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
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 dup zero? [ 2drop 0 ] [ / ] if ; inline
20 : unbiased-mean ( seq -- x )
23 : sum-of-squares ( seq -- x )
24 [ sq ] map-sum ; inline
26 : sum-of-squared-errors ( seq -- x )
27 [ mean ] keep [ - sq ] with map-sum ; inline
29 : sum-of-absolute-errors ( seq -- x )
30 [ mean ] keep [ - ] with map-sum ; inline
32 : quadratic-mean ( seq -- x ) ! root-mean-square
33 [ sum-of-squares ] [ length ] bi / sqrt ; inline
35 : geometric-mean ( seq -- x )
36 [ length ] [ product ] bi nth-root ; inline
38 : harmonic-mean ( seq -- x )
39 [ recip ] map-sum recip ; inline
41 : contraharmonic-mean ( seq -- x )
42 [ sum-of-squares ] [ sum ] bi / ; inline
46 : trim-points ( p seq -- from to seq )
47 [ length [ * >integer ] keep over - ] keep ;
51 : trimmed-mean ( seq p -- x )
52 swap natural-sort trim-points <slice> mean ;
54 : winsorized-mean ( seq p -- x )
55 swap natural-sort trim-points
57 [ nip dupd nth <array> ]
58 [ [ 1 - ] dip nth <array> ] 3tri
63 :: ((kth-object)) ( seq k nth-quot exchange-quot quot: ( x y -- ? ) -- elt )
64 #! Wirth's method, Algorithm's + Data structues = Programs p. 84
65 k seq bounds-check 2drop
78 [ i seq nth-quot call x quot call ] [ i 1 + i! ] while
79 [ x j seq nth-quot call quot call ] [ j 1 - j! ] while
81 i j seq exchange-quot call
92 : (kth-object) ( seq k nth-quot exchange-quot quot: ( x y -- ? ) -- elt )
93 #! The algorithm modifiers seq, so we clone it
94 [ clone ] 4dip ((kth-object)) ; inline
96 : kth-object-unsafe ( seq k quot: ( x y -- ? ) -- elt )
97 [ [ nth-unsafe ] [ exchange-unsafe ] ] dip (kth-object) ; inline
99 : kth-objects-unsafe ( seq kths quot: ( x y -- ? ) -- elts )
101 '[ [ nth-unsafe ] [ exchange-unsafe ] _ ((kth-object)) ] with map ; inline
105 : kth-object ( seq k quot: ( x y -- ? ) -- elt )
106 [ [ nth ] [ exchange ] ] dip (kth-object) ; inline
108 : kth-objects ( seq kths quot: ( x y -- ? ) -- elts )
110 '[ [ nth ] [ exchange ] _ ((kth-object)) ] with map ; inline
112 : kth-smallests ( seq kths -- elts ) [ < ] kth-objects-unsafe ;
114 : kth-smallest ( seq k -- elt ) [ < ] kth-object-unsafe ;
116 : kth-largests ( seq kths -- elts ) [ > ] kth-objects-unsafe ;
118 : kth-largest ( seq k -- elt ) [ > ] kth-object-unsafe ;
120 : count-relative ( seq k -- lt eq gt )
123 { +lt+ [ [ 1 + ] 2dip ] }
125 { +eq+ [ [ 1 + ] dip ] }
129 : minmax-relative ( seq k -- lt eq gt lt-max gt-min )
130 [ 0 0 0 -1/0. 1/0. ] 2dip '[
132 { +lt+ [ [ 1 + ] 5 ndip '[ _ max ] dip ] }
133 { +gt+ [ [ 1 + ] 3dip min ] }
134 { +eq+ [ [ 1 + ] 4dip drop ] }
138 : lower-median-index ( seq -- n )
140 [ length odd? [ 1 - ] unless ] bi ;
142 : lower-median ( seq -- elt )
143 [ ] [ lower-median-index ] bi kth-smallest ;
145 : upper-median ( seq -- elt )
146 dup midpoint@ kth-smallest ;
148 : medians ( seq -- lower upper )
150 [ [ lower-median-index ] [ midpoint@ ] bi 2array ]
151 bi kth-smallests first2 ;
153 : median ( seq -- x )
154 dup length odd? [ lower-median ] [ medians + 2 / ] if ;
156 ! quantile can be any n-tile. quartile is n = 4, percentile is n = 100
157 ! a,b,c,d parameters, N - number of samples, q is quantile (1/2 for median, 1/4 for 1st quartile)
158 ! http://mathworld.wolfram.com/Quantile.html
160 ! could subtract 1 from a
162 : quantile-x ( a b N q -- x )
163 [ + ] dip * + 1 - ; inline
167 >fraction [ /mod nip ] keep / ; inline
169 :: quantile-indices ( seq qs a b c d -- seq )
170 qs [ [ a b seq length ] dip quantile-x ] map ;
172 :: qabcd ( y-floor y-ceiling x c d -- qabcd )
173 y-floor y-ceiling y-floor - c d x frac * + * + ;
175 :: quantile-abcd ( seq qs a b c d -- quantile )
176 seq qs a b c d quantile-indices :> indices
177 indices [ [ floor ] [ ceiling ] bi 2array ] map
178 concat :> index-pairs
180 seq index-pairs kth-smallests
181 2 group indices [ [ first2 ] dip c d qabcd ] 2map ;
183 : quantile1 ( seq qs -- seq' )
184 0 0 1 0 quantile-abcd ;
186 : quantile3 ( seq qs -- seq' )
187 1/2 0 0 0 quantile-abcd ;
189 : quantile4 ( seq qs -- seq' )
190 0 0 0 1 quantile-abcd ;
192 : quantile5 ( seq qs -- seq' )
193 1/2 0 0 1 quantile-abcd ;
195 : quantile6 ( seq qs -- seq' )
196 0 1 0 1 quantile-abcd ;
198 : quantile7 ( seq qs -- seq' )
199 1 -1 0 1 quantile-abcd ;
201 : quantile8 ( seq qs -- seq' )
202 1/3 1/3 0 1 quantile-abcd ;
204 : quantile9 ( seq qs -- seq' )
205 3/8 1/4 0 1 quantile-abcd ;
207 : quartile ( seq -- seq' )
208 { 1/4 1/2 3/4 } quantile5 ;
212 : (sequence>assoc) ( seq map-quot insert-quot assoc -- assoc )
213 [ swap curry compose each ] keep ; inline
215 : (sequence-index>assoc) ( seq map-quot insert-quot assoc -- assoc )
216 [ swap curry compose each-index ] keep ; inline
220 : sequence>assoc! ( assoc seq map-quot: ( x -- ..y ) insert-quot: ( ..y assoc -- ) -- assoc )
221 4 nrot (sequence>assoc) ; inline
223 : sequence>assoc ( seq map-quot insert-quot exemplar -- assoc )
224 clone (sequence>assoc) ; inline
226 : sequence-index>assoc ( seq map-quot insert-quot exemplar -- assoc )
227 clone (sequence-index>assoc) ; inline
229 : sequence-index>hashtable ( seq map-quot insert-quot -- hashtable )
230 H{ } sequence-index>assoc ; inline
232 : sequence>hashtable ( seq map-quot insert-quot -- hashtable )
233 H{ } sequence>assoc ; inline
235 : histogram! ( hashtable seq -- hashtable )
236 [ ] [ inc-at ] sequence>assoc! ;
238 : histogram-by ( seq quot: ( x -- bin ) -- hashtable )
239 [ inc-at ] sequence>hashtable ; inline
241 : histogram ( seq -- hashtable )
244 : sorted-histogram ( seq -- alist )
245 histogram sort-values ;
247 : normalized-histogram ( seq -- alist )
248 [ histogram ] [ length ] bi '[ _ / ] assoc-map ;
250 : collect-index-by ( seq quot -- hashtable )
251 [ swap ] prepose [ push-at ] sequence-index>hashtable ; inline
253 : collect-by ( seq quot -- hashtable )
254 [ dup ] prepose [ push-at ] sequence>hashtable ; inline
256 : equal-probabilities ( n -- array )
257 dup recip <array> ; inline
261 [ ] [ [ [ second ] bi@ > ] most ] map-reduce first ;
263 : minmax ( seq -- min max )
264 [ first dup ] keep [ [ min ] [ max ] bi-curry bi* ] each ;
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 )
281 var-ddof sqrt ; inline
283 : population-std ( seq -- x ) 0 std-ddof ; inline
285 : sample-std ( seq -- x ) 1 std-ddof ; inline
287 ALIAS: std sample-std
289 : signal-to-noise ( seq -- x ) [ mean ] [ population-std ] bi / ;
291 : mean-dev ( seq -- x ) dup mean v-n vabs mean ;
293 : median-dev ( seq -- x ) dup median v-n vabs mean ;
295 : ste-ddof ( seq n -- x ) '[ _ std-ddof ] [ length ] bi sqrt / ;
297 : population-ste ( seq -- x ) 0 ste-ddof ;
299 : sample-ste ( seq -- x ) 1 ste-ddof ;
301 : ((r)) ( mean(x) mean(y) {x} {y} -- (r) )
302 ! finds sigma((xi-mean(x))(yi-mean(y))
303 0 [ [ [ pick ] dip swap - ] bi@ * + ] 2reduce 2nip ;
305 : (r) ( mean(x) mean(y) {x} {y} sx sy -- r )
306 * recip [ [ ((r)) ] keep length 1 - / ] dip * ;
308 : [r] ( {{x,y}...} -- mean(x) mean(y) {x} {y} sx sy )
309 first2 [ [ [ mean ] bi@ ] 2keep ] 2keep [ population-std ] bi@ ;
311 : r ( {{x,y}...} -- r )
314 : r^2 ( {{x,y}...} -- r )
317 : least-squares ( {{x,y}...} -- alpha beta )
318 [r] { [ 2dup ] [ ] [ ] [ ] [ ] } spread
319 ! stack is mean(x) mean(y) mean(x) mean(y) {x} {y} sx sy
320 [ (r) ] 2keep ! stack is mean(x) mean(y) r sx sy
321 swap / * ! stack is mean(x) mean(y) beta
324 : cov-ddof ( {x} {y} ddof -- cov )
325 [ [ dup mean v-n ] bi@ v* ] dip mean-ddof ;
327 : population-cov ( {x} {y} -- cov ) 0 cov-ddof ; inline
329 : sample-cov ( {x} {y} -- cov ) 1 cov-ddof ; inline
331 : corr-ddof ( {x} {y} n -- corr )
332 [ [ population-cov ] ] dip
333 '[ [ _ var-ddof ] bi@ * sqrt ] 2bi / ;
335 : population-corr ( {x} {y} -- corr ) 0 corr-ddof ; inline
337 : sample-corr ( {x} {y} -- corr ) 1 corr-ddof ; inline
339 : cum-map ( seq identity quot -- seq' )
340 swapd [ dup ] compose map nip ; inline
342 : cum-sum ( seq -- seq' )
345 : cum-sum0 ( seq -- seq' )
346 0 [ + ] accumulate nip ;
348 : cum-product ( seq -- seq' )
351 : cum-count ( seq quot -- seq' )
353 '[ _ call [ 1 + ] when ] cum-map ; inline
355 : cum-min ( seq -- seq' )
356 dup ?first [ min ] cum-map ;
358 : cum-max ( seq -- seq' )
359 dup ?first [ max ] cum-map ;
361 : entropy ( probabilities -- n )
362 dup sum '[ _ / dup log * ] map-sum neg ;
364 : maximum-entropy ( probabilities -- n )
367 : normalized-entropy ( probabilities -- n )
368 [ entropy ] [ maximum-entropy ] bi / ;
370 : binary-entropy ( p -- h )
371 [ dup log * ] [ 1 swap - dup log * ] bi + neg 2 log / ;
373 : standardize ( u -- v )
374 [ dup mean v-n ] [ sample-std ] bi
375 dup zero? [ drop ] [ v/n ] if ;
377 : standardize-2d ( u -- v )
378 flip dup [ [ mean ] [ sample-std ] bi 2array ] map
379 [ [ first v-n ] 2map ] keep [ second v/n ] 2map flip ;
381 : differences ( u -- v )
382 [ 1 tail-slice ] keep [ - ] 2map ;
385 dup minmax over - [ v-n ] [ v/n ] bi* ;
387 : rank-values ( seq -- seq' )
389 [ ] [ length iota ] bi zip sort-keys
390 [ [ first ] bi@ = ] monotonic-split
391 [ values ] map [ 0 [ length + ] accumulate nip ] [ ] bi zip
392 ] [ length f <array> ] bi
393 [ '[ first2 [ _ set-nth ] with each ] each ] keep ;
395 : z-score ( seq -- n )
396 dup [ mean ] [ sample-std ] bi '[ _ - _ / ] map ;