Homework4

To consume the code

Be sure to clone the submodules with the repository.

cd homework4/starting_point
cmake -S . -B build && cmake --build build && python build/main.py 365 1 "init.csv" "planet" 1

First part - Pybind: Python bindings for Particles Code

Exercise 1: Factory interface

**1.2 In class ParticlesFactory, createSimulation function has been overloaded to take functor as one of its argument. Comment on what this function is doing? ** The overloading allows to specify extra info for which type of compute class and values are used in the creation of the evolution system. The basic function presents already a call of default instance of compute objects.

we followed the pybind11 docu for overloading and we tried to implement this overload wrap for the createSimulation():

.def("createSimulation", py::overload_cast<const std::string&, Real>(&ParticlesFactoryInterface::createSimulation),
          py::arg("fname"), py::arg("timestep"), py::return_value_policy::reference)
.def("createSimulation", py::overload_cast<const std::string&, Real, py::function>(&ParticlesFactoryInterface::createSimulation<py::function>),
          py::arg("fname"), py::arg("timestep"), py::arg("create_computes"), py::return_value_policy::reference);

But we recived the error that it was an unspecified overload the first one, finally we decided to declare it as a simple function against the pybind11 documentation. Why this did not work?

Exercise 2: Compute

2.2 How will you ensure that references to Compute objects type are correctly managed in the python bindings? The memory for these object's bindings is done via shared pointers this ensures that once both the C++ and Python memory management counters reach 0, the memory garbage collector takes care of it. As a general rule, this is how every object with no particular instantiation rules should be wrapped.

Second part: Particle trajectory optimization

Exercise 5: Compute the error of the simulation

To run the code:

cd homework4/starting_point
python compute_error_simulation.py "mercury" "trajectories" "build/dumps"

Exercise 6: Launching the particle code from python by generating the input

To run the code:

cd homework4/starting_point
python generate_input_launcher.py 2 "mercury" "init.csv" "dumps" "build/dumps" "outVelocity.csv"

with args as (in order): scale : the scaling factor planet_name : the name of the planet to analyse input_filename : the starting input filename to modify the velocity directory : the generated dumps from this script directory_dir : the previously generated dumps from the vanilla launcher output_filename : the single output with modified data for veloctiy

Generated results can be found in the directory: homework/starting_point/dumps.

Exercice 7: Optimization

We did not menage to terminate the part 7 for the optimization in time. The optimization code is present but not working.

ibois-epfl / sp4e-andrea-petras