def vis_debug_rf(matcher, train, test, exclude_attrs, target_attr):
"""
Visual debugger for random forest matcher
Parameters
----------
matcher : object, RFMatcher object
train : MTable, containing training data with "True" labels
test : MTable, containing test data with "True labels.
The "True" labels are used for evaluation.
exclude_attrs : List, attributes to be excluded from train and test,
for training and testing.
target_attr : String, column name in validation_set containing 'True' labels
"""
assert set(test.columns) == set(train.columns), "The train and test columns are not same"
assert set(train.columns).intersection(exclude_attrs) == set(exclude_attrs), "Some of exclude attrs are not part of" \
"train columns"
# redundant
assert set(test.columns).intersection(exclude_attrs) == set(exclude_attrs), "Some of exclude attrs are not part of" \
"test columns"
# fit using training data
matcher.fit(table=train, exclude_attrs=exclude_attrs, target_attr=target_attr)
predict_attr_name = get_name_for_predict_column(test.columns)
predicted = matcher.predict(table=test, exclude_attrs=exclude_attrs, target_attr=predict_attr_name, append=True,
inplace=False)
eval_summary = mg.eval_matches(predicted, target_attr, predict_attr_name)
metric = get_metric(eval_summary)
fp_dataframe = get_dataframe(predicted, eval_summary['false_pos_ls'])
fn_dataframe = get_dataframe(predicted, eval_summary['false_neg_ls'])
app = mg._viewapp
m = MainWindowManager(matcher, "rf", exclude_attrs, metric, predicted, fp_dataframe, fn_dataframe)
m.show()
app.exec_()
def _vis_debug_dt(matcher, train, test, exclude_attrs, target_attr, show_window=True):
if not isinstance(matcher, DTMatcher):
logger.error('Input matcher is not of type Decision Tree matcher')
raise AssertionError('Input matcher is not of type Decision Tree matcher')
if not isinstance(target_attr, six.string_types):
logger.error('Target attribute is not of type string')
raise AssertionError('Target attribute is not of type string')
if not check_attrs_present(train, exclude_attrs):
logger.error('The exclude attrs are not in train table columns')
raise AssertionError('The exclude attrs are not in the train table columns')
if not check_attrs_present(train, target_attr):
logger.error('The target attr is not in train table columns')
raise AssertionError('The target attr is not in the train table columns')
if not check_attrs_present(test, exclude_attrs):
logger.error('The exclude attrs are not in test table columns')
raise AssertionError('The exclude attrs are not in the test table columns')
if not isinstance(exclude_attrs, list):
exclude_attrs = [exclude_attrs]
exclude_attrs = list_drop_duplicates(exclude_attrs)
if target_attr not in exclude_attrs:
exclude_attrs.append(target_attr)
# fit using training data
matcher.fit(table=train, exclude_attrs=exclude_attrs, target_attr=target_attr)
predict_attr_name = get_name_for_predict_column(test.columns)
# predict using the test data
predicted = matcher.predict(table=test, exclude_attrs=exclude_attrs,
target_attr=predict_attr_name, append=True,
inplace=False)
eval_summary = mg.eval_matches(predicted, target_attr, predict_attr_name)
metric = get_metric(eval_summary)
fp_dataframe = get_dataframe(predicted, eval_summary['false_pos_ls'])
fn_dataframe = get_dataframe(predicted, eval_summary['false_neg_ls'])
app = mg._viewapp
m = MainWindowManager(matcher, "dt", exclude_attrs, metric, predicted, fp_dataframe,
fn_dataframe)
if show_window == True:
m.show()
app.exec_()
def vis_debug_dt(matcher, train, test, exclude_attrs, target_attr):
"""
Visual debugger for decision tree matcher
Parameters
----------
matcher : object, DTMatcher object
train : MTable, containing training data with "True" labels
test : MTable, containing test data with "True labels.
The "True" labels are used for evaluation.
exclude_attrs : List, attributes to be excluded from train and test,
for training and testing.
target_attr : String, column name in validation_set containing 'True' labels
"""
assert set(test.columns) == set(
train.columns), "The train and test columns are not same"
assert set(train.columns).intersection(exclude_attrs) == set(exclude_attrs), "Some of exclude attrs are not part of" \
"train columns"
# redundant
assert set(test.columns).intersection(exclude_attrs) == set(exclude_attrs), "Some of exclude attrs are not part of" \
"test columns"
# fit using training data
matcher.fit(table=train,
exclude_attrs=exclude_attrs,
target_attr=target_attr)
predict_attr_name = get_name_for_predict_column(test.columns)
predicted = matcher.predict(table=test,
exclude_attrs=exclude_attrs,
target_attr=predict_attr_name,
append=True,
inplace=False)
# print predicted
eval_summary = mg.eval_matches(predicted, target_attr, predict_attr_name)
# print eval_summary
metric = get_metric(eval_summary)
fp_dataframe = get_dataframe(predicted, eval_summary['false_pos_ls'])
# print fp_dataframe.dtypes
fn_dataframe = get_dataframe(predicted, eval_summary['false_neg_ls'])
app = mg._viewapp
m = MainWindowManager(matcher, "dt", exclude_attrs, metric, predicted,
fp_dataframe, fn_dataframe)
m.show()
app.exec_()
def vis_debug_rm(matcher, validation_set, target_attr, feature_table):
"""
Visual debugger for boolean rule-based matcher
Parameters
----------
matcher : object, Boolean rule-based matcher object
validation_set : MTable, used to debug
target_attr : String, column name in validation_set containing 'True' labels
feature_table : pandas dataframe, containing feature information
"""
predict_attr_name = get_name_for_predict_column(validation_set.columns)
predicted = matcher.predict(validation_set, predict_attr_name, append=True, inplace=False)
eval_summary = mg.eval_matches(predicted, target_attr, predict_attr_name)
metric = get_metric(eval_summary)
fp_dataframe = get_dataframe(predicted, eval_summary['false_pos_ls'])
fn_dataframe = get_dataframe(predicted, eval_summary['false_neg_ls'])
app = mg._viewapp
m = MainWindowManager(matcher, "rm", feature_table, metric, predicted, fp_dataframe, fn_dataframe)
m.show()
app.exec_()
import sys
import magellan as mg
from magellan.gui.debug_gui_base import vis_debug_dt
sys.path.append('/Users/Pradap/Documents/Research/Python-Package/enrique/')
mg.init_jvm()
A = mg.load_dataset('table_A')
B = mg.load_dataset('table_B')
ab = mg.AttrEquivalenceBlocker()
C = ab.block_tables(A, B, 'zipcode', 'zipcode', ['name'], ['name'])
L = mg.read_csv('label_demo.csv', ltable=A, rtable=B)
feat_table = mg.get_features_for_matching(A, B)
G = mg.extract_feature_vecs(L, feature_table=feat_table, attrs_after='gold')
S = mg.train_test_split(G, 8, 7)
dt = mg.DTMatcher(name='DecisionTree')
dt.fit(table = S['train'], exclude_attrs=['_id', 'ltable.ID', 'rtable.ID', 'gold'], target_attr='gold')
dt.predict(table=S['test'], exclude_attrs=['_id', 'ltable.ID', 'rtable.ID', 'gold'], target_attr='predicted',
append=True)
d = mg.eval_matches(S['test'], 'gold', 'predicted')
vis_debug_dt(dt, d, S['test'], exclude_attrs=['_id', 'ltable.ID', 'rtable.ID', 'gold'], feat_table=feat_table)
print "Hi"
def _vis_debug_dt(matcher,
train,
test,
exclude_attrs,
target_attr,
show_window=True):
if not isinstance(matcher, DTMatcher):
logger.error('Input matcher is not of type Decision Tree matcher')
raise AssertionError(
'Input matcher is not of type Decision Tree matcher')
if not isinstance(target_attr, six.string_types):
logger.error('Target attribute is not of type string')
raise AssertionError('Target attribute is not of type string')
if not check_attrs_present(train, exclude_attrs):
logger.error('The exclude attrs are not in train table columns')
raise AssertionError(
'The exclude attrs are not in the train table columns')
if not check_attrs_present(train, target_attr):
logger.error('The target attr is not in train table columns')
raise AssertionError(
'The target attr is not in the train table columns')
if not check_attrs_present(test, exclude_attrs):
logger.error('The exclude attrs are not in test table columns')
raise AssertionError(
'The exclude attrs are not in the test table columns')
if not isinstance(exclude_attrs, list):
exclude_attrs = [exclude_attrs]
exclude_attrs = list_drop_duplicates(exclude_attrs)
if target_attr not in exclude_attrs:
exclude_attrs.append(target_attr)
# fit using training data
matcher.fit(table=train,
exclude_attrs=exclude_attrs,
target_attr=target_attr)
predict_attr_name = get_name_for_predict_column(test.columns)
# predict using the test data
predicted = matcher.predict(table=test,
exclude_attrs=exclude_attrs,
target_attr=predict_attr_name,
append=True,
inplace=False)
eval_summary = mg.eval_matches(predicted, target_attr, predict_attr_name)
metric = get_metric(eval_summary)
fp_dataframe = get_dataframe(predicted, eval_summary['false_pos_ls'])
fn_dataframe = get_dataframe(predicted, eval_summary['false_neg_ls'])
app = mg._viewapp
m = MainWindowManager(matcher, "dt", exclude_attrs, metric, predicted,
fp_dataframe, fn_dataframe)
if show_window == True:
m.show()
app.exec_()
import sys
import magellan as mg
from magellan.gui.debug_gui_base import vis_debug_dt
sys.path.append('/Users/Pradap/Documents/Research/Python-Package/enrique/')
mg.init_jvm()
A = mg.load_dataset('table_A')
B = mg.load_dataset('table_B')
ab = mg.AttrEquivalenceBlocker()
C = ab.block_tables(A, B, 'zipcode', 'zipcode', ['name'], ['name'])
L = mg.read_csv('label_demo.csv', ltable=A, rtable=B)
feat_table = mg.get_features_for_matching(A, B)
G = mg.extract_feature_vecs(L, feature_table=feat_table, attrs_after='gold')
S = mg.train_test_split(G, 8, 7)
dt = mg.DTMatcher(name='DecisionTree')
dt.fit(table=S['train'],
exclude_attrs=['_id', 'ltable.ID', 'rtable.ID', 'gold'],
target_attr='gold')
dt.predict(table=S['test'],
exclude_attrs=['_id', 'ltable.ID', 'rtable.ID', 'gold'],
target_attr='predicted',
append=True)
d = mg.eval_matches(S['test'], 'gold', 'predicted')
vis_debug_dt(dt,
d,
S['test'],
exclude_attrs=['_id', 'ltable.ID', 'rtable.ID', 'gold'],
feat_table=feat_table)
print "Hi"
#rf.fit(table=train, exclude_attrs=['_id', 'ltable.id', 'rtable.id', 'gold'], target_attr='gold')
#mg.vis_debug_rf(rf, train, test, exclude_attrs=['_id', 'ltable.id', 'rtable.id', 'gold'],
# target_attr='gold')
# ret_val, node_list = mg.vis_tuple_debug_rf_matcher(rf, G.iloc[[0]],
# exclude_attrs=['_id', 'ltable.id', 'rtable.id', 'gold'])
# print ret_val
# print node_list
rm = mg.BooleanRuleMatcher()
rm.add_rule(['title_title_jac_qgm_3_qgm_3(ltuple, rtuple) > 0.6', 'lang_lang_lev(ltuple, rtuple) > 0.5'], feature_table=f)
rm.add_rule(['author_author_jac_qgm_3_qgm_3(ltuple, rtuple) > 0.6'], feature_table=f)
rm.add_rule(['binding_binding_jac_qgm_3_qgm_3(ltuple, rtuple) > 0.5'], feature_table=f)
X = rm.predict(table=L, target_attr='predicted', append=True, inplace=False)
metric = mg.eval_matches(X, 'gold', 'predicted')
print metric
# ret_val, node_list = mg.vis_tuple_debug_rm_matcher(rm, wal.iloc[0], bwk.iloc[1], feature_table=f)
# print ret_val
# print node_list
#mg.vis_debug_rm(rm, L, 'gold', feature_table=f)
trigger = mg.MatchTrigger()
trigger.add_cond_rule('title_title_jac_qgm_3_qgm_3(ltuple, rtuple) < 0.8', f)
trigger.add_cond_status(True)
trigger.add_action(0)
# print P[['_id', 'gold', 'predicted']]
Y = trigger.execute(X, 'predicted', inplace=False)
print Y
print 'Hi'