Validate dimension of Constraint #420
Description
Background
In Python Constraint does not have a method like dimension; this is defined in derived classes like Rectangle. This was deliberate so we can do
constraint = Ball2(radius=1.0)without having to specify a size. As a result, sometimes the dimension is not determinable.
The issue
The following works and does not raise an exception:
import sys
import os
import casadi.casadi as cs
import opengen as og
optimizers_dir = "my_optimizers"
optimizer_name = "rosenbrock"
def get_open_local_absolute_path():
return os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "rust"))
nu, np = 5, 2
u = cs.SX.sym("u", nu) # decision variable (nu = 5)
p = cs.SX.sym("p", 2) # parameter (np = 2)
phi = og.functions.rosenbrock(u, p) + 1500*cs.sum1(u)
# c_f1 = cs.vertcat(p[0] * u[0] - u[1], p[1]*u[2] - u[3] + 0.1)
c_f2 = cs.vertcat(0.2 + 1.5 * u[0] - u[1], u[2] - u[3] - 0.1)
bounds = og.constraints.Rectangle(xmin=[1, 2], xmax=[3, 4])
problem = og.builder.Problem(u, p, phi) \
.with_penalty_constraints(c_f2)\
.with_constraints(bounds)
meta = og.config.OptimizerMeta()\
.with_optimizer_name(optimizer_name)
build_config = og.config.BuildConfiguration() \
.with_build_directory(optimizers_dir) \
.with_build_mode(og.config.BuildConfiguration.DEBUG_MODE) \
.with_build_python_bindings() \
.with_open_version(local_path=get_open_local_absolute_path())
solver_cfg = og.config.SolverConfiguration() \
.with_tolerance(1e-6) \
.with_max_inner_iterations(1000) \
.with_max_outer_iterations(30) \
.with_penalty_weight_update_factor(1.5) \
.with_preconditioning(True)
builder = og.builder.OpEnOptimizerBuilder(problem,
meta,
build_config,
solver_cfg)
builder.build()Note that we have a decision variable of dimension 5:
u = cs.SX.sym("u", nu)and a set of constraints of dimension 2:
bounds = og.constraints.Rectangle(xmin=[1, 2], xmax=[3, 4])Then, the Rust code panics:
thread '<unnamed>' (6694911) panicked at /Users/pantelis/Documents/Development/optimization-engine/rust/src/constraints/rectangle.rs:76:13:
assertion `left == right` failed: x and xmin have incompatible dimensions
left: 5
right: 2
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
Traceback (most recent call last):
File "/Users/pantelis/Documents/Development/optimization-engine/python/main.py", line 52, in <module>
response = solver.run(p=[0.5, 8.5], initial_guess=[1, 2, 3, 4, 0]) # SolverResponse
pyo3_runtime.PanicException: assertion `left == right` failed: x and xmin have incompatible dimensions
left: 5
right: 2
Solution
Mainly we need to implement dimension() -> int for all constraints. This method will sometimes have to return None (e.g., in case of Ball*). Then we need to verify the user input.