sample_crp_tablecounts in cstats fails in certain cases

[jonahpearl]

Trying to model some data with a negative binomial duration distribution...the /util/cstats/sample_crp_tablecounts function was failing because its second argument integral[:,:] customers was being passed in as a float. The numbers being passed in appear to be durations in n_samples, so they're guaranteed to be integers, so I fixed by changing line 31 in pyhsmm/distributions (within the _StartAtOneMixin.resample() method) from return super(_StartAtOneMixin,self).resample(data-1, *args,**kwargs) to return super(_StartAtOneMixin,self).resample((data-1).astype('int'), *args,**kwargs).

If I have time I'll try to track down what caused the error in the first place.

*** Code to reproduce ***

data = np.arange(10).reshape(-1,1)
Nmax = 25

# and some hyperparameters
obs_dim = data.shape[1]
obs_hypparams = {'mu_0':np.zeros(obs_dim),
                'sigma_0':np.eye(obs_dim),
                'kappa_0':0.25,
                'nu_0':obs_dim+2}
obs_distns = [pyhsmm.distributions.Gaussian(**obs_hypparams) for state in range(Nmax)]

dur_hypparams = {'r':3., 'p':0.5, 'k_0':2., 'theta_0':2., 'alpha_0':2., 'beta_0':2.}
dur_distns = [pyhsmm.distributions.NegativeBinomialDuration(**dur_hypparams) for state in range(Nmax)]

posteriormodel = pyhsmm.models.WeakLimitHDPHSMM(
        alpha=6.,gamma=6., # these can matter; see concentration-resampling.py
        init_state_concentration=6., # pretty inconsequential
        obs_distns=obs_distns,
        dur_distns=dur_distns)

posteriormodel.add_data(data, trunc=60) # duration truncation speeds things up when it's possible

for idx in progprint_xrange(150):
    posteriormodel.resample_model()

*** Traceback from error ***

TypeError Traceback (most recent call last) /tmp/ipykernel_16301/156295421.py in 35 36 for idx in progprint_xrange(150): ---> 37 posteriormodel.resample_model()

~/datta-lab/pyhsmm/pyhsmm/models.py in resample_model(self, num_procs) 454 @line_profiled 455 def resample_model(self,num_procs=0): --> 456 self.resample_parameters() 457 self.resample_states(num_procs=num_procs) 458

~/datta-lab/pyhsmm/pyhsmm/models.py in resample_parameters(self, **kwargs) 964 @line_profiled 965 def resample_parameters(self,**kwargs): --> 966 self.resample_dur_distns() 967 super(_HSMMGibbsSampling,self).resample_parameters(**kwargs) 968

~/datta-lab/pyhsmm/pyhsmm/models.py in resample_dur_distns(self) 969 def resample_dur_distns(self): 970 for state, distn in enumerate(self.dur_distns): --> 971 distn.resample_with_censoring_and_truncation( 972 data= 973 [s.durations_censored[s.untrunc_slice][s.stateseq_norep[s.untrunc_slice] == state]

~/datta-lab/pyhsmm/pyhsmm/basic/abstractions.py in resample_with_censoring_and_truncation(self, data, censored_data, left_truncation_level) 101 rejected_observations = [] 102 --> 103 self.resample(data=combinedata((data,filled_in,rejected_observations))) 104 105 @property

~/datta-lab/pyhsmm/pyhsmm/basic/distributions.py in resample(self, data, *args, **kwargs) 33 else: 34 # return super(_StartAtOneMixin,self).resample([(d-1).astype('int') for d in data],*args,**kwargs) ---> 35 return super(_StartAtOneMixin,self).resample([d-1 for d in data],*args,**kwargs) 36 37 def max_likelihood(self,data,weights=None,*args,**kwargs):

~/miniconda3/envs/dataPy_NWB/lib/python3.9/site-packages/pybasicbayes/distributions/negativebinomial.py in resample(self, data, niter) 113 for itr in range(niter): 114 ### resample r --> 115 msum = sample_crp_tablecounts(self.r,data).sum() 116 self.r = np.random.gamma(self.k_0 + msum, 1/(1/self.theta_0 - N*np.log(1-self.p))) 117 ### resample p

pybasicbayes/util/cstats.pyx in pybasicbayes.util.cstats.__pyx_fused_cpdef()

TypeError: Function call with ambiguous argument types