Errors with sampling

1

I’m using v5.I’m using normal likelihood, but I got an error and I don’t know why.My codes are

from pytensor.tensor import conv
def Ft(A,sigma2,mu2,a):
    t_min=0
    t_max=(t4_s.shape[0])*(t4_s[1]-t4_s[0])
    num=t4_s.shape[0]    
    t=np.linspace(t_min,t_max,num)
    tau=1/sigma2**2
    ft1=A*pm.math.sqrt(tau/(2*np.pi))*pm.math.exp(-tau/2*(t+a-mu2)**2)    
    IRF1=IRF[None,None,:]
    ft2=ft1[None,None,:]
    ft=(conv.causal_conv1d(ft2, IRF1, filter_shape=(1,1,IRF1.shape[2]))).squeeze()
    print(type(ft1))
    print(ft1.type)
    print(type(ft1[None, None, :]))
    print(ft1[None, None, :].type)
    print(type(ft))
    print(ft.type)
    print(ft)
    return ft
#%%
with pm.Model() as final_model:
    amp = pm.Uniform('amp',lower=-1.0,upper=0)
    mu1 = pm.Uniform('mu1', lower=-50,upper=50)
    sigma1 = pm.Uniform('sigma1',lower=0,upper=50)
    a1=pm.Uniform('a1',lower=-10,upper=10)    
    y_observed=pm.Normal(
        "y_observed",
        mu= Ft(amp,sigma1,mu1,a1),
        sigma=noise_sig,
        observed=v4_s,
        )
    output = pm.Deterministic('output',Ft(amp,sigma1,mu1,a1))
    prior = pm.sample_prior_predictive()
    posterior_f = pm.sample(draws =500, target_accept = 0.9,chains=4,cores=4)
    posterior_f = pm.sample_posterior_predictive(posterior_f)
    az.plot_trace(posterior_f, var_names = ['amp','mu1','sigma1','a1'])
    only_background=az.summary(posterior_f, var_names = ['amp','mu1','sigma1','a1'])

I got error

ValueError: Can not extract posterior from Inference data with groups:
	> posterior_predictive
	> observed_data! See /Users/majunfei/anaconda3/lib/python3.11/site-packages/arviz/data/converters.py for other conversion utilities.

I think this may caused by passing parameters ,so I print every parameters in the function which uses for cauculate mu.I got:

print(type(ft1))
  >   <class 'pytensor.tensor.var.TensorVariable'>
print(ft1.type)
  >  Vector(float64, shape=(5000,))
print(type(ft1[None, None, :]))
  >  <class 'pytensor.tensor.var.TensorVariable'>
 print(ft1[None, None, :].type)
  >  Tensor3(float64, shape=(1, 1, 5000))
print(type(ft))
  > <class 'pytensor.tensor.var.TensorVariable'>
print(ft.type)
  > Vector(float64, shape=(?,))
print(ft)
  > DropDims{axes=[0, 1]}.0
2

You overwrote your posterior when you called

posterior_f = pm.sample_posterior_predictive(posterior_f)

You can do

posterior_f = pm.sample_posterior_predictive(posterior_f, extend_inferencedata=True)

3

Thank you! I have another problem,I’m wondering how PyMC pass parameters to a called function . In my problem, when I creates t,I I hope to create a array with a central value of mu.But when I pass mu to numpy, I get errors, so I create t use

    t_min=0
    t_max=(t4_s.shape[0])*(t4_s[1]-t4_s[0])
    num=t4_s.shape[0]    
    t=np.linspace(t_min,t_max,num)

instead.But I think this may cause problems in my data processing process.

4

You should use pytensor.tensor operations and not numpy. There is a pytensor.tensor.linspace (and almost any numpy function you may need)