def __init__(self, node, **kwargs):
super(CVBestSearchRefit, self).__init__(estimator=None)
####?????
score = kwargs.pop("score") if "score" in kwargs else 'y/test/score_recall_mean'
arg_max = kwargs.pop("arg_max") if "arg_max" in kwargs else True
from epac.workflow.splitters import CV
#methods = Methods(*tasks)
self.cv = CV(node=node, reducer=ClassificationReport(keep=False), **kwargs)
self.score = score
self.arg_max = arg_max
def __init__(self, node, **kwargs):
super(CVBestSearchRefit, self).__init__(wrapped_node=None)
#### 'y/test/score_recall_mean'
default_score = "y" + conf.SEP + \
conf.TEST + conf.SEP + \
conf.SCORE_RECALL_MEAN
score = kwargs.pop("score") if "score" in kwargs else default_score
arg_max = kwargs.pop("arg_max") if "arg_max" in kwargs else True
from epac.workflow.splitters import CV
#methods = Methods(*tasks)
self.cv = CV(node=node,
reducer=ClassificationReport(keep=False),
**kwargs)
self.score = score
self.arg_max = arg_max
def __init__(self, node, **kwargs):
super(CVBestSearchRefit, self).__init__(wrapped_node=None)
#### 'y/test/score_recall_mean'
default_score = "y" + conf.SEP + \
conf.TEST + conf.SEP + \
conf.SCORE_RECALL_MEAN
score = kwargs.pop("score") if "score" in kwargs else default_score
arg_max = kwargs.pop("arg_max") if "arg_max" in kwargs else True
from epac.workflow.splitters import CV
#methods = Methods(*tasks)
self.cv = CV(node=node, reducer=ClassificationReport(keep=False),
**kwargs)
self.score = score
self.arg_max = arg_max
class CVBestSearchRefit(Estimator):
"""Cross-validation + grid-search then refit with optimals parameters.
Average results over first axis, then find the arguments that maximize or
minimise a "score" over other axis.
Parameters
----------
See CV parameters, plus other parameters:
score: string
the score name to be optimized (default "mean_score_te").
arg_max: boolean
True/False take parameters that maximize/minimize the score. Default
is True.
"""
def __init__(self, node, **kwargs):
super(CVBestSearchRefit, self).__init__(estimator=None)
####?????
score = kwargs.pop("score") if "score" in kwargs else 'y/test/score_recall_mean'
arg_max = kwargs.pop("arg_max") if "arg_max" in kwargs else True
from epac.workflow.splitters import CV
#methods = Methods(*tasks)
self.cv = CV(node=node, reducer=ClassificationReport(keep=False), **kwargs)
self.score = score
self.arg_max = arg_max
def get_signature(self):
return self.__class__.__name__
def transform(self, **Xy):
Xy_train, Xy_test = train_test_split(Xy)
if Xy_train is Xy_test:
to_refit, best_params = self._search_best(**Xy)
else:
to_refit, best_params = self._search_best(**Xy_train)
out = to_refit.top_down(**Xy)
out[conf.BEST_PARAMS] = best_params
self.refited = to_refit
self.best_params = best_params
return out
def _search_best(self, **Xy):
# Fit/predict CV grid search
self.cv.store = StoreMem() # local store erased at each fit
from epac.workflow.pipeline import Pipe
self.cv.top_down(**Xy)
# Pump-up results
cv_result_set = self.cv.reduce(store_results=False)
key_val = [(result.key(), result[self.score]) \
for result in cv_result_set]
scores = np.asarray(zip(*key_val)[1])
scores_opt = np.max(scores) if self.arg_max else np.min(scores)
idx_best = np.where(scores == scores_opt)[0][0]
best_key = key_val[idx_best][0]
# Find nodes that match the best
nodes_dict = {n.get_signature(): n for n in self.cv.walk_true_nodes() \
if n.get_signature() in key_split(best_key)}
to_refit = Pipe(*[nodes_dict[k].estimator for k in key_split(best_key)])
best_params = [dict(sig) for sig in key_split(best_key, eval=True)]
return to_refit, best_params
def reduce(self, store_results=True):
# Terminaison (leaf) node return result_set
return self.load_state(name="result_set")
class CVBestSearchRefit(Wrapper):
"""Cross-validation + grid-search then refit with optimals parameters.
Average results over first axis, then find the arguments that maximize or
minimise a "score" over other axis.
Parameters
----------
See CV parameters, plus other parameters:
score: string
the score name to be optimized (default "mean_score_te").
arg_max: boolean
True/False take parameters that maximize/minimize the score. Default
is True.
Example
-------
>>> from sklearn import datasets
>>> from sklearn.svm import SVC
>>> from epac import Methods
>>> from epac.workflow.splitters import CVBestSearchRefit
>>> X, y = datasets.make_classification(n_samples=12,
... n_features=10,
... n_informative=2,
... random_state=1)
>>> n_folds_nested = 2
>>> C_values = [.1, 0.5, 1, 2, 5]
>>> kernels = ["linear", "rbf"]
>>> methods = Methods(*[SVC(C=C, kernel=kernel)
... for C in C_values for kernel in kernels])
>>> wf = CVBestSearchRefit(methods, n_folds=n_folds_nested)
>>> wf.transform(X=X, y=y)
{'best_params': [{'kernel': 'rbf', 'C': 0.1, 'name': 'SVC'}], 'y/true': array([1, 0, 0, 1, 0, 0, 1, 0, 1, 1, 0, 1]), 'y/pred': array([1, 0, 0, 1, 0, 0, 1, 0, 1, 1, 0, 1])}
>>> wf.reduce()
>>> wf.run(X=X, y=y)
{'best_params': [{'kernel': 'rbf', 'C': 0.1, 'name': 'SVC'}], 'y/true': array([1, 0, 0, 1, 0, 0, 1, 0, 1, 1, 0, 1]), 'y/pred': array([1, 0, 0, 1, 0, 0, 1, 0, 1, 1, 0, 1])}
>>> wf.reduce()
ResultSet(
[{'key': CVBestSearchRefit, 'best_params': [{'kernel': 'rbf', 'C': 0.1, 'name': 'SVC'}], 'y/true': [1 0 0 1 0 0 1 0 1 1 0 1], 'y/pred': [1 0 0 1 0 0 1 0 1 1 0 1]}])
"""
def __init__(self, node, **kwargs):
super(CVBestSearchRefit, self).__init__(wrapped_node=None)
#### 'y/test/score_recall_mean'
default_score = "y" + conf.SEP + \
conf.TEST + conf.SEP + \
conf.SCORE_RECALL_MEAN
score = kwargs.pop("score") if "score" in kwargs else default_score
arg_max = kwargs.pop("arg_max") if "arg_max" in kwargs else True
from epac.workflow.splitters import CV
#methods = Methods(*tasks)
self.cv = CV(node=node,
reducer=ClassificationReport(keep=False),
**kwargs)
self.score = score
self.arg_max = arg_max
# warnings.warn("%s is deprecated. Please use %s instead." % \
# (self.__class__.__name__,\
# CVBestSearchRefitParallel.__name__),
# category=DeprecationWarning)
def get_signature(self):
return self.__class__.__name__
def transform(self, **Xy):
Xy_train, Xy_test = train_test_split(Xy)
if Xy_train is Xy_test:
to_refit, best_params = self._search_best(**Xy)
else:
to_refit, best_params = self._search_best(**Xy_train)
out = to_refit.top_down(**Xy)
out[conf.BEST_PARAMS] = best_params
self.refited = to_refit
self.best_params = best_params
return out
def _search_best(self, **Xy):
# Fit/predict CV grid search
self.cv.store = StoreMem() # local store erased at each fit
from epac.workflow.pipeline import Pipe
self.cv.top_down(**Xy)
# Pump-up results
cv_result_set = self.cv.reduce(store_results=False)
key_val = [(result.key(), result[self.score])
for result in cv_result_set]
scores = np.asarray(zip(*key_val)[1])
scores_opt = np.max(scores) if self.arg_max else np.min(scores)
idx_best = np.where(scores == scores_opt)[0][0]
best_key = key_val[idx_best][0]
# Find nodes that match the best
nodes_dict = {
n.get_signature(): n
for n in self.cv.walk_true_nodes()
if n.get_signature() in key_split(best_key)
}
to_refit = Pipe(
*[nodes_dict[k].wrapped_node for k in key_split(best_key)])
best_params = [dict(sig) for sig in key_split(best_key, eval=True)]
return to_refit, best_params
def reduce(self, store_results=True):
# Terminaison (leaf) node return result_set
return self.load_results()
class CVBestSearchRefit(Wrapper):
"""Cross-validation + grid-search then refit with optimals parameters.
Average results over first axis, then find the arguments that maximize or
minimise a "score" over other axis.
Parameters
----------
See CV parameters, plus other parameters:
score: string
the score name to be optimized (default "mean_score_te").
arg_max: boolean
True/False take parameters that maximize/minimize the score. Default
is True.
Example
-------
>>> from sklearn import datasets
>>> from sklearn.svm import SVC
>>> from epac import Methods
>>> from epac.workflow.splitters import CVBestSearchRefit
>>> X, y = datasets.make_classification(n_samples=12,
... n_features=10,
... n_informative=2,
... random_state=1)
>>> n_folds_nested = 2
>>> C_values = [.1, 0.5, 1, 2, 5]
>>> kernels = ["linear", "rbf"]
>>> methods = Methods(*[SVC(C=C, kernel=kernel)
... for C in C_values for kernel in kernels])
>>> wf = CVBestSearchRefit(methods, n_folds=n_folds_nested)
>>> wf.transform(X=X, y=y)
{'best_params': [{'kernel': 'rbf', 'C': 0.1, 'name': 'SVC'}], 'y/true': array([1, 0, 0, 1, 0, 0, 1, 0, 1, 1, 0, 1]), 'y/pred': array([1, 0, 0, 1, 0, 0, 1, 0, 1, 1, 0, 1])}
>>> wf.reduce()
>>> wf.run(X=X, y=y)
{'best_params': [{'kernel': 'rbf', 'C': 0.1, 'name': 'SVC'}], 'y/true': array([1, 0, 0, 1, 0, 0, 1, 0, 1, 1, 0, 1]), 'y/pred': array([1, 0, 0, 1, 0, 0, 1, 0, 1, 1, 0, 1])}
>>> wf.reduce()
ResultSet(
[{'key': CVBestSearchRefit, 'best_params': [{'kernel': 'rbf', 'C': 0.1, 'name': 'SVC'}], 'y/true': [1 0 0 1 0 0 1 0 1 1 0 1], 'y/pred': [1 0 0 1 0 0 1 0 1 1 0 1]}])
"""
def __init__(self, node, **kwargs):
super(CVBestSearchRefit, self).__init__(wrapped_node=None)
#### 'y/test/score_recall_mean'
default_score = "y" + conf.SEP + \
conf.TEST + conf.SEP + \
conf.SCORE_RECALL_MEAN
score = kwargs.pop("score") if "score" in kwargs else default_score
arg_max = kwargs.pop("arg_max") if "arg_max" in kwargs else True
from epac.workflow.splitters import CV
#methods = Methods(*tasks)
self.cv = CV(node=node, reducer=ClassificationReport(keep=False),
**kwargs)
self.score = score
self.arg_max = arg_max
# warnings.warn("%s is deprecated. Please use %s instead." % \
# (self.__class__.__name__,\
# CVBestSearchRefitParallel.__name__),
# category=DeprecationWarning)
def get_signature(self):
return self.__class__.__name__
def transform(self, **Xy):
Xy_train, Xy_test = train_test_split(Xy)
if Xy_train is Xy_test:
to_refit, best_params = self._search_best(**Xy)
else:
to_refit, best_params = self._search_best(**Xy_train)
out = to_refit.top_down(**Xy)
out[conf.BEST_PARAMS] = best_params
self.refited = to_refit
self.best_params = best_params
return out
def _search_best(self, **Xy):
# Fit/predict CV grid search
self.cv.store = StoreMem() # local store erased at each fit
from epac.workflow.pipeline import Pipe
self.cv.top_down(**Xy)
# Pump-up results
cv_result_set = self.cv.reduce(store_results=False)
key_val = [(result.key(), result[self.score])
for result in cv_result_set]
scores = np.asarray(zip(*key_val)[1])
scores_opt = np.max(scores) if self.arg_max else np.min(scores)
idx_best = np.where(scores == scores_opt)[0][0]
best_key = key_val[idx_best][0]
# Find nodes that match the best
nodes_dict = {n.get_signature(): n for n in self.cv.walk_true_nodes()
if n.get_signature() in key_split(best_key)}
to_refit = Pipe(*[nodes_dict[k].wrapped_node
for k in key_split(best_key)])
best_params = [dict(sig) for sig in key_split(best_key, eval=True)]
return to_refit, best_params
def reduce(self, store_results=True):
# Terminaison (leaf) node return result_set
return self.load_results()
