ImageCaptioningDB/thirdai_utils.py at main · SunnyStardust/ImageCaptioningDB · GitHub

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
153
154
155
156
from thirdai import neural_db as ndb
from pathlib import Path
import pandas as pd
import shutil
from thirdai import bolt, licensing

class CSV(ndb.Document):
    def __init__(self, path, id_column, strong_columns, weak_columns, reference_columns) -> None:
        self.df = pd.read_csv(path)
        self.df = self.df.sort_values(id_column)
        assert len(self.df[id_column].unique()) == len(self.df[id_column])
        assert self.df[id_column].min() == 0
        assert self.df[id_column].max() == len(self.df[id_column]) - 1

        for col in strong_columns + weak_columns:
            self.df[col] = self.df[col].fillna("")

        self.path = Path(path)
        self._hash = ndb.utils.hash_file(path)
        self.id_column = id_column
        self.strong_columns = strong_columns
        self.weak_columns = weak_columns
        self.reference_columns = reference_columns

    def hash(self) -> str:
        return self._hash

    def size(self) -> int:
        return len(self.df)

    def name(self) -> str:
        return self.path.name

    def strong_text(self, element_id: int) -> str:
        row = self.df.iloc[element_id]
        return " ".join([row[col] for col in self.strong_columns])

    def weak_text(self, element_id: int) -> str:
        row = self.df.iloc[element_id]
        return " ".join([row[col] for col in self.weak_columns])

    def reference(self, element_id: int) -> ndb.Reference:
        row = self.df.iloc[element_id]
        text = " ".join([row[col] for col in self.reference_columns])
        return ndb.Reference(
            document=self,
            element_id=element_id,
            text=text,
            source=str(self.path.absolute()),
            metadata={})

    def context(self, element_id: int, radius) -> str:
        rows = self.df.iloc[
            max(0, element_id - radius):
            min(len(self.df), element_id + radius)]
        return " ".join([row[col] for col in self.reference_columns for row in rows])

    def save_meta(self, directory: Path):
        # Let's copy the original CSV file to the provided directory
        shutil.copy(self.path, directory)

    def load_meta(self, directory: Path):
        # Since we've moved the CSV file to the provided directory, let's make
        # sure that we point to this CSV file.
        self.path = directory / self.path.name


def bolt_and_csv_to_checkpoint(
    raw_model,
    csv_location,
    checkpoint_location,
    model_input_dim = 50000,
    model_hidden_dim = 2048,
    id_col = "DOC_ID",
    query_col = "QUERY",
    id_delimiter = ":",
    strong_cols = ["passage"],
    weak_cols = ["para"],
    display_cols = ["passage"],
):
    # raw_model = bolt.UniversalDeepTransformer.load(bolt_location)
    raw_model._get_model().freeze_hash_tables()

    from student import models

    model = models.Mach(
        id_col=id_col,
        id_delimiter=id_delimiter,
        query_col=query_col,
        input_dim=model_input_dim,
        hidden_dim=model_hidden_dim,
    )

    model.model = raw_model

    from student.documents import CSV, DocList

    doc = CSV(
        path=csv_location,
        id_col=id_col,
        strong_cols=strong_cols,
        weak_cols=weak_cols,
        display_cols=display_cols,
    )

    doclist = DocList()
    doclist.add_document(doc)

    model.n_ids = doc.get_num_new_ids()

    model.balancing_samples = models.make_balancing_samples(doc)

    from student.loggers import InMemoryLogger, LoggerList
    from student.state import State

    state = State(model, logger=InMemoryLogger())
    state.documents = doclist

    state.save(checkpoint_location)

def bolt_to_checkpoint(
    raw_model,
    checkpoint_location,
    model_input_dim = 50000,
    model_hidden_dim = 2048,
    id_col = "DOC_ID",
    query_col = "QUERY",
    id_delimiter = ":",
):
    # raw_model = bolt.UniversalDeepTransformer.load(bolt_location)
    raw_model.clear_index()
    raw_model._get_model().freeze_hash_tables()

    from student import models

    model = models.Mach(
        id_col=id_col,
        id_delimiter=id_delimiter,
        query_col=query_col,
        input_dim=model_input_dim,
        hidden_dim=model_hidden_dim,
    )

    model.model = raw_model

    from student.documents import CSV, DocList

    doclist = DocList()

    from student.loggers import InMemoryLogger, LoggerList
    from student.state import State

    state = State(model, logger=InMemoryLogger())
    state.documents = doclist

    state.save(checkpoint_location)