def fit(self, x, y=None):
x = np.asfortranarray(x)
_, self.n_features = x.shape
# 1) Get all simple features
simple = (SimpleFeature(e, index=i) for e, i in product(self.exponents, range(self.n_features)))
simple = get_valid((s, s.transform(x)) for s in simple)
# 2) Get all operator features
operator = (OperatorFeature(s, op, operator_name=op_name) for (s, _), (op_name, op) in product(simple, self.operators.items()))
operator = get_valid((o, o.transform(x)) for o in operator)
# 3) Get all product features
all_ = simple + operator
if self.consider_products:
combs = chain(product(operator, simple), combinations(simple, 2))
prod = [ProductFeature(feat1, feat2) for (feat1, _) , (feat2, _) in combs]
all_ += get_valid((p, p.transform(x)) for p in prod)
all_ = get_valid(all_)
feat_cls, features = zip(*[(c, np.array(f)) for c, f in all_])
self._precomputed_features = np.array(list(features)).T # speed up fit_transform
self._precompute_hash = _hash(x)
self.feat_cls = list(feat_cls)
return self
def inner(x):
key = _hash(x)
if key not in cache:
cache[key] = _hash(x)
return cache[key]
def transform(self, x):
if self._precompute_hash == _hash(x):
return self._precomputed_features
else:
features = [c.transform(x) for c in self.feat_cls]
return np.array(list(features)).T