Difference between pymc2 and pymc3 in pm.Deterministic?

Question

I have model in pymc2 like this.

p = Uniform(name='p', lower=p_lowers, upper=p_uppers, value=p_values)
MIM_Model =  pm.Deterministic(eval = MIM_eval,
                         name = 'NEE_Model',
                         parents = {'p': p, 'Csom_0': Csom_0, 
                                    'Tave': Tave, 'Tmax': Tmax, 'Rg': Rg, 'RH': RH, 'VPD':VPD,'Lat':Lat,'a2': a2},
                         doc = 'model construction',
                         trace = True,
                         verbose = 0,
                         dtype = float,
                         plot = False,
                         cache_depth = 2)

All of the variables (Rg': Rg, 'RH': RH, 'VPD':VPD,'Lat':Lat) are arrays.

def MIM_eval(p, Csom_0, Tave, Tmax, Rg, RH, VPD, Lat, a2):
  p1 = p[0]#*1.e5
  p2 = p[1]#*1.e5
  p3 = p[2]#*1.e5
  p4 = p[3]#*1.e5
  p5 = p[4]#*1.e5
....
  Lf[0][i] = p6*Cf_0
  Ln[0][i] = p8*Cn_0
  Lr[0][i] = p7*Cr_0
....
  _results = []
  for j in range(obs_size):
    tmp_result = []
    for k in idx[j]:
      tmp_result.append(round(all_results[j][0][k], 8)) 
    _results.append(tmp_result)
  # Return the model result.
  return _results

And this runs well in pymc2, but i can't use it in pymc3, cause params in Deterministic must be theano variable, i searched a lot of examples but no one fits this situation(cause too many params);

• Example online, the theanovar is too simple...

 Normal('v3', mu=mean, sigma=half_cauchy)
 # Deterministic variables can be used in usual way
 Deterministic('v3_sq', self.v3 ** 2)

In my situation, i tried below, but i don't know how to write MIM_eval in pymc3, cause the deterministic's params can't recieve type of array.

MIM_Model =  pm.Deterministic('MIM_Model',MIM_eval(p, Cf_0, Cr_0, Cw_0, Clit_0, Csom_0, Nt, Ta))
def MIM_eval(p, Csom_0, Tave, Tmax, Rg, RH, VPD, Lat, a2):
   # ???
   # can't return results