def _debug_decisiontree_matcher(decision_tree,
tuple_1,
tuple_2,
feature_table,
table_columns,
exclude_attrs,
ensemble_flag=False):
"""
This function is used to print the debug information for decision tree
and random forest matcher.
"""
# Get the classifier from the input object.
if isinstance(decision_tree, DTMatcher):
clf = decision_tree.clf
else:
clf = decision_tree
# Based on the exclude attributes derive the feature names.
if exclude_attrs is None:
feature_names = table_columns
else:
cols = [c not in exclude_attrs for c in table_columns]
feature_names = table_columns[cols]
# Get the python code based on the classifier, feature names and the
# boolean results.
code = _get_code(clf, feature_names, ['False', 'True'])
# Apply feature functions to get feature vectors.
feature_vectors = apply_feat_fns(tuple_1, tuple_2, feature_table)
# Wrap the code in a a function
code = _get_dbg_fn(code)
# Initialize a dictionary with the given feature vectors. This is
# important because the code must be linked with the values in the
# feature vectors.
code_dict = {}
code_dict.update(feature_vectors)
six.exec_(code, code_dict)
ret_val = code_dict['debug_fn']()
# Based on the ensemble flag, indent the output (as in RF, we need to
# indent it a bit further right).
if ensemble_flag is True:
spacer = " "
else:
spacer = ""
# Further, if the ensemble flag is True, then print the prob. for match
# and non-matches.
if ensemble_flag is True:
p = _get_prob(clf, tuple_1, tuple_2, feature_table, feature_names)
print(spacer + "Prob. for non-match : " + str(p[0]))
print(spacer + "Prob for match : " + str(p[1]))
return p
else:
# Else, just print the match status.
print(spacer + "Match status : " + str(ret_val))
def _debug_decisiontree_matcher(decision_tree, tuple_1, tuple_2,
feature_table, table_columns,
exclude_attrs,
ensemble_flag=False):
"""
This function is used to print the debug information for decision tree
and random forest matcher.
"""
# Get the classifier from the input object.
if isinstance(decision_tree, DTMatcher):
clf = decision_tree.clf
else:
clf = decision_tree
# Based on the exclude attributes derive the feature names.
if exclude_attrs is None:
feature_names = table_columns
else:
cols = [c not in exclude_attrs for c in table_columns]
feature_names = table_columns[cols]
# Get the python code based on the classifier, feature names and the
# boolean results.
code = _get_code(clf, feature_names, ['False', 'True'])
# Apply feature functions to get feature vectors.
feature_vectors = apply_feat_fns(tuple_1, tuple_2, feature_table)
# Wrap the code in a a function
code = _get_dbg_fn(code)
# Initialize a dictionary with the given feature vectors. This is
# important because the code must be linked with the values in the
# feature vectors.
code_dict = {}
code_dict.update(feature_vectors)
six.exec_(code, code_dict)
ret_val = code_dict['debug_fn']()
# Based on the ensemble flag, indent the output (as in RF, we need to
# indent it a bit further right).
if ensemble_flag is True:
spacer = " "
else:
spacer = ""
# Further, if the ensemble flag is True, then print the prob. for match
# and non-matches.
if ensemble_flag is True:
p = _get_prob(clf, tuple_1, tuple_2, feature_table, feature_names)
print(spacer + "Prob. for non-match : " + str(p[0]))
print(spacer + "Prob for match : " + str(p[1]))
return p
else:
# Else, just print the match status.
print(spacer + "Match status : " + str(ret_val))
def _get_prob(clf, t1, t2, feature_table, feature_names):
"""
Get the probability of the match status.
"""
# Get the feature vectors from the feature table and the input tuples.
feat_values = apply_feat_fns(t1, t2, feature_table)
feat_values = pd.Series(feat_values)
feat_values = feat_values[feature_names]
v = feat_values.values
v = v.reshape(1, -1)
# Use the classifier to predict the probability.
p = clf.predict_proba(v)
return p[0]
def _get_prob(clf, t1, t2, feature_table, feature_names):
"""
Get the probability of the match status.
"""
# Get the feature vectors from the feature table and the input tuples.
feat_values = apply_feat_fns(t1, t2, feature_table)
feat_values = pd.Series(feat_values)
feat_values = feat_values[feature_names]
v = feat_values.values
v = v.reshape(1, -1)
# Use the classifier to predict the probability.
p = clf.predict_proba(v)
return p[0]