Save and reuse model with shared variables

Questions
1

Let’s say I have a simple model with shared variables to make out-of-sample predictions:

N=5000
p=2
batch_size=100
X=np.random.normal(0,1,(N,p))
b=np.random.uniform(0,0.1,p)
y=np.random.poisson(np.exp(X.dot(b)),N)

model = pm.Model()
with model:

x0_s=shared(X[:,0])
x1_s=shared(X[:,1])

b0=pm.Normal('b0',mu=0,sd=.1)
b1=pm.Normal('b1',mu=0,sd=.1)
mu=pm.math.exp(b0*x0_s+b1*x1_s)
y_=pm.Poisson('Y_obs',mu=mu,observed=y)

approx=pm.fit(10000,method='ADVI')
trace=approx.sample(1000)

I need to save the fitted model and then to make prediction in a separate session.
I can easily pickle the trace and the model but how can I manage the shared variables?
One solution would be to redefine the model in the new session so that I have access to the shared variables but I would like to avoid that.
Any suggestions?

Thank you.

2

Did you try pickling them?

Save them with the approx variable and set the values before you call approx.sample(). For me this works just fine.

x0_s.set_value(newX0)
x1_s.set_value(newX1)
trace = approx.sample(1000)
1 Like
3

You are right! The easiest solution is the one that works :slight_smile:

Thank you.