Strange shape of internal logp function .stats._log_post_trace() test case

pymc3test0.py

import numpy as np
import matplotlib.pyplot as plt
import pymc3 as pm
import theano

print("numpy version",np.__version__)
print("theano version",theano.__version__)
print("pymc version",pm.__version__)

floatX = theano.config.floatX # "float32"

#np.random.seed(1337)
## ALEPH: GENERATE DATA
mean_t = np.array((0,0),dtype=floatX)
cov_t = np.eye(2,dtype=floatX)
n = 2222
data= np.random.multivariate_normal(mean=mean_t,cov=cov_t,size=n).astype(floatX)

# look at the data
plt.plot(data[:,0],data[:,1],'o')
plt.gca().set_aspect('equal')

## BETH: FORMULATE THE MODEL
mvg_model = pm.Model()
with mvg_model:
    mu = pm.Uniform('mu',lower=-1,upper=1,shape=2)
    y = pm.MvNormal('y',mu=mu,cov=cov_t,observed=data)


with mvg_model:
    trace = pm.sample(2000,njobs = 2)

_ = pm.traceplot(trace,combined=True)
plt.show()


lnp = np.array([mvg_model.logp(trace.point(i,chain=c)) for c in trace.chains for i in range(len(trace))])
print('lnp.shape ',lnp.shape)


lnp_strange = pm.stats._log_post_trace(trace,mvg_model)
print('lnp_strange.shape ',lnp_strange.shape)