def fit(self, X, y):
if not isinstance(X, pd.DataFrame):
X = pd.DataFrame(X)
X = pd.DataFrame(self.sparsify.fit_transform(X).A)
if self.scoring is None:
tp = classification_or_regression(y)
self.scoring = [f1_weighted_scorer,
mse_scorer][tp != 'classification']
print(self.scoring)
self.pick_model(X, y)
self.get_pipeline()
unique_combinations = np.prod(list(map(len, self.grid.values())))
print("unique_combinations", unique_combinations)
if 'population_size' not in self.kwargs:
self.kwargs['population_size'] = np.clip(
int(unique_combinations / 1000), 5, 10)
if 'generations_number' not in self.kwargs:
self.kwargs['generations_number'] = np.clip(
int(unique_combinations / 20), 10, 50)
self.evo = evo_search(self.pipeline,
self.grid,
scoring=self.scoring,
cv=self.cv,
n_jobs=self.n_jobs,
**self.kwargs)
self.evo.fit(X, y)
return self.evo.best_estimator_
def fit(self, X, y):
evos = []
if not isinstance(X, pd.DataFrame):
X = pd.DataFrame(X)
self._feature_name = np.array(list(X.columns))
if isinstance(y, pd.DataFrame):
y = np.array(y)
elif hasattr(y, "__array__"):
y = y.__array__()
elif len(y.shape) > 1 and y.shape[1] == 1:
y = y.ravel()
X = pd.DataFrame(self.featurizer.fit_transform(X).A)
self.cl_or_reg = self.cl_or_reg or classification_or_regression(y)
cl_or_reg = self.cl_or_reg
if self.scoring is None:
self.scoring = [f1_weighted_scorer, mse_scorer][cl_or_reg != "classification"]
print(self.scoring)
complexity = calculate_complexity(X, y, cl_or_reg)
for model in self.get_models(X, y):
try:
print("estimator:", model["name"])
grid = model["grid"]
if complexity > model["max_complexity"]:
continue
clf = self.handle_multi_output(y, model["name"], model["clf"])
if clf is None:
continue
pipeline = self.get_pipeline(clf)
unique_combinations = np.prod(list(map(len, grid.values())))
print("unique_combinations", unique_combinations)
kwargs = self.kwargs.copy()
if "population_size" not in self.kwargs:
kwargs["population_size"] = np.clip(int(unique_combinations / 100), 5, 10)
if "generations_number" not in kwargs:
kwargs["generations_number"] = np.clip(int(unique_combinations / 10), 10, 50)
evo = evo_search(
pipeline, grid, scoring=self.scoring, cv=self.cv, n_jobs=self.n_jobs, **kwargs
)
evo.fit(X, y)
evos.append((evo.best_score_, evo))
except KeyboardInterrupt:
if not evos:
print("Stopped by user. No models finished trained; failed to fit.")
raise
print("Stopped by user. {} models trained.".format(len(evos)))
self.evos = evos
self.best_evo = sorted(self.evos, key=lambda x: x[0])[-1][1]
# import warnings
# warnings.warn("best: {}".format(self.best_evo.best_estimator_))
print("best: {}".format(self.best_evo.best_estimator_))
return self.best_evo.best_estimator_
def apply_toy_on(X, y, cl_or_reg=None, n=500, max_tries=3):
n = min(len(y), n)
cl_or_reg = cl_or_reg if cl_or_reg else classification_or_regression(y)
if cl_or_reg == 'classification':
cross_split = StratifiedShuffleSplit(y, 1, int(0.2 * n), int(0.8 * n))
else:
cross_split = ShuffleSplit(len(y), max_tries, int(0.2 * n), int(0.8 * n))
if not isinstance(X, pd.DataFrame):
X = pd.DataFrame(X)
for train_index, test_index in cross_split:
try:
toy = Toy(cv=2, cl_or_reg=cl_or_reg)
toy.fit(X.iloc[train_index], y[train_index])
score = toy.score(X.iloc[test_index], y[test_index])
except NotFittedError: # probably have to find the real fix here...
score = -10000
print(score)
yield score
def fit(self, X, y):
self.X = X
self.y = y
if self.scoring is None:
tp = classification_or_regression(self.y)
# MULTIOUTPUT NOT SUPPORTED FOR REGRESSION (probably DEAPs fault)
self.scoring = ['f1_weighted', 'r2'][tp != 'classification']
self.pick_model()
self.get_pipeline()
unique_combinations = np.prod(list(map(len, self.grid.values())))
print("unique_combinations", unique_combinations)
if 'population_size' not in self.kwargs:
self.kwargs['population_size'] = np.clip(int(unique_combinations / 1000), 5, 10)
if 'generations_number' not in self.kwargs:
self.kwargs['generations_number'] = np.clip(int(unique_combinations / 200), 10, 50)
self.evo = evo_search(self.pipeline, self.grid, cv=self.cv,
scoring=self.scoring, n_jobs=self.n_jobs, **self.kwargs)
self.evo.fit(X, y)
return self.evo.best_estimator_
