@casadi/casadi-wasm
Version:
CasADi — symbolic framework for algorithmic differentiation and numerical optimization, compiled to WebAssembly. Runs in Node.js with on-demand solver plugins (ipopt, fatrop, sundials, ...).
126 lines (106 loc) • 5.51 kB
JavaScript
//
// MIT No Attribution
//
// Copyright (C) 2010-2023 Joel Andersson, Joris Gillis, Moritz Diehl, KU Leuven.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy of this
// software and associated documentation files (the "Software"), to deal in the Software
// without restriction, including without limitation the rights to use, copy, modify,
// merge, publish, distribute, sublicense, and/or sell copies of the Software, and to
// permit persons to whom the Software is furnished to do so.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED,
// INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A
// PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
// HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
// SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
//
//
// JS port of docs/examples/python/sensitivity_analysis.py.
//
// Forward / adjoint / second-order integrator sensitivities obtained with
// Function.factory and compared against finite differences.
//
// JS notes (see README.md):
// * The Python script also loops over a DAE example with the "idas" and
// "cvodes" integrators and a "kinsol" root-finder; all three abort the
// wasm runtime, so only the ODE (rocket) example is ported here, run
// with the "rk" and "collocation" integrators (which support sensitivity
// factories just fine).
async function example(M, log) {
log("Testing sensitivity analysis in CasADi");
log("******");
log("Testing ODE example");
// Time
const t = M.SX.sym("t");
// Parameter
const u = M.SX.sym("u");
// Differential states
const s = M.SX.sym("s"), v = M.SX.sym("v"), m = M.SX.sym("m");
const x = M.vertcat(s, v, m);
// Constants
const alpha = 0.05; // friction
const beta = 0.1; // fuel consumption rate
// Differential equation
const ode = M.vertcat(
v,
M.rdivide(M.minus(u, M.times(alpha, M.times(v, v))), m),
M.times(-beta, M.times(u, u)));
// Quadrature
const quad = M.plus(M.power(v, 3), M.power(M.minus(M.minus(3, M.sin(t)), u), 2));
// DAE callback
const dae = { t: t, x: x, p: u, ode: ode, quad: quad };
const tf = 0.5; // Time length
const x0 = M.DM([0, 0, 1]); // Initial position
const u0 = 0.4; // Parameter
const h = 0.001; // FD step
const fmt = (dm) => dm.nonzeros().map((z) => z.toPrecision(8)).join(", ");
for (const MyIntegrator of ["rk", "collocation"]) {
log("========");
log("Integrator: " + MyIntegrator);
log("========");
// Integrator
const I = M.integrator("I", MyIntegrator, dae, 0, tf, { number_of_finite_elements: 100 });
// Integrate to get results
let res = I.call({ x0: x0, p: u0 });
const xf = res["xf"], qf = res["qf"];
log("Unperturbed solution: xf = [" + fmt(xf) + "], qf = [" + fmt(qf) + "]");
// Finite-difference approximation
res = I.call({ x0: x0, p: u0 + h });
const fd_xf = M.rdivide(M.minus(res["xf"], xf), h);
const fd_qf = M.rdivide(M.minus(res["qf"], qf), h);
log("Finite difference: d(xf)/d(p) = [" + fmt(fd_xf) + "], d(qf)/d(p) = [" + fmt(fd_qf) + "]");
// Forward sensitivities
const I_fwd = I.factory("I_fwd", ["x0", "z0", "p", "fwd:p"], ["fwd:xf", "fwd:qf"]);
res = I_fwd.call({ x0: x0, p: u0, fwd_p: 1 });
log("Forward: d(xf)/d(p) = [" + fmt(res["fwd_xf"]) + "], d(qf)/d(p) = [" + fmt(res["fwd_qf"]) + "]");
// Adjoint sensitivities
const I_adj = I.factory("I_adj", ["x0", "z0", "p", "adj:qf"], ["adj:x0", "adj:p"]);
res = I_adj.call({ x0: x0, p: u0, adj_qf: 1 });
let adj_x0 = res["adj_x0"], adj_p = res["adj_p"];
log("Adjoint: d(qf)/d(x0) = [" + fmt(adj_x0) + "], d(qf)/d(p) = [" + fmt(adj_p) + "]");
// FD of adjoint -> second order
res = I_adj.call({ x0: x0, p: u0 + h, adj_qf: 1 });
const fd_adj_x0 = M.rdivide(M.minus(res["adj_x0"], adj_x0), h);
const fd_adj_p = M.rdivide(M.minus(res["adj_p"], adj_p), h);
log("FD of adjoint: d2(qf)/d(x0)d(p) = [" + fmt(fd_adj_x0) + "], d2(qf)/d(p)d(p) = [" + fmt(fd_adj_p) + "]");
// Forward over adjoint -> second order
const I_foa = I_adj.factory("I_foa", ["x0", "z0", "p", "adj_qf", "fwd:p"], ["fwd:adj_x0", "fwd:adj_p"]);
res = I_foa.call({ x0: x0, p: u0, adj_qf: 1, fwd_p: 1 });
log("Forward over adjoint: d2(qf)/d(x0)d(p) = [" + fmt(res["fwd_adj_x0"]) + "], d2(qf)/d(p)d(p) = [" + fmt(res["fwd_adj_p"]) + "]");
// Adjoint over adjoint -> second order
const I_aoa = I_adj.factory("I_aoa", ["x0", "z0", "p", "adj_qf", "adj:adj_p"], ["adj:x0", "adj:p"]);
res = I_aoa.call({ x0: x0, p: u0, adj_qf: 1, adj_adj_p: 1 });
log("Adjoint over adjoint: d2(qf)/d(x0)d(p) = [" + fmt(res["adj_x0"]) + "], d2(qf)/d(p)d(p) = [" + fmt(res["adj_p"]) + "]");
}
}
if (typeof require !== "undefined" && typeof module !== "undefined" && require.main === module) {
const path = require("path");
const casadiPath = process.env.CASADI_JS
|| path.resolve(__dirname, "../../../build-wasm/swig/wasm-js/casadi.js");
require(casadiPath)()
.then((M) => example(M, (...a) => console.log(...a)))
.catch((e) => { console.error("FATAL:", e.message || e); process.exit(1); });
}
if (typeof module !== "undefined" && module.exports) module.exports = example;