def fit(self, data, **kwargs):
"""
Fit the classifier to given training data.
:param data: instance of DataManager
:return: self
"""
metric = accuracy_score if 'metric' not in kwargs else kwargs['metric']
# TODO:Automated feature engineering
if isinstance(data, pd.DataFrame):
self.pre_pipeline = DP_Pipeline(None)
data = self.pre_pipeline.execute(data, phase='train')
# Check the task type: {binary, multiclass}
task_type = type_of_target(data.train_y)
if task_type in [
'multiclass-multioutput', 'continuous',
'continuous-multioutput', 'unknown'
]:
raise ValueError("UNSUPPORTED TASK TYPE: %s!" % task_type)
self.task_type = task_type
kwargs['task_type'] = task_type
# Options for multiclass averaging.
average = 'weighted'
metric = get_metric(metric)
kwargs['metric'] = metric
super().fit(data, **kwargs)
return self
def fit(self, data, **kwargs):
"""
Fit the regressor to given training data.
:param data: instance of DataManager
:return: self
"""
metric = mean_squared_error if 'metric' not in kwargs else kwargs[
'metric']
# TODO:Automated feature engineering
if isinstance(data, pd.DataFrame):
self.pre_pipeline = DP_Pipeline(None)
data = self.pre_pipeline.execute(data,
phase='train',
stratify=False)
# Check the task type: {continuous}
task_type = type_of_target(data.train_y)
if task_type != 'continuous':
raise ValueError("UNSUPPORTED TASK TYPE: %s!" % task_type)
self.task_type = task_type
kwargs['task_type'] = task_type
metric = get_metric(metric)
kwargs['metric'] = metric
super().fit(data, **kwargs)
return self
def fit(self, data, **kwargs):
"""Fit the classifier to given training data.
Parameters
----------
data : instance of DataManager or DataFrame
metric : callable, optional (default='autosklearn.metrics.accuracy_score').
feat_type : list, optional (default=None)
List of str of `len(X.shape[1])` describing the attribute type.
Possible types are `Categorical` and `Numerical`. `Categorical`
attributes will be automatically One-Hot encoded. The values
used for a categorical attribute must be integers, obtained for
example by `sklearn.preprocessing.LabelEncoder
<http://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.LabelEncoder.html>`_.
dataset_name : str, optional (default=None)
Create nicer output. If None, a string will be determined by the
md5 hash of the dataset.
Returns
-------
self
"""
metric = accuracy_score if 'metric' not in kwargs else kwargs['metric']
# feat_type = None if 'feat_type' not in kwargs else kwargs['feat_type']
# dataset_name = None if 'dataset_name' not in kwargs else kwargs['dataset_name']
# # The number of evaluations.
# runcount = None if 'runcount' not in kwargs else kwargs['runcount']
# TODO:Automated feature engineering
if isinstance(data, pd.DataFrame):
self.pre_pipeline = DP_Pipeline(None)
data = self.pre_pipeline.execute(data, phase='train')
# Check the task type: {binary, multiclass}
task_type = type_of_target(data.train_y)
if task_type in [
'multiclass-multioutput', 'continuous',
'continuous-multioutput', 'unknown'
]:
raise ValueError("UNSUPPORTED TASK TYPE: %s!" % task_type)
self.task_type = task_type
kwargs['task_type'] = task_type
# Options for multiclass averaging.
average = 'weighted'
metric = get_metric(metric)
kwargs['metric'] = metric
super().fit(data, **kwargs)
return self
def _get_performance(self, train_x, train_y, valid_x, valid_y, model):
metric = get_metric(self.metricstr)
model.fit(train_x, train_y)
if self.metricstr == 'auc':
pred = model.predict_proba(valid_x)[:, 1]
else:
pred = model.predict(valid_x)
return metric(pred, valid_y)
def __init__(self,
max_iter,
metrics,
stratify,
model=LogisticRegression(multi_class='auto',
solver='liblinear')):
self.max_iter = max_iter
self.model = model
self.metricstr = metrics
self.stratify = stratify
self.metric = get_metric(metrics)
self.feature_sets = None
self.feature_cols = dict()
self.train_data = None
self.valid_data = None
self.numerical_features = None
self.init_length = None
def fit(self, data, **kwargs):
"""Fit the regressor to given training data.
Parameters
----------
data : instance of DataManager.
metric : callable, optional (default='autosklearn.metrics.mean_squared_error').
dataset_name : str, optional (default=None)
Create nicer output. If None, a string will be determined by the
md5 hash of the dataset.
Returns
-------
self
"""
# feat_type = None if 'feat_type' not in kwargs else kwargs['feat_type']
# dataset_name = None if 'dataset_name' not in kwargs else kwargs['dataset_name']
# # The number of evaluations.
# runcount = None if 'runcount' not in kwargs else kwargs['runcount']
# TODO:Automated feature engineering
if isinstance(data, pd.DataFrame):
self.pre_pipeline = DP_Pipeline(None)
data = self.pre_pipeline.execute(data, phase='train', stratify=False)
# Check the task type: {continuous}
task_type = type_of_target(data.train_y)
if task_type != 'continuous':
raise ValueError("UNSUPPORTED TASK TYPE: %s!" % task_type)
self.task_type = task_type
kwargs['task_type'] = task_type
metric = mean_squared_error if 'metric' not in kwargs else kwargs['metric']
metric = get_metric(metric)
kwargs['metric'] = metric
super().fit(data, **kwargs)
return self
def _selection_process(self, dm):
generated_train_data, generated_valid_data, generated_test_data = self.solver.transform(
)
features = np.load("features_27.npz")
generated_train_data, generated_valid_data, generated_test_data = features[
"train"], features["valid"], None
feature_num = dm.train_X.shape[1]
if feature_num < 20:
dm.train_X = generated_train_data
dm.val_X = generated_valid_data
dm.test_X = generated_test_data
else:
print("start selection process...............")
selector = RandomForestSelector()
selector.fit(dm.train_X, dm.train_y)
lr = LogisticRegression()
best_perf = get_metric(self.metrics, generated_train_data,
dm.train_y, generated_valid_data, dm.val_y,
lr)
best_k = 0
for percentile in range(1, 10):
k = int((percentile / 10.0) * feature_num)
selected_train_data = selector.transform(dm.train_X, k)
selected_valid_data = selector.transform(dm.val_X, k)
perf = get_metric(metrics=self.metrics,
x_train=np.hstack((generated_train_data,
selected_train_data)),
y_train=dm.train_y,
x_valid=np.hstack((generated_valid_data,
selected_valid_data)),
y_valid=dm.val_y,
model=lr)
if perf <= best_perf:
break
else:
print("selected pertentile:",
percentile,
"perf:",
best_perf,
flush=True)
best_perf = perf
best_k = k
if best_k != 0:
dm.train_X = np.hstack((generated_train_data,
selector.transform(dm.train_X,
best_k)))
dm.val_X = np.hstack((generated_valid_data,
selector.transform(dm.val_X, best_k)))
if dm.test_X is not None:
dm.test_X = np.hstack(
(generated_test_data,
selector.transform(dm.test_X, best_k)))
else:
dm.train_X = generated_train_data
dm.val_X = generated_valid_data
dm.test_X = generated_test_data