A Threaded PyQt GUI and Particles

I wrote a post about adding a GUI to your command line interface program. Sometimes people just want to click stuff. If you tried it / implemented it yourself, you'd have seen that the program would have gone from responsive to non-responsive when you hit the run button. Why? Because the program can only do ONE think at a time and frankly, keeping the controls available for clicking at all times, updating the graphics and calculating particle movement do not get along nicely. This post is about making it behave like we expect.

I present to you threading. Specifically for PyQt there is the QThread class which is actually discussed rather nicely in this PyQtWiki and this page by Jo Plaete which I used for inspiration.

What complicates the whole scene is that we need to realize that we have two independant things we want to accomplish: 1) move particles and 2) update the graphics. Both independent of the GUI.

I started by making my own class called BaseThread which subclasses QThread. The code is

where I believe the magic for everything not just crashing and burning is the subtle use of the custom boolean flag exiting (which we will use later) and the use of self.wait() in the __del__ method. Again, for details you should read the blog posts mentioned above.

To come around many of the syncronization problems that arises with threads the Qt framework allows us to use custom signals to make everything talk together. Here is my version of a StepThread which defines the run method (you do NEVER call this explicitly!) and a convenience function called simulate to start a simulation for N steps. The run function emits a custom signal and sleeps for a little while before simulating again.

Very similarly we have the DrawThread which emits a signal so we can update the matplotlib surface.

Notice that it updates less frequently. I have yet to figure out something clever in this regard.

Finally, I had to make some tweaks to the Simulator class (nothing very fancy) to hook up our custom signals.

What I need now is a video of it and maybe to implement some different potentials instead of only the non-interacting particles. Stay tuned.

The entire code can be downloaded from the latest gist I made.

Adding a PyQt GUI to your program

I am all in for command line interface (CLI) programs. I'll admit that. They work in terminals. You can run them without an X Server and usually option flags are easy to use and obvious*. But sometimes I'd like to just hit a button. Click and magic happens. No more 200 character command line statement to get the ball rolling. A good example of this is from my colleague Anders Christensen who made a graphical user interface (GUI) for the Phaistos program.

In this post, we'll take the first small steps towards actually making a GUI for a simulation to run the ideal gas in and not just watch some numbers printed in a terminal. I'll be using Qt4 and the PyQt4 bindings. If you have a fairly recent linux distribution, you can apt-get install yourself to success very fast. Remember to also install the Qt-Designer app so you visually can layout your controls.

We'll start out simple by constructing buttons, textedits and a frame. I use a regular QWidget for my main form. The end goal for a user is to push the 1) setup button to initialize particles and 2) push the run button and make the simulation run. The simulation (i.e. the particles) should be displayed in the frame via matplotlib. It looks like this

If you want the resulting simulator.ui code its listed as part the gist for this blog post.

To move on from here, there is a nifty little tool called pyuic4 which will convert your .ui file into a python class with the name Simulator_UI. I always just parse it down to a file called simulator_ui.py. The code is

$ pyuic4 simulator.ui > simulator_ui.py

The simulator_ui class contains the framework for the QWidget form we will be using. The most clever approach I could think of was to subclass the simulator_ui class to separate the "setting up the form" code with the "what happens when I push a button" code. The QWidget I'll use to represent the window the user will see and I call it Simulator. It imports the Simulator_UI class, subclasses it and sets up the GUI. The two buttons are hooked up to either initialize new particles or run for some steps


The bread and butter of this application is the ParticleCanvas. Its tailored to draw and move the particles around. Look at its code


You see that it subclasses the Canvas class which is my best bet on how one should make the code for fast blitting of a matplotlib canvas. A class which subclasses the Canvas class is only responsible for redrawing the actual plot (and do it fast). However,  I would gladly appreciate comments of you have a better way to implement it


Finally, the code that makes it all run is just my main executable which is defined like this


The final result looks like this

We now have a finished simulation GUI. However, the main problem is that the GUI and the simulation code is intimately hooked up so the GUI will freeze at some point. Fortunately, it leaves a new post on the horizon.

*I never understood anything of the find command, however.

Programming the Ideal Gas

You've heard about it before. Many times. It's a model system used a lot in chemistry. The Ideal Gas law: A system where the particles can be treated as non-interacting. How do you program such a system? What do you need to describe it? Given 1) the basic introduction covered by Prof. Jan Jensen in another lecture, and 2) the following assignment (notice it is part 2), here is what the students are supposed to do once this exercise is complete:

1. Take a box with dimensions $(-1<x<1$ and $-1<y<1)$.
2. Make npart particles placed randomly in the right side of a box ($0<x<1$) but evenly from the top through the bottom
3. Give the npart particles random velocities in both directions and make it so they can move in both the positive and negative direction.
4. Move the particles around via the equation: $X_i^{n+1} = X_i^{n}+dt*dX_i^{n}$
5. Keep them inside the box at all times.
6. Plot the result.

The students start out with 1, 2 and 6 presented to them, both for inspiration and because they contain some subtle hints. The code is

One of the tricky parts is knowing "what is the i'th particle" because we need this information to update the system according to the equation above in point 4. Once you realize that you can access this information in the X, Y, dX and dY arrays as X[i] under a loop, then there is really no problem with that point either, for instance, to update all particles according to the equation in 4 in one go, we could write

Assigning a loop to iterate over the particles is not really troublesome, but understanding why is hard - even to explain it to yourself. A simple for-loop might not be the smartest thing to do, but I think it is the easiest one to go about and more importantly I often feel that many things can get lost in the whole numpy magic, especially for newcomers.

The next part is, that we should keep the particles inside the box at all times. The exercise has a little discussion about which version is the better, but in my opinion the particles should always be inside at all times. To keep them in, we ask whether a step (equation in point 4) in X and Y will result in the particle being outside the box and if that is the case, we reverse velocity that potentially brought the particle outside. We could use the following code to do that

Inserting the above two snippets of code, under a loop over the number of steps we want to take, we finally end up with the code that simulates non-interacting particles in a container.

Try and code it for yourself or use it in a course you teach. The students do find it interesting, both in the form of "what do I need to tell the computer to make it do stuff?" and "oh - the particles do move!".

Disclaimer: This code is part of a course I help teach at the University of Copenhagen - just imagine you never saw python before.