|
1
|
|
|
""" |
|
2
|
|
|
Unit initialization. |
|
3
|
|
|
""" |
|
4
|
1 |
|
class Unit(): |
|
5
|
|
|
""" |
|
6
|
|
|
Performs general statistics over the units in the jobs. |
|
7
|
|
|
""" |
|
8
|
|
|
|
|
9
|
1 |
|
@staticmethod |
|
10
|
|
|
def aggregate(judgments, config): |
|
11
|
|
|
""" |
|
12
|
|
|
Aggregates information for each unit in the job. For each unit we save the |
|
13
|
|
|
data that was used as input (in the crowdsourcing template), the job in which |
|
14
|
|
|
it appeared, the number of workers that annotated the unit and the total |
|
15
|
|
|
amount of time spent by the workers to annotate it. |
|
16
|
|
|
|
|
17
|
|
|
Args: |
|
18
|
|
|
judgments: Judgments contained in the job. |
|
19
|
|
|
config: Job configuration as provided as input for the metrics. |
|
20
|
|
|
|
|
21
|
|
|
Returns: |
|
22
|
|
|
A dataframe containing all units that appear in the jobs and the |
|
23
|
|
|
statistics relevant for them. |
|
24
|
|
|
""" |
|
25
|
1 |
|
agg = {} |
|
26
|
1 |
|
for col in config.input.values(): |
|
27
|
|
|
# for each input column the first value is taken. |
|
28
|
|
|
# all rows have the same value for each unit. |
|
29
|
1 |
|
agg[col] = 'first' |
|
30
|
1 |
|
for col in config.output.values(): |
|
31
|
|
|
# each output column dict is summed |
|
32
|
1 |
|
agg[col] = 'sum' |
|
33
|
1 |
|
agg['job'] = 'first' |
|
34
|
1 |
|
agg['worker'] = 'count' |
|
35
|
1 |
|
agg['duration'] = 'mean' |
|
36
|
|
|
|
|
37
|
1 |
|
units = judgments.groupby('unit').agg(agg) |
|
38
|
|
|
|
|
39
|
|
|
# |
|
40
|
|
|
# get unit metrics |
|
41
|
|
|
# |
|
42
|
|
|
# for each vector in the unit get the unit metrics |
|
43
|
1 |
|
units = units.apply(lambda row: Unit.get_metrics(row, config), axis=1) |
|
44
|
|
|
|
|
45
|
|
|
# sort columns |
|
46
|
1 |
|
units = units.reindex(sorted(units.columns), axis=1) |
|
47
|
|
|
|
|
48
|
1 |
|
return units |
|
49
|
|
|
|
|
50
|
1 |
|
@staticmethod |
|
51
|
|
|
def get_metrics(row, config): |
|
52
|
|
|
""" |
|
53
|
|
|
Counts the number of annotations and the number of unique annotations for each unit. |
|
54
|
|
|
""" |
|
55
|
1 |
|
for col in config.output.values(): |
|
56
|
1 |
|
row[col+'.unique_annotations'] = len(row[col]) |
|
57
|
1 |
|
row[col+'.annotations'] = sum(row[col].values()) |
|
58
|
|
|
return row |
|
59
|
|
|
|
CrowdTruth /
CrowdTruth-core
