/
Changepoint.jl
394 lines (330 loc) · 10 KB
/
Changepoint.jl
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
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
function update_changepoint_model!(m, σ, d, boundaries, np)
@inbounds for i=1:np+1
r = boundaries[i]:boundaries[i+1]
for col = 1:size(d,2)
m[r,col] .= nanmean(view(d,r,col))
σ[r,col] .= nanstd(view(d,r,col))
end
end
end
function update_changepoint_mu!(m, d, boundaries, np)
@inbounds for i=1:np+1
r = boundaries[i]:boundaries[i+1]
for col = 1:size(d,2)
m[r,col] .= nanmean(view(d,r,col))
end
end
end
# # Not currently used
# function update_changepoint_sigma!(σ, d, boundaries, np)
# @inbounds for i=1:np+1
# r = boundaries[i]:boundaries[i+1]
# for col = 1:size(d,2)
# σ[r,col] .= nanstd(view(d,r,col))
# end
# end
# end
"""
```julia
changepoint(data, [sigma], nsteps; np, npmin, npmax)
```
Given an ordered array of `data` points, optionally with uncertainties `sigma`,
use a Markov chain Monte Carlo approach based on that of Gallagher et al., 2010
(10.1016/j.epsl.2011.09.015) to estimate the position (by index) and optionally
number of changepoints that best explain the `data`. Will return the results for
`nsteps` steps of the Markov chain.
Optional keyword arguments:
np
Specify an exact integer number of changepoints.
npmin
nmpax
Specify the maximum and minimum possible integer number of changepoints. If `np`
is not specified, the number of changepoints will allowed to vary freely between
these bounds, as in the "transdimensional" approach of Gallagher et al.
### Examples
```julia
julia> A = [randn(100).-2; randn(100).+2];
julia> dist = changepoint(A, 10000; np=1);
julia> dist[9000:end] # after burnin
1001-element Vector{Int64}:
101
101
101
101
⋮
101
101
101
julia> dist = changepoint(A, ones(size(A)), 10000; np=1)
10000×1 Matrix{Int64}:
61
61
61
61
⋮
101
101
101
```
"""
function changepoint(data::AbstractArray, nsteps::Integer; np::Integer=0, npmin::Integer=0, npmax::Integer=min(size(data,1) ÷ 2, 11))
MOVE = 0.70
BIRTH = 0.15
DEATH = 0.15
DEBUG = false
FORMATTED = true
T = float(eltype(data))
nrows = size(data,1)
ncolumns = size(data,2)
m = similar(data, T)
σ = similar(data, T) #Array{T}(undef, ncolumns)
σₚ = similar(data, T) #Array{T}(undef, ncolumns)
# Number of possible changepoint locations
K = nrows-1
# Parse provided options
if 0 < np <= K
# If np is specified, use that
npmin = npmax = np
else
# Otherwise, ensure all provided values are plausible and go with that
npmax > K && (npmax = K)
npmin < 0 && (npmin = 0)
np = min(max(npmin, 2), npmax)
end
# Allocate output array of changepoints
result = fill(0, nsteps, npmax)
# Create and fill initial boundary point array
boundaries = Array{Int}(undef, K+2)
boundaries[1] = 1
boundaries[np+2] = nrows
boundaries[2:np+1] .= rand(2:nrows-1, np)
boundariesₚ = similar(boundaries)
boundary_sigma = nrows/np
np = count_unique!(view(boundaries,1:np+2)) - 2
# Calculate initial proposal and log likelihood
update_changepoint_model!(m, σ, data, boundaries, np)
ll = normpdf_ll(m, σ, data)
# The actual loop
@inbounds for i = 1:nsteps
# Randomly choose a type of modification to the model
r = rand()
u = rand()
# Update the model with the chosen modification
if r < MOVE && np>0
# Move a changepoint
copyto!(boundariesₚ,1,boundaries,1,np+2)
# Pick which changepoint to move
pick = rand(2:np+1)
# Move the changepoint
boundary_adj = randn()*boundary_sigma
boundary_prop = boundariesₚ[pick] + round(Int, boundary_adj)
# Treat ends of array as periodic boundary conditions
boundariesₚ[pick] = mod(boundary_prop - 1, nrows-1) + 1
# Check if this has caused any redundancies
npₚ = count_unique!(view(boundariesₚ,1:np+2)) - 2
# Update the model
update_changepoint_mu!(m, data, boundariesₚ, npₚ)
# Calculate log likelihood for proposal
if (1 < boundary_prop < nrows)
llₚ = normpdf_ll(m, σ, data)
else
llₚ = -Inf
end
DEBUG && println("Move: llₚ-ll = $llₚ - $ll")
if log(u) < llₚ-ll
DEBUG && println("Accepted!")
ll = llₚ
boundary_sigma = abs(boundary_adj)*2.9
# println("sigma: $boundary_sigma")
copyto!(boundaries,1,boundariesₚ,1,np+2)
# for n=1:np
# print("$(boundariesₚ[n+1]),")
# end
# FORMATTED && print("\n")
end
elseif r < MOVE+BIRTH
# Add a changepoint
if np < npmax
copyto!(boundariesₚ,1,boundaries,1,np+2)
# Propose a new changepoint
boundariesₚ[np+3] = rand(2:nrows-1)
npₚ = count_unique!(view(boundariesₚ,1:np+3)) - 2
# Update the model
# update_changepoint_model!(m, σ, data, boundariesₚ, npₚ)
update_changepoint_mu!(m, data, boundariesₚ, npₚ)
# Calculate log likelihood for proposal
lqz = sum(1 ./ (2*σ.*σ))
llₚ = normpdf_ll(m, σ, data)
DEBUG && println("Birth: -lqz+llₚ-ll = $(-lqz) + $llₚ - $ll")
if log(u) < llₚ-lqz-ll
DEBUG && println("Accepted!")
ll = llₚ
np = npₚ
copyto!(boundaries,1,boundariesₚ,1,np+2)
# for n=1:np
# print("$(boundariesₚ[n+1]),")
# end
# FORMATTED && print("\n")
end
end
elseif r < MOVE+BIRTH+DEATH
# Delete a changepoint
if np > npmin
copyto!(boundariesₚ,1,boundaries,1,np+2)
# Pick which changepoint to delete
pick = rand(2:np+1)
boundariesₚ[pick]=boundariesₚ[pick+1]
npₚ = count_unique!(view(boundariesₚ,1:np+2)) - 2
# Update the model
# update_changepoint_model!(m, σ, data, boundariesₚ, npₚ)
update_changepoint_mu!(m, data, boundariesₚ, npₚ)
# Calculate log likelihood for proposal
llₚ = normpdf_ll(m, σ, data)
lqz = sum(1 ./ (2*σ.*σ))
DEBUG && println("Death: lqz+llₚ-ll = $lqz + $llₚ - $ll")
if log(u) < llₚ+lqz-ll
DEBUG && println("Accepted!")
ll = llₚ
np = npₚ
copyto!(boundaries,1,boundariesₚ,1,np+2)
# for n=1:np
# print("$(boundariesₚ[n+1]),")
# end
# FORMATTED && print("\n")
end
end
end
# Record results
result[i, 1:np] .= boundaries[2:(np+1)]
end
return result
end
function changepoint(data::AbstractArray, sigma::AbstractArray, nsteps::Integer; np::Integer=0, npmin::Integer=0, npmax::Integer=min(size(data,1) ÷ 2, 11))
MOVE = 0.70
BIRTH = 0.15
DEATH = 0.15
DEBUG = false
FORMATTED = true
T = float(eltype(data))
nrows = size(data,1)
ncolumns = size(data,2)
m = similar(data, T)
# Number of possible changepoint locations
K = nrows-1
# Parse provided options
if 0 < np <= K
# If np is specified, use that
npmin = npmax = np
else
# Otherwise, ensure all provided values are plausible and go with that
npmax > K && (npmax = K)
npmin < 0 && (npmin = 0)
np = min(max(npmin, 2), npmax)
end
# Allocate output array of changepoints
result = fill(0, nsteps, npmax)
# Create and fill initial boundary point array
boundaries = Array{Int}(undef, K+2)
boundaries[1] = 1
boundaries[np+2] = nrows
boundaries[2:np+1] .= rand(2:nrows-1, np)
boundariesₚ = similar(boundaries)
boundary_sigma = nrows/np
np = count_unique!(view(boundaries,1:np+2)) - 2
# Calculate initial proposal and log likelihood
update_changepoint_mu!(m, data, boundaries, np)
ll = normpdf_ll(m, sigma, data)
# The actual loop
@inbounds for i = 1:nsteps
# Randomly choose a type of modification to the model
r = rand()
u = rand()
# Update the model with the chosen modification
if r < MOVE && np>0
# Move a changepoint
copyto!(boundariesₚ,1,boundaries,1,np+2)
# Pick which changepoint to move
pick = rand(2:np+1)
# Move the changepoint
boundary_adj = randn()*boundary_sigma
boundary_prop = boundariesₚ[pick] + round(Int, boundary_adj)
# Treat ends of array as periodic boundary conditions
boundariesₚ[pick] = mod(boundary_prop - 1, nrows-1) + 1
# Check if this has caused any redundancies
npₚ = count_unique!(view(boundariesₚ,1:np+2)) - 2
# Update the model
update_changepoint_mu!(m, data, boundariesₚ, npₚ)
# Calculate log likelihood for proposal
if (1 < boundary_prop < nrows)
llₚ = normpdf_ll(m, sigma, data)
else
llₚ = -Inf
end
DEBUG && println("Move: llₚ-ll = $llₚ - $ll")
if log(u) < llₚ-ll
DEBUG && println("Accepted!")
ll = llₚ
boundary_sigma = abs(boundary_adj)*2.9
# println("sigma: $boundary_sigma")
copyto!(boundaries,1,boundariesₚ,1,np+2)
# for n=1:np
# print("$(boundariesₚ[n+1]),")
# end
# FORMATTED && print("\n")
end
elseif r < MOVE+BIRTH
# Add a changepoint
if np < npmax
copyto!(boundariesₚ,1,boundaries,1,np+2)
# Propose a new changepoint
boundariesₚ[np+3] = rand(2:nrows-1)
npₚ = count_unique!(view(boundariesₚ,1:np+3)) - 2
# Update the model
update_changepoint_mu!(m, data, boundariesₚ, npₚ)
# Calculate log likelihood for proposal
lqz = sum(1 ./ (2*sigma.*sigma))
llₚ = normpdf_ll(m, sigma, data)
DEBUG && println("Birth: -lqz+llₚ-ll = $(-lqz) + $llₚ - $ll")
if log(u) < llₚ-lqz-ll
DEBUG && println("Accepted!")
ll = llₚ
np = npₚ
copyto!(boundaries,1,boundariesₚ,1,np+2)
# for n=1:np
# print("$(boundariesₚ[n+1]),")
# end
# FORMATTED && print("\n")
end
end
elseif r < MOVE+BIRTH+DEATH
# Delete a changepoint
if np > npmin
copyto!(boundariesₚ,1,boundaries,1,np+2)
# Pick which changepoint to delete
pick = rand(2:np+1)
boundariesₚ[pick]=boundariesₚ[pick+1]
npₚ = count_unique!(view(boundariesₚ,1:np+2)) - 2
# Update the model
update_changepoint_mu!(m, data, boundariesₚ, npₚ)
# Calculate log likelihood for proposal
llₚ = normpdf_ll(m, sigma, data)
lqz = sum(1 ./ (2*sigma.*sigma))
DEBUG && println("Death: lqz+llₚ-ll = $lqz + $llₚ - $ll")
if log(u) < llₚ+lqz-ll
DEBUG && println("Accepted!")
ll = llₚ
np = npₚ
copyto!(boundaries,1,boundariesₚ,1,np+2)
# for n=1:np
# print("$(boundariesₚ[n+1]),")
# end
# FORMATTED && print("\n")
end
end
end
# Record results
result[i, 1:np] .= boundaries[2:(np+1)]
end
return result
end
export changepoint