NLPExample

NLP Example

Initial setup

Before starting the example, you'll have to setup some variables.

In the file create_problem.m, set the value of FROST_PATH to be the path to Frost installation.

In the file CMakeLists.txt, set the value of C_FROST_ROOT_PATH to be the path to C-Frost root folder. Note that if you're using WSL the path would look like /mnt/c/....

Running the example

• Run create_problem.m.
• Afterwards you'll see a new folder created called c_code. This folder will contain the c++ Ipopt code.
• In a bash directory, go to the folder c_code.
• Run the following:

  mkdir build
  cd build
  cmake ..
  make
  make install
  cd ..
  ./program --initial 'res/init.json' --options '../ipopt.opt' --data 'res/data.json' --bounds 'res/bounds.json' --output '../output.json'
  

• You'll see a file called output.json created in the main example folder.
• In matlab, run sol = loadjson('output.json');. Now you have the solution from the problem in the variable sol.

Explanation of the code

Anything prior to the line,

solver = IpoptApplication(nlp, extraOpts);

including the line itself, is the same as when using the typical Frost.

The section (C-Frost specific code) Create c ipopt problem includes the unique commands needed to be called when wanting to run the Ipopt problem natively on C++.

Here is a description of some important lines:

---

[funcs] = frost_c.getAllFuncs(solver);
frost_c.createFunctionListHeader(funcs, src_path, include_dir);
frost_c.createIncludeHeader(funcs, include_dir);

These three lines do the following, [funcs] = frost_c.getAllFuncs(solver); returns the names of all the important mathematical expressions/functions that will be generated. Then, frost_c.createFunctionListHeader(funcs, src_path, include_dir); and frost_c.createIncludeHeader(funcs, include_dir); generate some necessary C files that are dependent on the content of funcs (both the names of the functions and the order of the elements in the list). Typically, if a new constraint/function is created, or if the index of an element in the list changes, run all three lines of code. Then for convinence, save the content of funcs (just to make sure you're always using the same order of elements).

---

frost_c.createConstraints(nlp,[],[],src_gen_path, include_dir);
frost_c.createObjectives(nlp,[],[],src_gen_path, include_dir);

These two lines generate the actual mathematical expressions of all the constraints as C code. Their usage is identical to:

nlp.compileConstraint(export_path);
nlp.compileObjective(export_path);

Except you have to provide two paths instead of one. Note that you can also ignore some functions or generate specific functions.

---

frost_c.createDataFile(solver, funcs, res_path, 'data');
frost_c.createBoundsFile(solver, funcs, res_path, 'bounds');
frost_c.createInitialGuess(solver, res_path);

These three lines generate three json files: a data file, a bounds file, and an init file.

• data.json contain information about all the constraint/objective functions. This includes the indices of the dependant varaibles and the list of functions (including which functions from the funcs list each constraint/objective must call).
• bounds.json contain the bounds for each variable and constraint. In addition, the file includes all Auxilary Data used in the problem.
• init.json is simply a vector containing the initial condition.

Generally, if only the bounds have changed, you only need new bounds files. If a constraint expression was modified (or if you removed a constraint), you need to generate new a new data file (and a new bounds file). If new constraints were created (or some were renamed), you'll need to run:

[funcs] = frost_c.getAllFuncs(solver);
frost_c.createFunctionListHeader(funcs, src_path, include_dir);
frost_c.createIncludeHeader(funcs, include_dir);

then generate a new data and bounds file.