def clear_classifier(db, feature_type, data, algorithm):
print "function: " + __name__
if algorithm==None:
return my_classifier.error_json('algorithm must be designated')
clf_id = my_classifier.generate_clf_id(algorithm,feature_type,data)
query = {'_id':clf_id}
collection = db['classifiers']
data_count = collection.find(query).count()
if data_count==0:
return my_classifier.error_json("No classifiers are hit.")
try:
db['classifiers'].remove(query)
except:
return my_classifier.error_json(sys.exc_info()[1])
result = my_classifier.success_json()
result['event'] = {'_id': generate_event_id('clear_classifier', feature_type, clf_id )}
return result
def evaluate(db,feature_type, data,algorithm):
#print "function: evaluate"
# class_name2idのために識別器のデータを呼ぶ
clf_id = my_classifier.generate_clf_id(algorithm,feature_type,data)
#print "clf_id: " + clf_id
try:
record = db["classifiers"].find_one({'_id':clf_id})
if record == None:
return my_classifier.error_json("No classifier was found.")
except:
return my_classifier.error_json(sys.exc_info()[1])
#print record
name2id = record['class_name2id']
y = []
y_pred = []
weights = []
selector = data['selector']
selector['likelihood.'+clf_id] = {"$exists":True}
print selector
samples = get_training_samples(db,feature_type,False,selector)
for s in samples:
if not s['likelihood'].has_key(clf_id):
continue
y.append(name2id[s['ground_truth']])
likelihood = dict(s['likelihood'][clf_id])
pred_name = max([(v,k) for k,v in likelihood.items()])[1]
y_pred.append(name2id[pred_name])
if s.has_key('weight'):
weights.append(float(s['weight']))
else:
weights.append(1.0)
if not y:
return my_classifier.error_json("ERROR: samples are not found.")
result = my_classifier.success_json()
result['event'] = {'_id':generate_event_id('evaluate', feature_type, clf_id)}
id2name = record['class_id2name']
result['class_list'] = [id2name[k] for k in sorted(id2name.keys())]
# confution_matrix
cm = confusion_matrix(y, y_pred)
cm_json_searizable = []
for line in cm:
cm_json_searizable.append(line.tolist())
# id2name = record['class_id2name']
result['confusion_matrix'] = json.dumps(cm_json_searizable)
#print precision_score(y,y_pred,sample_weight=weights)
#print precision_score(y,y_pred)
#print weights
result['precision_score'] = precision_score(y,y_pred,average=None).tolist()#,sample_weight=weights)
result['recall_score'] = recall_score(y,y_pred,average=None).tolist()#,sample_weight=weights)
result['f1_score'] = f1_score(y,y_pred,average=None).tolist()#,sample_weight=weights)
return result
