File size: 4,949 Bytes
3da1d9d
d08fbc6
066c396
d08fbc6
066c396
9d5b4c0
d08fbc6
0a1b314
d08fbc6
 
 
066c396
 
 
 
 
 
 
d08fbc6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
066c396
3da1d9d
0a1b314
066c396
 
 
 
d08fbc6
 
 
 
 
 
9d5b4c0
d08fbc6
066c396
 
b462f85
 
 
 
 
7f6dcb7
d08fbc6
 
 
 
 
 
 
 
 
 
 
 
7cdc7d0
 
7f6dcb7
b462f85
f6ebc4f
7cdc7d0
 
 
b462f85
7cdc7d0
d08fbc6
7cdc7d0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d08fbc6
3da1d9d
7f6dcb7
066c396
f4655a2
 
066c396
d08fbc6
f4655a2
 
 
 
d08fbc6
7cdc7d0
d08fbc6
 
 
 
 
 
 
 
 
 
 
 
 
 
9d5b4c0
 
 
 
 
 
 
 
d08fbc6
066c396
 
7f6dcb7
066c396
7f6dcb7
 
 
 
d08fbc6
066c396
d08fbc6
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
import json
from typing import Any, Dict, List, Optional

from datasets import Audio, Features, Image, Sequence, Value

from .artifact import Artifact
from .dict_utils import dict_get
from .operator import InstanceOperatorValidator
from .settings_utils import get_constants

constants = get_constants()

UNITXT_DATASET_SCHEMA = Features(
    {
        "source": Value("string"),
        "target": Value("string"),
        "references": Sequence(Value("string")),
        "metrics": Sequence(Value("string")),
        "groups": Sequence(Value("string")),
        "subset": Sequence(Value("string")),
        "media": {
            "images": Sequence(Image()),
            "audios": Sequence(Audio()),
        },
        "postprocessors": Sequence(Value("string")),
        "task_data": Value(dtype="string"),
        "data_classification_policy": Sequence(Value("string")),
    }
)

UNITXT_INFERENCE_SCHEMA = Features(
    {
        "source": Value("string"),
        "metrics": Sequence(Value("string")),
        "groups": Sequence(Value("string")),
        "subset": Sequence(Value("string")),
        "postprocessors": Sequence(Value("string")),
        "task_data": Value(dtype="string"),
        "data_classification_policy": Sequence(Value("string")),
    }
)


def get_schema(stream_name):
    if stream_name == constants.inference_stream:
        return UNITXT_INFERENCE_SCHEMA
    return UNITXT_DATASET_SCHEMA


class Finalize(InstanceOperatorValidator):
    group_by: List[List[str]]
    remove_unnecessary_fields: bool = True

    @staticmethod
    def artifact_to_jsonable(artifact):
        if artifact.__id__ is None:
            return artifact.to_dict()
        return artifact.__id__

    def _prepare_media(self, instance):
        if "media" not in instance:
            instance["media"] = {}

        if "images" not in instance["media"]:
            instance["media"]["images"] = []

        if "audios" not in instance["media"]:
            instance["media"]["audios"] = []

        return instance

    def _get_instance_task_data(
        self, instance: Dict[str, Any], use_reference_fields=True
    ) -> Dict[str, Any]:
        task_data = {
            **instance["input_fields"],
            "metadata": {
                "data_classification_policy": instance["data_classification_policy"],
            },
        }
        if use_reference_fields:
            task_data = {**task_data, **instance["reference_fields"]}
        return task_data

    def process(
        self, instance: Dict[str, Any], stream_name: Optional[str] = None
    ) -> Dict[str, Any]:
        task_data = self._get_instance_task_data(
            instance,
            use_reference_fields=stream_name != constants.inference_stream,
        )

        task_data["metadata"]["num_demos"] = instance["recipe_metadata"]["num_demos"]
        task_data["metadata"]["template"] = self.artifact_to_jsonable(
            instance["recipe_metadata"]["template"]
        )
        if "demos" in instance:
            task_data["demos"] = [
                self._get_instance_task_data(instance)
                for instance in instance.pop("demos")
            ]

        instance["task_data"] = json.dumps(task_data)

        if self.remove_unnecessary_fields:
            keys_to_delete = []

            for key in instance.keys():
                if key not in get_schema(stream_name):
                    keys_to_delete.append(key)

            for key in keys_to_delete:
                del instance[key]

        data = {**task_data, **task_data["metadata"]}
        groups = []
        for group_attributes in self.group_by:
            group = {}
            if isinstance(group_attributes, str):
                group_attributes = [group_attributes]
            for attribute in group_attributes:
                group[attribute] = dict_get(data, attribute)
            groups.append(json.dumps(group))

        instance["groups"] = groups
        instance["subset"] = []

        instance = self._prepare_media(instance)

        instance["metrics"] = [
            metric.to_json() if isinstance(metric, Artifact) else metric
            for metric in instance["metrics"]
        ]
        instance["postprocessors"] = [
            processor.to_json() if isinstance(processor, Artifact) else processor
            for processor in instance["postprocessors"]
        ]

        return instance

    def validate(self, instance: Dict[str, Any], stream_name: Optional[str] = None):
        # verify the instance has the required schema
        assert instance is not None, "Instance is None"
        assert isinstance(
            instance, dict
        ), f"Instance should be a dict, got {type(instance)}"
        schema = get_schema(stream_name)
        assert all(
            key in instance for key in schema
        ), f"Instance should have the following keys: {schema}. Instance is: {instance}"
        schema.encode_example(instance)