def fit(self, X, y, sample_weight=None):
import eval
preds = [[] for _ in self.sub_dags]
for train_idx, test_idx in cross_validation.StratifiedKFold(y,
n_folds=5):
tr_X, tr_y = X.iloc[train_idx], y.iloc[train_idx]
tst_X, tst_y = X.iloc[test_idx], y.iloc[test_idx]
wf_init = Workflow(self.initial_dag)
wf_init.fit(tr_X, tr_y, sample_weight=sample_weight)
preproc_X, preproc_y = eval.test_dag(self.initial_dag,
wf_init.models,
test_data=(tr_X, tr_y),
output='all')
pp_tst_X = wf_init.transform(tst_X)
if pp_tst_X.empty:
continue
for i, dag in enumerate(self.sub_dags):
wf = Workflow(dag)
wf.fit(preproc_X, preproc_y)
res = wf.predict(pp_tst_X)
preds[i].append(pd.DataFrame(res, index=pp_tst_X.index))
preds = [pd.concat(ps) for ps in preds]
self.train = pd.concat(preds, axis=1)
self.train.columns = ['p' + str(x) for x in range(len(preds))]
return self
def fit(self, X, y, sample_weight=None):
import eval
preds = [[] for _ in self.sub_dags]
for train_idx, test_idx in cross_validation.StratifiedKFold(y, n_folds=5):
tr_X, tr_y = X.iloc[train_idx], y.iloc[train_idx]
tst_X, tst_y = X.iloc[test_idx], y.iloc[test_idx]
wf_init = Workflow(self.initial_dag)
wf_init.fit(tr_X, tr_y, sample_weight=sample_weight)
preproc_X, preproc_y = eval.test_dag(self.initial_dag, wf_init.models, test_data=(tr_X, tr_y), output='all')
pp_tst_X = wf_init.transform(tst_X)
if pp_tst_X.empty:
continue
for i, dag in enumerate(self.sub_dags):
wf = Workflow(dag)
wf.fit(preproc_X, preproc_y)
res = wf.predict(pp_tst_X)
preds[i].append(pd.DataFrame(res, index=pp_tst_X.index))
preds = [pd.concat(ps) for ps in preds]
self.train = pd.concat(preds, axis=1)
self.train.columns = ['p' + str(x) for x in range(len(preds))]
return self
def transform(self, X):
import eval
return eval.test_dag(self.dag,
self.models,
test_data=(X, None),
output='feats_only')
def predict(self, X):
import eval #TODO: Refactor to remove circular imports
return np.array(
eval.test_dag(self.dag, self.models, test_data=(X, None)))
def transform(self, X):
import eval
return eval.test_dag(self.dag, self.models, test_data=(X, None), output='feats_only')
def predict(self, X):
import eval #TODO: Refactor to remove circular imports
return np.array(eval.test_dag(self.dag, self.models, test_data=(X, None)))