Calling gs.Gaussian() with dim=3 leads to TypeError: 'float' object cannot be interpreted as an integer

[MatusGasparik]

This works:

model = gs.Gaussian(dim=2, var=1)
srf = gs.SRF(model, seed=20170519)

But this doesn't:

model = gs.Gaussian(dim=3, var=1)
srf = gs.SRF(model, seed=20170519)

Traceback:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-76-0274372fd2dd> in <module>
      3 
      4 model = gs.Gaussian(dim=3, var=1) #, len_scale=10)
----> 5 srf = gs.SRF(model, seed=20170519)
      6 
      7 #x = y = z = np.arange(0, 10)

~/miniconda3/envs/indust39/lib/python3.9/site-packages/gstools/field/srf.py in __init__(self, model, mean, normalizer, trend, upscaling, generator, **generator_kwargs)
     98         # initialize attributes
     99         self.upscaling = upscaling
--> 100         self.set_generator(generator, **generator_kwargs)
    101 
    102     def __call__(

~/miniconda3/envs/indust39/lib/python3.9/site-packages/gstools/field/srf.py in set_generator(self, generator, **generator_kwargs)
    159         if generator in GENERATOR:
    160             gen = GENERATOR[generator]
--> 161             self._generator = gen(self.model, **generator_kwargs)
    162             self.value_type = self._generator.value_type
    163         else:

~/miniconda3/envs/indust39/lib/python3.9/site-packages/gstools/field/generator.py in __init__(self, model, mode_no, seed, verbose, sampling, **kwargs)
    106         self.sampling = sampling
    107         # set model and seed
--> 108         self.update(model, seed)
    109 
    110     def __call__(self, pos, add_nugget=True):

~/miniconda3/envs/indust39/lib/python3.9/site-packages/gstools/field/generator.py in update(self, model, seed)
    172                 self._model = dcp(model)
    173                 if seed is None or not np.isnan(seed):
--> 174                     self.reset_seed(seed)
    175                 else:
    176                     self.reset_seed(self._seed)

~/miniconda3/envs/indust39/lib/python3.9/site-packages/gstools/field/generator.py in reset_seed(self, seed)
    240             )
    241         else:
--> 242             rad = self._rng.sample_ln_pdf(
    243                 ln_pdf=self.model.ln_spectral_rad_pdf,
    244                 size=self._mode_no,

~/miniconda3/envs/indust39/lib/python3.9/site-packages/gstools/random/rng.py in sample_ln_pdf(self, ln_pdf, size, sample_around, nwalkers, burn_in, oversampling_factor)
     96         initial_state = State(burn_in_state, copy=True)
     97         initial_state.random_state = self.random.get_state()
---> 98         sampler.run_mcmc(initial_state=initial_state, nsteps=sample_size)
     99         samples = sampler.get_chain(flat=True)[:, 0]
    100 

~/miniconda3/envs/indust39/lib/python3.9/site-packages/emcee/ensemble.py in run_mcmc(self, initial_state, nsteps, **kwargs)
    436 
    437         results = None
--> 438         for results in self.sample(initial_state, iterations=nsteps, **kwargs):
    439             pass
    440 

~/miniconda3/envs/indust39/lib/python3.9/site-packages/emcee/ensemble.py in sample(self, initial_state, log_prob0, rstate0, blobs0, iterations, tune, skip_initial_state_check, thin_by, thin, store, progress)
    374             checkpoint_step = thin_by
    375             if store:
--> 376                 self.backend.grow(iterations, state.blobs)
    377 
    378         # Set up a wrapper around the relevant model functions

~/miniconda3/envs/indust39/lib/python3.9/site-packages/emcee/backends/backend.py in grow(self, ngrow, blobs)
    173         self._check_blobs(blobs)
    174         i = ngrow - (len(self.chain) - self.iteration)
--> 175         a = np.empty((i, self.nwalkers, self.ndim), dtype=self.dtype)
    176         self.chain = np.concatenate((self.chain, a), axis=0)
    177         a = np.empty((i, self.nwalkers), dtype=self.dtype)

TypeError: 'float' object cannot be interpreted as an integer

The problem seems to originate in gs.random.rng.RNG.sample_ln_pdf():

        #.....
        if size is None:  # pragma: no cover
            sample_size = burn_in
        else:
            sample_size = max(burn_in, (size / nwalkers) * oversampling_factor)

...which later calls sampler.run_mcmc() with nsteps=sample_size but with sample_size=200.0 being a float:

sampler.run_mcmc(initial_state=initial_state, nsteps=sample_size)

[MuellerSeb]

Hey there!

Good catch. This should have been converted to integer in the past already. I guess the problem never ocured, since it was not a problem with older versions of emcee, that got a new release last week: https://github.com/dfm/emcee/releases/tag/v3.1.0

I will come up with a bugfix. Sorry for the inconvenience and thank you very much for the investigation of this error.

For now you could downgrade your emcee version to 3.0.2.

Cheers,
Sebastian

[MuellerSeb]

This is now fixed in the latest release 1.3.2