problem with gta.optimize (Value error: function value is NaN)
Closed this issue · 1 comments
lylol commented
Dear all,
I am trying to run a fermipy analysis following: https://fermipy.readthedocs.io/en/latest/quickstart.html.
The following gta.setup function operates normally,
from fermipy.gtanalysis import GTAnalysis
gta = GTAnalysis('config.yaml',logging={'verbosity' : 3})
gta.setup()however when I run
gta.optimize()it gives the following error (I use the 4FGL-DR4 catalogs from https://fermi.gsfc.nasa.gov/ssc/data/access/lat/14yr_catalog/):
error report
2023-09-21 10:20:13 INFO GTAnalysis.optimize(): Starting
Joint fit ['galdiff', 'isodiff', '4FGL J1224.9+2122', '4FGL J1239.5+0443', '4FGL J1222.5+0414']
/Users/yonlinli/mambaforge/envs/fermipy/lib/python3.9/site-packages/scipy/interpolate/_fitpack2.py:313: UserWarning:
The maximal number of iterations maxit (set to 20 by the program)
allowed for finding a smoothing spline with fp=s has been reached: s
too small.
There is an approximation returned but the corresponding weighted sum
of squared residuals does not satisfy the condition abs(fp-s)/s < tol.
warnings.warn(message)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/Users/yonlinli/mambaforge/envs/fermipy/lib/python3.9/site-packages/fermipy/gtanalysis.py", line 2339, in optimize
self.fit(loglevel=logging.DEBUG, **config['optimizer'])
File "/Users/yonlinli/mambaforge/envs/fermipy/lib/python3.9/site-packages/fermipy/gtanalysis.py", line 3074, in fit
self.update_source(name, reoptimize=config['reoptimize'])
File "/Users/yonlinli/mambaforge/envs/fermipy/lib/python3.9/site-packages/fermipy/gtanalysis.py", line 4106, in update_source
sd = self.get_src_model(name, paramsonly, reoptimize, npts,
File "/Users/yonlinli/mambaforge/envs/fermipy/lib/python3.9/site-packages/fermipy/gtanalysis.py", line 4264, in get_src_model
lnlp = self.profile_norm(name, savestate=True,
File "/Users/yonlinli/mambaforge/envs/fermipy/lib/python3.9/site-packages/fermipy/gtanalysis.py", line 2439, in profile_norm
lims = utils.get_parameter_limits(lnlp['xvals'],
File "/Users/yonlinli/mambaforge/envs/fermipy/lib/python3.9/site-packages/fermipy/utils.py", line 804, in get_parameter_limits
x0 = find_function_root(sd, xval[ilo], xval[ihi])
File "/Users/yonlinli/mambaforge/envs/fermipy/lib/python3.9/site-packages/fermipy/utils.py", line 700, in find_function_root
return brentq(lambda t: fn(t) + delta, x0, xb, xtol=xtol)
File "/Users/yonlinli/mambaforge/envs/fermipy/lib/python3.9/site-packages/scipy/optimize/_zeros_py.py", line 799, in brentq
r = _zeros._brentq(f, a, b, xtol, rtol, maxiter, args, full_output, disp)
File "/Users/yonlinli/mambaforge/envs/fermipy/lib/python3.9/site-packages/scipy/optimize/_zeros_py.py", line 98, in f_raise
raise err
ValueError: The function value at x=0.0 is NaN; solver cannot continue.
Corresponding reports from fermipy.log :
2023-09-21 10:20:13 INFO GTAnalysis.optimize(): Starting
2023-09-21 10:20:13 DEBUG GTAnalysis.optimize(): LogLike: -129050.599035
2023-09-21 10:20:14 DEBUG GTAnalysis.free_source(): Freeing parameters for galdiff : ['Prefactor']
2023-09-21 10:20:14 DEBUG GTAnalysis.free_source(): Freeing parameters for isodiff : ['Normalization']
2023-09-21 10:20:14 DEBUG GTAnalysis.free_source(): Freeing parameters for 4FGL J1224.9+2122 : ['norm']
2023-09-21 10:20:14 DEBUG GTAnalysis.free_source(): Freeing parameters for 4FGL J1239.5+0443 : ['norm']
2023-09-21 10:20:14 DEBUG GTAnalysis.free_source(): Freeing parameters for 4FGL J1222.5+0414 : ['norm']
2023-09-21 10:20:14 DEBUG GTAnalysis.fit(): Starting fit.
2023-09-21 10:20:15 DEBUG GTAnalysis.fit(): Fit complete.
2023-09-21 10:20:15 DEBUG GTAnalysis.get_src_model(): Generating source dict for 4FGL J1222.5+0414
2023-09-21 10:20:15 DEBUG GTAnalysis.profile_norm(): Profiling 4FGL J1222.5+0414
2023-09-21 10:20:16 DEBUG GTAnalysis.profile_norm(): Finished
2023-09-21 10:20:16 DEBUG GTAnalysis.get_src_model(): Generating source dict for 4FGL J1224.9+2122
2023-09-21 10:20:16 DEBUG GTAnalysis.profile_norm(): Profiling 4FGL J1224.9+2122
2023-09-21 10:20:16 DEBUG GTAnalysis.profile_norm(): Finished
2023-09-21 10:20:16 DEBUG GTAnalysis.get_src_model(): Generating source dict for 4FGL J1239.5+0443
2023-09-21 10:20:16 DEBUG GTAnalysis.profile_norm(): Profiling 4FGL J1239.5+0443
No more outputs, so I suppose there are some problems with the profiling procedure.
Environment :
- MacOS
- Darwin Kernel Version 22.4.0: Mon Mar 6 21:00:17 PST 2023; root:xnu-8796.101.5~3/RELEASE_X86_64
- fermipy 1.2.0, Sciencetools 2.2.0
Besides I find that when I decrease the 'src_roiwidth' parameter (in the 'model' section of the configuration file) from 20.0 to 5.0 , sometimes the optimize function reports no error (sometimes it reports the same NaN error described above, confusingly):
2023-09-21 15:47:17 INFO GTAnalysis.optimize(): Starting
Joint fit ['galdiff', 'isodiff']
/Users/yonlinli/mambaforge/envs/fermipy/lib/python3.9/site-packages/scipy/interpolate/_fitpack2.py:313: UserWarning:
The maximal number of iterations maxit (set to 20 by the program)
allowed for finding a smoothing spline with fp=s has been reached: s
too small.
There is an approximation returned but the corresponding weighted sum
of squared residuals does not satisfy the condition abs(fp-s)/s < tol.
warnings.warn(message)
Fitting shape galdiff TS: 824.873
Fitting shape isodiff TS: 694.017
Fitting shape 4FGL J1230.8+1223 TS: 59.605
2023-09-21 15:47:19 INFO GTAnalysis.optimize(): Finished
2023-09-21 15:47:19 INFO GTAnalysis.optimize(): LogLike: -129566.256655 Delta-LogLike: 1115.703656
2023-09-21 15:47:19 INFO GTAnalysis.optimize(): Execution time: 1.69 s
{'loglike0': -130681.96031147361, 'loglike1': -129566.25665539311, 'dloglike': 1115.703656080499, 'config': {'npred_threshold': 1.0, 'npred_frac': 0.95, 'shape_ts_threshold': 25.0, 'max_free_sources': 5, 'skip': [], 'optimizer': {'optimizer': 'MINUIT', 'tol': 0.001, 'max_iter': 100, 'init_lambda': 0.0001, 'retries': 3, 'min_fit_quality': 2, 'verbosity': 0}}}
Then I run
gta.print_roi()the outputs are:
2023-09-21 15:47:23 INFO GTAnalysis.print_roi():
name SpatialModel SpectrumType offset ts npred
--------------------------------------------------------------------------------
4FGL J1230.8+1223 PointSource LogParabola 0.047 87.80 71.9
4FGL J1233.0+1333 PointSource LogParabola 1.250 0.41 12.6
4FGL J1235.9+1136 PointSource LogParabola 1.443 1.62 8.3
4FGL J1223.3+1213 PointSource PowerLaw 1.880 3.39 77.3
4FGL J1231.5+1421 PointSource PowerLaw 1.951 3.44 15.1
4FGL J1223.0+1100 PointSource LogParabola 2.393 8.95 7.8
isodiff ConstantValue FileFunction ----- 15837.94 22086.4
galdiff MapCubeFunctio PowerLaw ----- 21913.07 25110.9
But I really need to perform the analysis in a wider range ('src_roiwidth'=20 deg).
I sincerely appreciate every help on this problem.
lylol commented
It seems like a matter of memory. I successfully run the same program on a different device .