File size: 2,969 Bytes
2890711
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from typing import Optional

import torchdata.datapipes.iter
import webdataset as wds
from omegaconf import DictConfig
from pytorch_lightning import LightningDataModule

try:
    from sdata import create_dataset, create_dummy_dataset, create_loader
except ImportError as e:
    print("#" * 100)
    print("Datasets not yet available")
    print("to enable, we need to add stable-datasets as a submodule")
    print("please use ``git submodule update --init --recursive``")
    print("and do ``pip install -e stable-datasets/`` from the root of this repo")
    print("#" * 100)
    exit(1)


class StableDataModuleFromConfig(LightningDataModule):
    def __init__(
        self,
        train: DictConfig,
        validation: Optional[DictConfig] = None,
        test: Optional[DictConfig] = None,
        skip_val_loader: bool = False,
        dummy: bool = False,
    ):
        super().__init__()
        self.train_config = train
        assert (
            "datapipeline" in self.train_config and "loader" in self.train_config
        ), "train config requires the fields `datapipeline` and `loader`"

        self.val_config = validation
        if not skip_val_loader:
            if self.val_config is not None:
                assert (
                    "datapipeline" in self.val_config and "loader" in self.val_config
                ), "validation config requires the fields `datapipeline` and `loader`"
            else:
                print(
                    "Warning: No Validation datapipeline defined, using that one from training"
                )
                self.val_config = train

        self.test_config = test
        if self.test_config is not None:
            assert (
                "datapipeline" in self.test_config and "loader" in self.test_config
            ), "test config requires the fields `datapipeline` and `loader`"

        self.dummy = dummy
        if self.dummy:
            print("#" * 100)
            print("USING DUMMY DATASET: HOPE YOU'RE DEBUGGING ;)")
            print("#" * 100)

    def setup(self, stage: str) -> None:
        print("Preparing datasets")
        if self.dummy:
            data_fn = create_dummy_dataset
        else:
            data_fn = create_dataset

        self.train_datapipeline = data_fn(**self.train_config.datapipeline)
        if self.val_config:
            self.val_datapipeline = data_fn(**self.val_config.datapipeline)
        if self.test_config:
            self.test_datapipeline = data_fn(**self.test_config.datapipeline)

    def train_dataloader(self) -> torchdata.datapipes.iter.IterDataPipe:
        loader = create_loader(self.train_datapipeline, **self.train_config.loader)
        return loader

    def val_dataloader(self) -> wds.DataPipeline:
        return create_loader(self.val_datapipeline, **self.val_config.loader)

    def test_dataloader(self) -> wds.DataPipeline:
        return create_loader(self.test_datapipeline, **self.test_config.loader)