Upload schema.py with huggingface_hub
Browse files
schema.py
ADDED
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from datasets import Features, Sequence, Value
|
2 |
+
from .operator import StreamInstanceOperatorValidator
|
3 |
+
|
4 |
+
from typing import Dict, Any, List
|
5 |
+
|
6 |
+
from dataclasses import field
|
7 |
+
|
8 |
+
UNITXT_DATASET_SCHEMA = Features(
|
9 |
+
{
|
10 |
+
"source": Value("string"),
|
11 |
+
"target": Value("string"),
|
12 |
+
"references": Sequence(Value("string")),
|
13 |
+
"metrics": Sequence(Value("string")),
|
14 |
+
"group": Value("string"),
|
15 |
+
"postprocessors": Sequence(Value("string")),
|
16 |
+
}
|
17 |
+
)
|
18 |
+
|
19 |
+
# UNITXT_METRIC_SCHEMA = Features({
|
20 |
+
# "predictions": Value("string", id="sequence"),
|
21 |
+
# "target": Value("string", id="sequence"),
|
22 |
+
# "references": Value("string", id="sequence"),
|
23 |
+
# "metrics": Value("string", id="sequence"),
|
24 |
+
# 'group': Value('string'),
|
25 |
+
# 'postprocessors': Value("string", id="sequence"),
|
26 |
+
# })
|
27 |
+
|
28 |
+
|
29 |
+
class ToUnitxtGroup(StreamInstanceOperatorValidator):
|
30 |
+
group: str
|
31 |
+
metrics: List[str] = None
|
32 |
+
postprocessors: List[str] = field(default_factory=lambda: ["to_string"])
|
33 |
+
remove_unnecessary_fields: bool = True
|
34 |
+
|
35 |
+
def process(self, instance: Dict[str, Any], stream_name: str = None) -> Dict[str, Any]:
|
36 |
+
if self.remove_unnecessary_fields:
|
37 |
+
for key in instance.keys():
|
38 |
+
if key not in UNITXT_DATASET_SCHEMA:
|
39 |
+
del instance[key]
|
40 |
+
|
41 |
+
instance["group"] = self.group
|
42 |
+
if self.metrics is not None:
|
43 |
+
instance["metrics"] = self.metrics
|
44 |
+
if self.postprocessors is not None:
|
45 |
+
instance["postprocessors"] = self.postprocessors
|
46 |
+
|
47 |
+
return instance
|
48 |
+
|
49 |
+
def validate(self, instance: Dict[str, Any], stream_name: str = None):
|
50 |
+
# verify the instance has the required schema
|
51 |
+
assert instance is not None, f"Instance is None"
|
52 |
+
assert isinstance(instance, dict), f"Instance should be a dict, got {type(instance)}"
|
53 |
+
assert all(
|
54 |
+
[key in instance for key in UNITXT_DATASET_SCHEMA]
|
55 |
+
), f"Instance should have the following keys: {UNITXT_DATASET_SCHEMA}"
|
56 |
+
UNITXT_DATASET_SCHEMA.encode_example(instance)
|