -- r ) r-stats (r) ;
296 : least-squares ( xy-pairs -- alpha beta )
297 r-stats [ 2dup ] 4dip
298 ! stack is x-mean y-mean x-mean y-mean x-seq y-seq x-std y-std
299 [ (r) ] 2keep ! stack is mean(x) mean(y) r sx sy
300 swap / * ! stack is mean(x) mean(y) beta
303 : cov-ddof ( x-seq y-seq ddof -- cov )
304 [ [ demean ] bi@ v* ] dip mean-ddof ;
306 : population-cov ( x-seq y-seq -- cov ) 0 cov-ddof ; inline
308 : sample-cov ( x-seq y-seq -- cov ) 1 cov-ddof ; inline
310 : corr-ddof ( x-seq y-seq n -- corr )
311 [ [ population-cov ] ] dip
312 '[ [ _ var-ddof ] bi@ * sqrt ] 2bi / ;
314 : population-corr ( x-seq y-seq -- corr ) 0 corr-ddof ; inline
316 : sample-corr ( x-seq y-seq -- corr ) 1 corr-ddof ; inline
318 : cum-sum ( seq -- seq' )
319 0 [ + ] accumulate* ;
321 : cum-sum0 ( seq -- seq' )
322 0 [ + ] accumulate nip ;
324 : cum-product ( seq -- seq' )
325 1 [ * ] accumulate* ;
327 : cum-product1 ( seq -- seq' )
328 1 [ * ] accumulate nip ;
330 : cum-mean ( seq -- seq' )
331 0 swap [ [ + dup ] dip 1 + / ] map-index nip ;
333 : cum-count ( seq quot: ( elt -- ? ) -- seq' )
334 [ 0 ] dip '[ @ [ 1 + ] when ] accumulate* ; inline
336 : cum-min ( seq -- seq' )
337 dup ?first [ min ] accumulate* ;
339 : cum-max ( seq -- seq' )
340 dup ?first [ max ] accumulate* ;
342 : entropy ( probabilities -- n )
343 dup sum '[ _ / dup log * ] map-sum neg ;
345 : maximum-entropy ( probabilities -- n )
348 : normalized-entropy ( probabilities -- n )
349 [ entropy ] [ maximum-entropy ] bi / ;
351 : binary-entropy ( p -- h )
352 [ dup log * ] [ 1 swap - dup log * ] bi + neg 2 log / ;
354 : standardize ( u -- v )
355 [ demean ] [ sample-std ] bi [ v/n ] unless-zero ;
357 : standardize-2d ( u -- v )
358 flip [ standardize ] map flip ;
360 : differences ( u -- v )
361 [ rest-slice ] keep v- ;
364 dup minmax over - [ v-n ] [ v/n ] bi* ;
366 : rankings ( histogram -- assoc )
367 sort-keys 0 swap [ rot [ + ] keep swapd ] H{ } assoc-map-as nip ;
369 : rank-values ( seq -- seq' )
370 dup histogram rankings '[ _ at ] map ;
372 : z-score ( seq -- n )
373 [ demean ] [ sample-std ] bi v/n ;
375 : dcg ( scores -- dcg )
376 dup length 1 + 2 swap [a,b] [ log 2 log /f ] map v/ sum ;
378 : ndcg ( scores -- ndcg )
380 [ dcg ] [ natural-sort <reversed> dcg ] bi /f