summaryrefslogtreecommitdiff
path: root/training/callbacks/wandb.py
blob: d9bb9b8f811aa4985f43f1be81a0451e1b2f22a7 (plain)
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
"""Weights and Biases callbacks."""
from pathlib import Path
from typing import Tuple

import wandb
from torch import Tensor
from pytorch_lightning import Callback, LightningModule, Trainer
from pytorch_lightning.loggers import WandbLogger
from pytorch_lightning.utilities import rank_zero_only


def get_wandb_logger(trainer: Trainer) -> WandbLogger:
    """Safely get W&B logger from Trainer."""

    for logger in trainer.loggers:
        if isinstance(logger, WandbLogger):
            return logger

    raise Exception("Weight and Biases logger not found for some reason...")


class WatchModel(Callback):
    """Make W&B watch the model at the beginning of the run."""

    def __init__(
        self,
        log_params: str = "gradients",
        log_freq: int = 100,
        log_graph: bool = False,
    ) -> None:
        self.log_params = log_params
        self.log_freq = log_freq
        self.log_graph = log_graph

    @rank_zero_only
    def on_train_start(self, trainer: Trainer, pl_module: LightningModule) -> None:
        """Watches model weights with wandb."""
        logger = get_wandb_logger(trainer)
        logger.watch(
            model=trainer.model,
            log=self.log_params,
            log_freq=self.log_freq,
            log_graph=self.log_graph,
        )


class UploadConfigAsArtifact(Callback):
    """Upload all *.py files to W&B as an artifact, at the beginning of the run."""

    def __init__(self) -> None:
        self.config_dir = Path(".hydra/")

    @rank_zero_only
    def on_train_start(self, trainer: Trainer, pl_module: LightningModule) -> None:
        """Uploads project code as an artifact."""
        logger = get_wandb_logger(trainer)
        experiment = logger.experiment
        artifact = wandb.Artifact("experiment-config", type="config")
        for filepath in self.config_dir.rglob("*.yaml"):
            artifact.add_file(str(filepath))

        experiment.use_artifact(artifact)


class UploadCheckpointsAsArtifact(Callback):
    """Upload checkpoint to wandb as an artifact, at the end of a run."""

    def __init__(
        self, ckpt_dir: str = "checkpoints/", upload_best_only: bool = False
    ) -> None:
        self.ckpt_dir = Path(ckpt_dir)
        self.upload_best_only = upload_best_only

    @rank_zero_only
    def on_train_end(self, trainer: Trainer, pl_module: LightningModule) -> None:
        """Uploads model checkpoint to W&B."""
        logger = get_wandb_logger(trainer)
        experiment = logger.experiment
        ckpts = wandb.Artifact("experiment-ckpts", type="checkpoints")

        if self.upload_best_only:
            ckpts.add_file(trainer.checkpoint_callback.best_model_path)
        else:
            for ckpt in (self.ckpt_dir).rglob("*.ckpt"):
                ckpts.add_file(ckpt)

        experiment.use_artifact(ckpts)


class ImageToCaptionLogger(Callback):
    """Logs the image and output caption."""

    def __init__(self, num_samples: int = 8, on_train: bool = True) -> None:
        self.num_samples = num_samples
        self.on_train = on_train
        self._required_keys = ("predictions", "ground_truths")

    def _log_captions(
        self, trainer: Trainer, batch: Tuple[Tensor, Tensor], outputs: dict, key: str
    ) -> None:
        xs, _ = batch
        preds, gts = outputs["predictions"], outputs["ground_truths"]
        xs, preds, gts = (
            list(xs[: self.num_samples]),
            preds[: self.num_samples],
            gts[: self.num_samples],
        )
        trainer.logger.log_image(key, xs, caption=preds)

    @rank_zero_only
    def on_train_batch_end(
        self,
        trainer: Trainer,
        pl_module: LightningModule,
        outputs: dict,
        batch: Tuple[Tensor, Tensor],
        batch_idx: int,
    ) -> None:
        """Logs predictions on validation batch end."""
        if self.has_metrics(outputs):
            self._log_captions(trainer, batch, outputs, "train/predictions")

    @rank_zero_only
    def on_validation_batch_end(
        self,
        trainer: Trainer,
        pl_module: LightningModule,
        outputs: dict,
        batch: Tuple[Tensor, Tensor],
        batch_idx: int,
        *args,
        # dataloader_idx: int,
    ) -> None:
        """Logs predictions on validation batch end."""
        if self.has_metrics(outputs):
            self._log_captions(trainer, batch, outputs, "val/predictions")

    @rank_zero_only
    def on_test_batch_end(
        self,
        trainer: Trainer,
        pl_module: LightningModule,
        outputs: dict,
        batch: Tuple[Tensor, Tensor],
        batch_idx: int,
        dataloader_idx: int,
    ) -> None:
        """Logs predictions on train batch end."""
        if self.has_metrics(outputs):
            self._log_captions(trainer, batch, outputs, "test/predictions")

    def has_metrics(self, outputs: dict) -> bool:
        return all(k in outputs.keys() for k in self._required_keys)