Using PyCall to call iminuit

New to Julia
1

Hello, I’m working in high energy physics, and there is a package called MINUIT, https://seal.web.cern.ch/seal/snapshot/work-packages/mathlibs/minuit/, that is commonly used in the field for Function Minimization and Error Analysis. It is in cernlib, and both fortran and c++ versions. Wrapping either of them in Julia does not exist yet. But there is a python wrapping of the c++ version, called iminuit, https://iminuit.readthedocs.io/en/latest/. It can be installed by conda install iminuit.

I then tried PyCall to use iminuit inside Julia.

minuit = pyimport(:iminuit)

datax = LinRange(0,10, 10)
σ = 1; μ = 0;
off = σ .*randn(10) .+ μ
line(x, a, b) = a + sin(x) * b
datay = line.(datax, 1, 2) .+ 0.2*off

function χsq(a, b)
    err = 0.01
    sum(@. (datay- line(datax, a, b))^2/err )
end

I then tried to get the best fit to the data datay by minimizing χsq with the following line

minuit.Minuit(χsq, a=5, b=5, error_a = 0.1, error_b = 0.1, errordef = 1)

Then I got an error starting with

PyError ($(Expr(:escape, :(ccall(#= /homepath/.julia/packages/PyCall/ttONZ/src/pyfncall.jl:44 =# @pysym(:PyObject_Call), PyPtr, (PyPtr, PyPtr, PyPtr), o, pyargsptr, kw))))) <class 'TypeError'>
TypeError('Unable to obtain function signature')
  File "iminuit/_libiminuit.pyx", line 475, in iminuit._libiminuit.Minuit.__init__
File "/homepath/anaconda3/lib/python3.7/site-packages/iminuit/util.py", line 350, in describe
    return better_arg_spec(f, verbose)
  File "/homepath/anaconda3/lib/python3.7/site-packages/iminuit/util.py", line 340, in better_arg_spec
    raise TypeError("Unable to obtain function signature")

Any help will be very grateful.
If there is some day a Julia wrapping of the MINUIT package, there would be great!

Thank you in advance

2

Not too be dismissive, but Julia already has some state of the art told for these problems. Why not use those?

3

MINUIT has lots of subroutines that are very useful and easy-to-use for the data analysis at least in high energy physics. Just copy one sentence from the manual:

The Minuit processor MINOS was probably the first, and may still be the only, generally available
program to calculate parameter errors taking into account both parameter correlations and non-linearities.
The MINOS error intervals are in general assymmetric, and may be expensive to calculate, especially if there are a lot of free parameters and the problem is very non-linear.

If this lib can be ported to julia, I’m sure there will be many more julia users from this community.

1 Like
4

This is not a Python function · Issue #268 · JuliaPy/PyCall.jl · GitHub

The workaround is to wrap the Julia function in a Python lambda:

pyfun = py"lambda fun: lambda x, y: fun(x,y)"

minuit.Minuit(pyfun(χsq), a=5, b=5, error_a = 0.1, error_b = 0.1, errordef = 1)
1 Like
5

Thanks a lot! I tested, to make it work, I need to use the same letters for the parameters as those appear in the lambda definition, i.e.,

pyfun = py"lambda fun: lambda a, b: fun(a, b)"

minuit.Minuit(pyfun(χsq), a=5, b=5, error_a = 0.1, error_b = 0.1, errordef = 1)

Then a, b and error_a, error_b can be correctly recognized.