-
Notifications
You must be signed in to change notification settings - Fork 0
/
floydrivest.go
280 lines (238 loc) · 7.4 KB
/
floydrivest.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
package kth
import (
"cmp"
"math"
"sort"
)
// FloydRivest implements the Floyd-Rivest selection algorithm to find the k-th smallest elements.
// It typically makes fewer comparisons than other selection algorithms by narrowing the search range
// based on order statistics estimates before partitioning.
func FloydRivest(data sort.Interface, k int) {
n := data.Len()
if k < 1 || k > n {
return
}
floydRivest(data, 0, n-1, k-1)
}
// rangeNarrowingThreshold represents the size above which we narrow the search range
// using order statistics estimates before partitioning.
const rangeNarrowingThreshold = 600
// The Floyd-Rivest algorithm maintains two core invariants:
// 1. After each iteration, elements known to be less than the k-th element
// are to its left, and elements known to be greater are to its right
// 2. The k-th element is always within our search bounds [left, right]
//
// The algorithm combines two strategies with proven optimality:
// - Range narrowing based on order statistics for large arrays
// - Efficient partitioning for reduced ranges
func floydRivest(data sort.Interface, left, right, k int) {
// Loop invariant: k-th element is within [left, right]
for right > left {
size := right - left
// For large arrays, attempt to narrow the search range
// This is a heuristic that can fail with pathological data distributions
// but the algorithm remains correct due to the outer loop's invariants
if size > rangeNarrowingThreshold {
n := size + 1
i := k - left + 1
// Statistical parameters for range estimation
z := math.Log(float64(n))
s := 0.5 * math.Exp(2*z/3)
// Standard deviation for range width
sd := 0.5 * math.Sqrt(z*s*(float64(n)-s)/float64(n))
if i < n/2 {
sd *= -1.0
}
// Estimate new boundaries
// These could be wrong for non-uniform distributions
// but the algorithm will still converge correctly
newLeft := max(left, int(float64(k)-float64(i)*s/float64(n)+sd))
newRight := min(right, int(float64(k)+float64(n-i)*s/float64(n)+sd))
floydRivest(data, newLeft, newRight, k)
}
// Partitioning section
// Core invariants during partitioning:
// 1. Elements strictly < pivot end up left of final pivot position
// 2. Elements strictly > pivot end up right of final pivot position
// 3. Elements = pivot cluster around final pivot position
i, j := left, right
// Initial pivot selection and positioning
// Using the larger value as pivot creates an important property:
// - When we encounter elements equal to the original pivot value,
// they are guaranteed to be strictly less than our chosen pivot
// - This removes ambiguity in the partitioning process and
// prevents infinite loops when many duplicates are present
data.Swap(left, k)
swap := data.Less(left, right)
pivot := right
if swap {
data.Swap(left, right)
pivot = left
}
// Main partitioning loop
// Initial swap before moving pointers serves multiple purposes:
// 1. Elements at i,j have been compared to pivot but not yet placed
// 2. Swapping before moving pointers ensures we don't skip elements
// 3. This ordering guarantees progress even when many elements equal pivot
for i < j {
data.Swap(i, j)
i++
j--
// Forward and backward scans maintain strict invariants:
// - All elements before i are strictly < pivot_value
// - All elements after j are strictly > pivot_value
// - Elements between i and j are yet to be classified
for data.Less(i, pivot) {
i++
}
for data.Less(pivot, j) {
j--
}
}
// Final pivot positioning requires careful handling because:
// 1. When pointers cross (i ≥ j), we have three regions:
// [left...j] < pivot_value
// [j...i] = pivot_value
// [i...right] > pivot_value
// 2. The pivot_value itself is still at either left or right
// 3. It must end up at the boundary between < and = regions
if swap {
data.Swap(left, j)
} else {
j++
data.Swap(right, j)
}
// Range reduction implements the key efficiency of selection:
// 1. After partitioning, j is the exact count of elements ≤ pivot_value
// 2. This gives us perfect information for reducing the search range:
// - If k ≤ j: k-th element must be in left portion
// - If k > j: k-th element must be in right portion
// 3. The reduction is optimal because:
// - We never discard elements that could be the k-th element
// - We always discard elements that cannot be the k-th element
if j <= k {
left = j + 1
}
if k <= j {
right = j - 1
}
}
}
// FloydRivestOrdered is a specialized version of FloydRivest that works with slices of
// ordered types (i.e. types that implement the cmp.Ordered interface).
func FloydRivestOrdered[T cmp.Ordered](data []T, k int) {
n := len(data)
if k < 1 || k > n {
return
}
floydRivestOrdered(data, 0, n-1, k-1)
}
func floydRivestOrdered[T cmp.Ordered](data []T, left, right, k int) {
for right > left {
size := right - left
if size > rangeNarrowingThreshold {
n := size + 1
i := k - left + 1
z := math.Log(float64(n))
s := 0.5 * math.Exp(2*z/3)
sd := 0.5 * math.Sqrt(z*s*(float64(n)-s)/float64(n))
if i < n/2 {
sd *= -1.0
}
newLeft := max(left, int(float64(k)-float64(i)*s/float64(n)+sd))
newRight := min(right, int(float64(k)+float64(n-i)*s/float64(n)+sd))
floydRivestOrdered(data, newLeft, newRight, k)
}
i, j := left, right
// Initial pivot selection and positioning
data[left], data[k] = data[k], data[left]
swap := data[left] < data[right]
pivot := right
if swap {
data[left], data[right] = data[right], data[left]
pivot = left
}
for i < j {
data[i], data[j] = data[j], data[i]
i++
j--
for data[i] < data[pivot] {
i++
}
for data[pivot] < data[j] {
j--
}
}
if swap {
data[left], data[j] = data[j], data[left]
} else {
j++
data[right], data[j] = data[j], data[right]
}
if j <= k {
left = j + 1
}
if k <= j {
right = j - 1
}
}
}
// FloydRivestFunc is a generic version of FloydRivest that allows the caller to provide
// a custom comparison function to determine the order of elements.
func FloydRivestFunc[E any](data []E, k int, less func(a, b E) bool) {
n := len(data)
if k < 1 || k > n {
return
}
floydRivestFunc(data, 0, n-1, k-1, less)
}
func floydRivestFunc[E any](data []E, left, right, k int, less func(a, b E) bool) {
for right > left {
size := right - left
if size > rangeNarrowingThreshold {
n := size + 1
i := k - left + 1
z := math.Log(float64(n))
s := 0.5 * math.Exp(2*z/3)
sd := 0.5 * math.Sqrt(z*s*(float64(n)-s)/float64(n))
if i < n/2 {
sd *= -1.0
}
newLeft := max(left, int(float64(k)-float64(i)*s/float64(n)+sd))
newRight := min(right, int(float64(k)+float64(n-i)*s/float64(n)+sd))
floydRivestFunc(data, newLeft, newRight, k, less)
}
i, j := left, right
// Initial pivot selection and positioning
data[left], data[k] = data[k], data[left]
swap := less(data[left], data[right])
pivot := right
if swap {
data[left], data[right] = data[right], data[left]
pivot = left
}
for i < j {
data[i], data[j] = data[j], data[i]
i++
j--
for less(data[i], data[pivot]) {
i++
}
for less(data[pivot], data[j]) {
j--
}
}
if swap {
data[left], data[j] = data[j], data[left]
} else {
j++
data[right], data[j] = data[j], data[right]
}
if j <= k {
left = j + 1
}
if k <= j {
right = j - 1
}
}
}