search

scikit-hep / iminuit

Jupyter-friendly Python interface for C++ MINUIT2
https://scikit-hep.org/iminuit
Other
283 stars 76 forks source link

Invitation #1

Closed piti118 closed 11 years ago

piti118 commented 11 years ago

@cdeil @mattbellis You might be interested in this package. No root required at all.

cdeil commented 11 years ago

This is great! I'll be using this all the time for interactive fitting.

piti118 commented 11 years ago

How did I miss @gitcheng.

mattbellis commented 11 years ago

Looks awesome! Is there any benefits/major differences with pyminuit that I should be aware of? Is this better than RTMinuit?

piti118 commented 11 years ago

just see the tutorial

mattbellis commented 11 years ago

Hey Piti, looking over the tutorial and it looks great. I've been using pyminuit for a while now and switching over to iminuit was pretty simple. I had written my own pretty-printing functions, but now I've got iminuit, I don't need them!

I've a question about your cython tutorial. I just started checking out cython earlier this week. I have a fairly involved PDF that I've coded up using numpy, so it's much faster than having the loops in python. In your cython tutorials, I see all the loops are hard-coded. Does that give you much improvement over numpy? Should I take the time to type def my .pyx file? Or does it not matter that much if it's numpy?

piti118 commented 11 years ago

Haha. Good to know you like it. Depending on what kind of things you did in numpy if it involves only numpy broadcast function (np.exp etc.) and no python function call then you won't see any significant gain. It might even be slower if you change your code to the one in tutorial. But the tutorial definitely use less ram.

The main cost of computing likelihood is actually python function call and making tuple to make a call. Eliminate them as much as you can.

I wrote a pretty primitive html frontend. You could fork it and make it prettier and push it back. I can give you admin access too if you like.

mattbellis commented 11 years ago

OK, sounds about what I was thinking. I made sure to only use the np.sqrt, np.exp, np.log, which was a huge gain over the straight python stuff. I'll stick with the PDF for now and just focus on the iminuit interface. Thanks!