-
Notifications
You must be signed in to change notification settings - Fork 87
/
nlp.jl
412 lines (320 loc) · 12.6 KB
/
nlp.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
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
# Copyright (c) 2017: Miles Lubin and contributors
# Copyright (c) 2017: Google Inc.
#
# Use of this source code is governed by an MIT-style license that can be found
# in the LICENSE.md file or at https://opensource.org/licenses/MIT.
# This API is imported from MathProgBase and is meant to be transitional.
# A more ideal API would pass expressions (with vector-valued nodes) directly
# let the solver call out to AD tools.
# NLP = Nonlinear Programming.
# The non-MOI-like name is intentional because this part of the API is not
# MOI-like.
"""
AbstractNLPEvaluator
Abstract supertype for the callback object that is used to query function
values, derivatives, and expression graphs.
It is used in [`NLPBlockData`](@ref).
"""
abstract type AbstractNLPEvaluator end
"""
NLPBlock()
An [`AbstractModelAttribute`](@ref) that stores an [`NLPBlockData`](@ref),
representing a set of nonlinear constraints, and optionally a nonlinear
objective.
"""
struct NLPBlock <: AbstractModelAttribute end
"""
NLPBlockDual(result_index::Int = 1)
An [`AbstractModelAttribute`](@ref) for the Lagrange multipliers on the
constraints from the [`NLPBlock`](@ref) in result `result_index`.
If `result_index` is omitted, it is `1` by default.
"""
struct NLPBlockDual <: AbstractModelAttribute
result_index::Int
NLPBlockDual(result_index::Int = 1) = new(result_index)
end
is_set_by_optimize(::NLPBlockDual) = true
"""
NLPBlockDualStart()
An [`AbstractModelAttribute`](@ref) for the initial assignment of the Lagrange
multipliers on the constraints from the [`NLPBlock`](@ref) that the solver may
use to warm-start the solve.
"""
struct NLPBlockDualStart <: AbstractModelAttribute end
"""
NLPBoundsPair(lower::Float64, upper::Float64)
A struct holding a pair of lower and upper bounds.
`-Inf` and `Inf` can be used to indicate no lower or upper bound, respectively.
"""
struct NLPBoundsPair
lower::Float64
upper::Float64
end
"""
struct NLPBlockData
constraint_bounds::Vector{NLPBoundsPair}
evaluator::AbstractNLPEvaluator
has_objective::Bool
end
A struct encoding a set of nonlinear constraints of the form
``lb \\le g(x) \\le ub`` and, if `has_objective == true`, a nonlinear objective
function ``f(x)``.
Nonlinear objectives *override* any objective set by using the
[`ObjectiveFunction`](@ref) attribute.
The `evaluator` is a callback object that is used to query function values,
derivatives, and expression graphs. If `has_objective == false`, then it is an
error to query properties of the objective function, and in
Hessian-of-the-Lagrangian queries, `σ` must be set to zero.
!!! note
Throughout the evaluator, all variables are ordered according to
[`ListOfVariableIndices`](@ref). Hence, MOI copies of nonlinear problems
must not re-order variables.
"""
struct NLPBlockData
constraint_bounds::Vector{NLPBoundsPair}
evaluator::AbstractNLPEvaluator
has_objective::Bool
end
"""
initialize(
d::AbstractNLPEvaluator,
requested_features::Vector{Symbol},
)::Nothing
Initialize `d` with the set of features in `requested_features`. Check
[`features_available`](@ref) before calling `initialize` to see what features
are supported by `d`.
!!! warning
This method must be called before any other methods.
## Features
The following features are defined:
* `:Grad`: enables [`eval_objective_gradient`](@ref)
* `:Jac`: enables [`eval_constraint_jacobian`](@ref)
* `:JacVec`: enables [`eval_constraint_jacobian_product`](@ref) and
[`eval_constraint_jacobian_transpose_product`](@ref)
* `:Hess`: enables [`eval_hessian_lagrangian`](@ref)
* `:HessVec`: enables [`eval_hessian_lagrangian_product`](@ref)
* `:ExprGraph`: enables [`objective_expr`](@ref) and [`constraint_expr`](@ref).
In all cases, including when `requested_features` is empty,
[`eval_objective`](@ref) and [`eval_constraint`](@ref) are supported.
## Examples
```julia
MOI.initialize(d, Symbol[])
MOI.initialize(d, [:ExprGraph])
MOI.initialize(d, MOI.features_available(d))
```
"""
function initialize end
"""
features_available(d::AbstractNLPEvaluator)::Vector{Symbol}
Returns the subset of features available for this problem instance.
See [`initialize`](@ref) for the list of defined features.
"""
function features_available end
"""
eval_objective(d::AbstractNLPEvaluator, x::AbstractVector{T})::T where {T}
Evaluate the objective ``f(x)``, returning a scalar value.
"""
function eval_objective end
"""
eval_constraint(d::AbstractNLPEvaluator,
g::AbstractVector{T},
x::AbstractVector{T},
)::Nothing where {T}
Given a set of vector-valued constraints ``l \\le g(x) \\le u``, evaluate the
constraint function ``g(x)``, storing the result in the vector `g`.
## Implementation notes
When implementing this method, you must not assume that `g` is
`Vector{Float64}`, but you may assume that it supports `setindex!` and `length`.
For example, it may be the `view` of a vector.
"""
function eval_constraint end
"""
eval_objective_gradient(
d::AbstractNLPEvaluator,
grad::AbstractVector{T},
x::AbstractVector{T},
)::Nothing where {T}
Evaluate the gradient of the objective function ``grad = \\nabla f(x)`` as a
dense vector, storing the result in the vector `grad`.
## Implementation notes
When implementing this method, you must not assume that `grad` is
`Vector{Float64}`, but you may assume that it supports `setindex!` and `length`.
For example, it may be the `view` of a vector.
"""
function eval_objective_gradient end
"""
jacobian_structure(d::AbstractNLPEvaluator)::Vector{Tuple{Int64,Int64}}
Returns a vector of tuples, `(row, column)`, where each indicates the position
of a structurally nonzero element in the Jacobian matrix:
``J_g(x) = \\left[ \\begin{array}{c} \\nabla g_1(x) \\\\ \\nabla g_2(x) \\\\ \\vdots \\\\ \\nabla g_m(x) \\end{array}\\right],``
where ``g_i`` is the ``i\\text{th}`` component of the nonlinear constraints
``g(x)``.
The indices are not required to be sorted and can contain duplicates, in which
case the solver should combine the corresponding elements by adding them
together.
The sparsity structure is assumed to be independent of the point ``x``.
"""
function jacobian_structure end
"""
hessian_lagrangian_structure(
d::AbstractNLPEvaluator,
)::Vector{Tuple{Int64,Int64}}
Returns a vector of tuples, `(row, column)`, where each indicates the position
of a structurally nonzero element in the Hessian-of-the-Lagrangian matrix:
``\\nabla^2 f(x) + \\sum_{i=1}^m \\nabla^2 g_i(x)``.
The indices are not required to be sorted and can contain duplicates, in which
case the solver should combine the corresponding elements by adding them
together.
Any mix of lower and upper-triangular indices is valid. Elements `(i,j)` and
`(j,i)`, if both present, should be treated as duplicates.
The sparsity structure is assumed to be independent of the point ``x``.
"""
function hessian_lagrangian_structure end
"""
eval_constraint_jacobian(d::AbstractNLPEvaluator,
J::AbstractVector{T},
x::AbstractVector{T},
)::Nothing where {T}
Evaluates the sparse Jacobian matrix
``J_g(x) = \\left[ \\begin{array}{c} \\nabla g_1(x) \\\\ \\nabla g_2(x) \\\\ \\vdots \\\\ \\nabla g_m(x) \\end{array}\\right]``.
The result is stored in the vector `J` in the same order as the indices returned
by [`jacobian_structure`](@ref).
## Implementation notes
When implementing this method, you must not assume that `J` is
`Vector{Float64}`, but you may assume that it supports `setindex!` and `length`.
For example, it may be the `view` of a vector.
"""
function eval_constraint_jacobian end
"""
eval_constraint_jacobian_product(
d::AbstractNLPEvaluator,
y::AbstractVector{T},
x::AbstractVector{T},
w::AbstractVector{T},
)::Nothing where {T}
Computes the Jacobian-vector product ``y = J_g(x)w``, storing the result in the
vector `y`.
The vectors have dimensions such that `length(w) == length(x)`, and `length(y)`
is the number of nonlinear constraints.
## Implementation notes
When implementing this method, you must not assume that `y` is
`Vector{Float64}`, but you may assume that it supports `setindex!` and `length`.
For example, it may be the `view` of a vector.
"""
function eval_constraint_jacobian_product end
"""
eval_constraint_jacobian_transpose_product(
d::AbstractNLPEvaluator,
y::AbstractVector{T},
x::AbstractVector{T},
w::AbstractVector{T},
)::Nothing where {T}
Computes the Jacobian-transpose-vector product ``y = J_g(x)^Tw``, storing the
result in the vector `y`.
The vectors have dimensions such that `length(y) == length(x)`, and `length(w)`
is the number of nonlinear constraints.
## Implementation notes
When implementing this method, you must not assume that `y` is
`Vector{Float64}`, but you may assume that it supports `setindex!` and `length`.
For example, it may be the `view` of a vector.
"""
function eval_constraint_jacobian_transpose_product end
"""
eval_hessian_lagrangian_product(
d::AbstractNLPEvaluator,
h::AbstractVector{T},
x::AbstractVector{T},
v::AbstractVector{T},
σ::T,
μ::AbstractVector{T},
)::Nothing where {T}
Given scalar weight `σ` and vector of constraint weights `μ`,
computes the Hessian-of-the-Lagrangian-vector product
``h = \\left(\\sigma\\nabla^2 f(x) + \\sum_{i=1}^m \\mu_i \\nabla^2 g_i(x)\\right)v``,
storing the result in the vector `h`.
The vectors have dimensions such that `length(h) == length(x) == length(v)`.
## Implementation notes
When implementing this method, you must not assume that `h` is
`Vector{Float64}`, but you may assume that it supports `setindex!` and `length`.
For example, it may be the `view` of a vector.
"""
function eval_hessian_lagrangian_product end
"""
eval_hessian_lagrangian(
d::AbstractNLPEvaluator,
H::AbstractVector{T},
x::AbstractVector{T},
σ::T,
μ::AbstractVector{T},
)::Nothing where {T}
Given scalar weight `σ` and vector of constraint weights `μ`, this function
computes the sparse Hessian-of-the-Lagrangian matrix:
``\\sigma\\nabla^2 f(x) + \\sum_{i=1}^m \\mu_i \\nabla^2 g_i(x)``,
storing the result in the vector `H` in the same order as the indices
returned by [`hessian_lagrangian_structure`](@ref).
## Implementation notes
When implementing this method, you must not assume that `H` is
`Vector{Float64}`, but you may assume that it supports `setindex!` and `length`.
For example, it may be the `view` of a vector.
"""
function eval_hessian_lagrangian end
"""
objective_expr(d::AbstractNLPEvaluator)::Expr
Returns a Julia `Expr` object representing the expression graph of the objective
function.
## Format
The expression has a number of limitations, compared with arbitrary Julia
expressions:
* All sums and products are flattened out as simple `Expr(:+, ...)` and
`Expr(:*, ...)` objects.
* All decision variables must be of the form
`Expr(:ref, :x, MOI.VariableIndex(i))`, where `i` is the ``i``th variable in
[`ListOfVariableIndices`](@ref).
* There are currently no restrictions on recognized functions; typically these
will be built-in Julia functions like `^`, `exp`, `log`, `cos`, `tan`, `sqrt`,
etc., but modeling interfaces may choose to extend these basic functions, or
error if they encounter unsupported functions.
## Examples
The expression ``x_1+\\sin(x_2/\\exp(x_3))`` is represented as
```julia
:(x[MOI.VariableIndex(1)] + sin(x[MOI.VariableIndex(2)] / exp(x[MOI.VariableIndex[3]])))
```
or equivalently
```julia
Expr(
:call,
:+,
Expr(:ref, :x, MOI.VariableIndex(1)),
Expr(
:call,
:/,
Expr(:call, :sin, Expr(:ref, :x, MOI.VariableIndex(2))),
Expr(:call, :exp, Expr(:ref, :x, MOI.VariableIndex(3))),
),
)
```
"""
function objective_expr end
"""
constraint_expr(d::AbstractNLPEvaluator, i::Integer)::Expr
Returns a Julia `Expr` object representing the expression graph for the
``i\\text{th}`` nonlinear constraint.
## Format
The format is the same as [`objective_expr`](@ref), with an additional
comparison operator indicating the sense of and bounds on the constraint.
For single-sided comparisons, the body of the constraint must be on the
left-hand side, and the right-hand side must be a constant.
For double-sided comparisons (that is, ``l \\le f(x) \\le u``), the body of the
constraint must be in the middle, and the left- and right-hand sides must be
constants.
The bounds on the constraints must match the [`NLPBoundsPair`](@ref)s passed to
[`NLPBlockData`](@ref).
## Examples
```julia
:(x[MOI.VariableIndex(1)]^2 <= 1.0)
:(x[MOI.VariableIndex(1)]^2 >= 2.0)
:(x[MOI.VariableIndex(1)]^2 == 3.0)
:(4.0 <= x[MOI.VariableIndex(1)]^2 <= 5.0)
```
"""
function constraint_expr end