def _predict_pset(self, pset, theta):
i,j = pset
X = self.X.loc[:,[j]+i].values
Y = self.Y.loc[:,[j]].values
return gp.rbf(X,Y,theta).predict(X)
def eval_gp(self, i, theta):
"Evaluate a GP for parental set @i using a RBF"
# extract the items for our given parental set
pi,gi = self.pset[i]
# pull out data for these items
X = self.X.loc[:,[gi]+pi].values
Y = self.Y.loc[:,[gi]].values
# evaluate the GP
return gp.rbf(X, Y, theta)
def __init__(self, res, pset, datasets=None):
self.target = res.target
self.hypers = res.hypers
self.pset = list(pset)
self.ix = [self.target]+self.pset
self.iy = [self.target]
if datasets is None:
datasets = (r.data for r in res.mod.reps)
X = [] # parent array
Y = [] # target array
for d in datasets:
X.append(d[self.ix,:-1])
Y.append(d[self.iy,1:])
X = np.hstack(X).T
Y = np.hstack(Y).T
self.gp = gp.rbf(X,Y,self.hypers)
