How to write & evaluate state space model in PyMC4

PyMC4
1

I’m tring to write down state space model in PyMC4 and sample posterior distributions of parameters as in PyMC3.
But the acceptance ratio of all variables are 0 and chain values do not change.
Is there some mistake in this code?


data=pd.read_csv("https://raw.githubusercontent.com/statsmodels/statsmodels/master/statsmodels/datasets/nile/nile.csv",index_col='year',)
N=len(data)
datay=np.array(data["volume"])

model = pm.Model(N=N, y=datay)

@model.define
def st(cfg):    
    mu0 = ed.Normal(name='muZero', loc=0.0, scale=1.0)
    sigmaW = ed.InverseGamma(name='sigmaW', concentration=1.0, rate=1.0)    
    sigmaV = ed.InverseGamma(name='sigmaV', concentration=1.0, rate=1.0)
    
    y=[0]*cfg.N
    mu =[0]*cfg.N
    mu[0] = ed.Normal(name='mu0', loc=mu０, scale=sigmaW)
    for n in range(1, N):
        mu[n] = ed.Normal(name='mu'+str(n), loc=mu[n-1], scale=sigmaW)    
    for n in range(N):
        y[n]= ed.Normal(name='y'+str(n), loc=mu[n], scale=sigmaV)
    
    return y

trace = pm.sample(model)
2

At this point PyMC4 is more of a prototype and I would expect many bugs. If you don’t want to actively develop the package I would stick to PyMC3 which we will continue to support.