diff --git a/Homework4/README.md b/Homework4/README.md index 929e3ce..d9ad7f3 100644 --- a/Homework4/README.md +++ b/Homework4/README.md @@ -1,65 +1,65 @@ # SP4E - Homework 4 ---- ## General Info This file provides a brief documentation and information related to the fourth (last) Homework of the course "Scientific Programming for Engineers", fall 2019. This homework is done by **O. Ashtari** and **A. Sieber**. Last update: 15.01.2020 ---- ## Project Description The first part of the Homework aims at using the `Pybind11` library in order to create Python binding for the C++ particles code. The ultimate goal of this part consists in allowing the user to run this code through a Python interface. The second part of the Homework focuses on the planet trajectory simulation module of the particles code. There, a Python optimization routine is written which main purpose is to modify the initial planets states (one planet at a time) in order to minimize the error of their computed trajectories when compared to reference trajectories. ---- ## Executable Files ---- ## Optimization routine The optimization routine `error_minimization.py` is intended to minimize the error on Mercury trajectory (or any other planet of the solar system) by scaling the initial velocity of the planet of interest. The minimization is performed by computing the error between the computed planet trajectory and a reference planet trajectory. This routine calls `main.py`, therefore, in order to run it, the user must first make sure she/he can launch the particles code through the python interface. Once the requirement is satisfied the `error_minimization.py` routine can be called as follow: ``` -$ python3 directory_comp directory_ref planet input_file scale nb_steps freq +$ python3 error_minimization.py directory_comp directory_ref planet input_file scale nb_steps freq ``` where: * `directory_comp`, the path to the directory containing the computed trajectory data. * `directory_ref`, the path to the directory containing the reference trajectory data. * `planet`, the planet of interest. * `input_file`, the file containing the initial state of the different planets. * `scale`, the initial scaling factor (initial guess of the minimization). * `nb_steps`, the number of time steps of the simulation within the particles code. * `freq`, the dumping frequency at which the particles code writes outputs. As an example a minimization on Mercury trajectory over 365 days with a dumping frequency of 1 day could be run as such: ``` -$ python3 dumps trajectories mercury init.csv 1.0 365 1 +$ python3 error_minimization.py dumps trajectories mercury init.csv 1.0 365 1 ``` The `error_minimization.py` routine then prints the scaling factor minimizing the error, the value of this error and the amount of minimization iterations needed to reach an optimum. It moreover plots the evolution of the error versus the scaling factor as shown in the link below: ![Optimization on Mercury initial velocity](/diffusion/9482/browse/master/Homework4/optimization_example/minimization_evolution.png) The above graph shows that the optimized scaling is evaluated at 0.3989. Comparing this result to the exact scaling of 0.4 (found by dividing the reference initial velocity by the one stored in `init.csv`) leads to a relative error of less than 0.3% in the optimization procedure. ---- ## Comment on why and how createSimulation function is overloaded To comment on this function overload, let's first discuss what is the role of `createComputes` in the whole code. In the `ParticlesFactoryInterface` class, `createComputes` is a Real-to-voind function which is to be defined later, if needed. In the constructor of `MaterialPointsFactory`, `createComputes` is defined to be the default function (i.e. to be `createDefaultComputes`.) In the default function, `ComputeTemperature` is added to the system evolution object. Similarly in the `PlanetsFactory` where `createComputes` is defined to be `createDefaultComputes` in which `verlet` is added to the system evolution object. As a summary, by constructing an object of type `MaterialPointsFactory` or `PlanetsFactory`, `createComputes` takes a default definition. It is then used inside the first definition of `createSimulation` method whose arguments are file name and time step size. `createSimulation` is overloaded such that `createComputes` is redefined by receiving and substituting the python functor with the desired functionality. The python object (functor) includes initializations and adding compute objects explained above. The whole idea behind overloading `createSimulation` and using the templated functor is to give the flexibility to the python user to easily tailor computing cobjects (i.e. `ComputeTemperature` or `verlet` etc.) based on his/her needs. The reason is that when a C++ code is wrapped to be used through python, the C++ side is supposed to not be touched anymore. However, default settings for material point, for instance, is hard-coded inside its factory. Therefore, to change default values, the python user should manipulate the C++ files. Or assume, like the previous homework, one wants to get length, conductivity, etc. from command line arguments. Or one wants to develop the code such that these variables are read from a file. Using the overloaded `createSimulation` with having very little insight into how the C++ code is designed, one can do all these by just writing few lines in python, with neither changing any C++ file nor editing the binding file.