def fit(self):
t0 = u.t()
self.fitted = False
np.random.seed(self.random_state)
if self.n_jobs_ != 1:
pool = multiprocessing.Pool(self.n_jobs_)
try:
iter_params = []
for it in self.yield_cv():
iter_params.append(it)
res = pool.map(_fit, iter_params)
for iCV, i_model, i_preprocessor in res:
self.models[iCV] = i_model
self.preprocessors[iCV] = i_preprocessor
finally:
pool.close()
else:
for args in self.yield_cv():
icv, i_model, i_preprocessor = KFoldPredBase.fit_static(args)
self.models[icv] = i_model
self.preprocessors[icv] = i_preprocessor
self.fitted = True
if self.verbose:
logd("Fit ALL CVs in %2.2f seconds" % u.td(t0))
return self
def fit(self, external_cols=None):
self.predictors = {}
self.predictions_cv = pd.DataFrame()
for i_model, model in enumerate(self.models):
model_name = stringify2(model, i_model)
if self.verbose:
logd(model_name)
t0 = u.t()
i_predictor = self.base_predictor(self.X, self.y, model, scoring=self.scoring,
n_folds=self.n_folds, random_state=self.random_state,
shuffle=self.shuffle, n_jobs=self.predictors_n_jobs,
preprocessor=self.preprocessor, verbose=self.verbose)
col = model_name
i_predictor.fit()
i_prediction_cv = i_predictor.predict()
if not len(self.predictions_cv):
self.predictions_cv = i_prediction_cv.rename(columns={i_predictor.cv_col: col}) # [i_predictor.cv_col]
else:
df = i_prediction_cv[[i_predictor.cv_col]].rename(columns={i_predictor.cv_col: col})
# TODO assert index is not duplicate
self.predictions_cv = self.predictions_cv.merge(df, left_index=True, right_index=True)
i_predictor.fit_test()
self.predictors[model_name] = i_predictor
if self.verbose:
logd("Fit %s in %2.2f seconds" % (model_name, u.td(t0)))
self.fit_ensemble(external_cols=external_cols)
def fit_test(self):
t0 = u.t()
self.fitted_test = False
x = self.preprocessor.fit_transform(self.X) if self.preprocessor else self.X
self.model.fit(x, self.y)
self.fitted_test = True
if self.verbose:
logd("Fit Test in %2.2f seconds | %s" % (u.td(t0), x.shape))
def fit_ensemble(self, external_cols=None):
t0 = u.t()
_x = self.predictions_cv[self.cols] if self.predictions_cv is not None else pd.DataFrame()
if external_cols is not None:
if not isinstance(external_cols, pd.DataFrame):
external_cols = pd.DataFrame(external_cols)
for col in external_cols.columns:
_x["ADD_%s" % col] = external_cols[col]
_y = self.predictions_cv[self.true_col]
self.ensemble_scaler = StandardScaler()
x = self.ensemble_scaler.fit_transform(_x)
if self.ensemble_grid_params:
scorer = make_scorer(self.scoring, greater_is_better=self.score_greater_is_better)
self.ensemble_model, _ = \
u.get_best_model(self.ensemble_model, self.ensemble_grid_params, x, _y,
scoring=scorer, cv=self.n_folds, refit=True)
else:
self.ensemble_model.fit(x, _y)
if self.verbose:
logd("Fit Ensemble in %2.2f seconds" % u.td(t0))
self.predictions_cv["ENS"] = self.ensemble_model.predict(x)
self.predictions_cv = self.predictions_cv[self.cols + ["ENS", self.true_col]]
