def get_attr_scores(self, data, stop_at=0, progress_callback=None):
"""Return a dictionary mapping attributes to scores.
A score is a step number at which the attribute
was removed from the recursive evaluation.
"""
iter = 1
attrs = data.domain.attributes
attrScores = {}
while len(attrs) > stop_at:
weights = get_linear_svm_weights(self.learner(data), sum=False)
if progress_callback:
progress_callback(100. * iter / (len(attrs) - stop_at))
score = dict.fromkeys(attrs, 0)
for w in weights:
for attr, wAttr in w.items():
score[attr] += wAttr ** 2
score = score.items()
score.sort(lambda a, b:cmp(a[1], b[1]))
numToRemove = max(int(len(attrs) * 1.0 / (iter + 1)), 1)
for attr, s in score[:numToRemove]:
attrScores[attr] = len(attrScores)
attrs = [attr for attr, s in score[numToRemove:]]
if attrs:
data = data.select(attrs + [data.domain.classVar])
iter += 1
return attrScores
def score_smiles_variables(data, candidates=None, max_sample=20):
if candidates is None:
candidates = data.domain.variables + data.domain.getmetas().values()
candidates = filter(
lambda v: isinstance(v, (Orange.feature.Discrete, Orange.feature.
String)), candidates)
if len(data) > max_sample:
indices = Orange.data.sample.SubsetIndices2(data, max_sample)
data = data.select(indices, 0)
scored = [(var, sum([is_valid_smiles(str(e[var])) for e in data]))
for var in candidates]
return scored
def score_smiles_variables(data, candidates=None, max_sample=20):
if candidates is None:
candidates = data.domain.variables + data.domain.getmetas().values()
candidates = filter(lambda v: isinstance(v, (Orange.feature.Discrete,
Orange.feature.String)),
candidates)
if len(data) > max_sample:
indices = Orange.data.sample.SubsetIndices2(data, max_sample)
data = data.select(indices, 0)
scored = [(var, sum([is_valid_smiles(str(e[var])) for e in data]))
for var in candidates]
return scored
def get_attr_scores(self, data, stop_at=0, progress_callback=None):
"""Return a dictionary mapping attributes to scores.
A score is a step number at which the attribute
was removed from the recursive evaluation.
"""
iter = 1
attrs = data.domain.attributes
attr_scores = {}
scorer = ScoreSVMWeights(learner=self.learner)
while len(attrs) > stop_at:
scores = [(scorer(attr, data), attr) for attr in attrs]
if progress_callback:
progress_callback(100. * iter / (len(attrs) - stop_at))
scores = sorted(scores)
num_to_remove = max(int(len(attrs) * 1.0 / (iter + 1)), 1)
for s, attr in scores[:num_to_remove]:
attr_scores[attr] = len(attr_scores)
attrs = [attr for s, attr in scores[num_to_remove:]]
if attrs:
data = data.select(attrs + [data.domain.class_var])
iter += 1
return attr_scores
