def get_tree_loc_auc(self, data, i):
if "loc" not in data:
return 8
# self.predict will add/modify the 'prediction' column to the data
self.predict(data, i)
sorted_data = data.sort_values(by=["prediction", "loc"],
ascending=[False, True])
return get_auc(sorted_data)
def eval_range_split(self, level, cur_selected, cur_performance, data, cue,
indexes, interval, decision):
TP, FP, TN, FN = cur_performance
pos = data.iloc[indexes]
neg = data.iloc[~data.index.isin(indexes)]
if decision == 1:
decided = pos
undecided = neg
else:
pos, neg = neg, pos
decided = neg
undecided = pos
# get auc for loc.
if "loc" in data:
sorted_data = pd.concat([
df.sort_values(by=["loc"], ascending=True)
for df in [pos, neg]
])
loc_auc = get_auc(sorted_data)
else:
loc_auc = 0
tp = pos.loc[pos[self.target] == 1]
fp = pos.loc[pos[self.target] == 0]
tn = neg.loc[neg[self.target] == 0]
fn = neg.loc[neg[self.target] == 1]
metrics = map(len, [tp, fp, tn, fn])
tp, fp, tn, fn = self.update_metrics(level, self.max_depth, decision,
metrics)
# if the decision lead to no data, punish the score
if sum([tp, fp, tn, fn]) == 0:
score = float('inf')
elif self.criteria == "LOC_AUC":
score = loc_auc
else:
score = get_score(self.criteria,
[TP + tp, FP + fp, TN + tn, FN + fn])
if not cur_selected or score < cur_selected['score']:
direction = "inside" if decision else "outside"
cur_selected = {'rule': (cue, direction, interval, decision),\
'undecided': undecided,\
'metrics': [TP + tp, FP + fp, TN + tn, FN + fn],\
'score': score}
'''
if self.sorted_cues:
self.sorted_cues = [cur_selected].extend(self.sorted_cues)
else:
self.sorted_cues = [cur_selected]
else:
direction = "inside" if decision else "outside"
new_cue = {'rule': (cue, direction, interval, decision),
'undecided': undecided,
'metrics': [TP + tp, FP + fp, TN + tn, FN + fn],
'score': score}
self.sorted_cues.append(new_cue)
'''
return cur_selected
def eval_decision(self, data, cue, direction, threshold, decision):
try:
if type(threshold) != type(np.ndarray(1)):
if direction == ">":
decided, undecided = data.loc[
data[cue] > threshold], data.loc[
data[cue] <= threshold]
else:
decided, undecided = data.loc[
data[cue] < threshold], data.loc[
data[cue] >= threshold]
else:
decided = data.loc[(data[cue] >= threshold[0])
& (data[cue] < threshold[1])]
undecided = data.loc[(data[cue] < threshold[0]) |
(data[cue] >= threshold[1])]
except:
print "Exception"
return 1, 2, 3
if decision == 1:
pos, neg = decided, undecided
else:
pos, neg = undecided, decided
# get auc for loc.
if "loc" in data:
sorted_data = pd.concat([
df.sort_values(by=["loc"], ascending=True)
for df in [pos, neg]
])
loc_auc = get_auc(sorted_data)
else:
loc_auc = 0
tp = pos.loc[pos[self.target] == 1]
fp = pos.loc[pos[self.target] == 0]
tn = neg.loc[neg[self.target] == 0]
fn = neg.loc[neg[self.target] == 1]
# pre, rec, spec, fpr, npv, acc, f1 = get_performance([tp, fp, tn, fn])
# return undecided, [tp, fp, tn, fn, pre, rec, spec, fpr, npv, acc, f1]
return undecided, map(len, [tp, fp, tn, fn]), loc_auc