Spaces:
Sleeping
Sleeping
File size: 10,292 Bytes
e45d058 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 |
import subprocess
from pathlib import Path
from typing import List
import matplotlib.pyplot as plt
import seaborn as sn
import torch
import wandb
from pytorch_lightning import Callback, Trainer
from pytorch_lightning.loggers import LoggerCollection, WandbLogger
from pytorch_lightning.utilities import rank_zero_only
from sklearn import metrics
from sklearn.metrics import f1_score, precision_score, recall_score
def get_wandb_logger(trainer: Trainer) -> WandbLogger:
"""Safely get Weights&Biases logger from Trainer."""
if trainer.fast_dev_run:
raise Exception(
"Cannot use wandb callbacks since pytorch lightning disables loggers in `fast_dev_run=true` mode."
)
if isinstance(trainer.logger, WandbLogger):
return trainer.logger
if isinstance(trainer.logger, LoggerCollection):
for logger in trainer.logger:
if isinstance(logger, WandbLogger):
return logger
raise Exception(
"You are using wandb related callback, but WandbLogger was not found for some reason..."
)
class WatchModel(Callback):
"""Make wandb watch model at the beginning of the run."""
def __init__(self, log: str = "gradients", log_freq: int = 100):
self.log = log
self.log_freq = log_freq
@rank_zero_only
def on_train_start(self, trainer, pl_module):
logger = get_wandb_logger(trainer=trainer)
logger.watch(model=trainer.model, log=self.log, log_freq=self.log_freq)
class UploadCodeAsArtifact(Callback):
"""Upload all code files to wandb as an artifact, at the beginning of the run."""
def __init__(self, code_dir: str, use_git: bool = True):
"""
Args:
code_dir: the code directory
use_git: if using git, then upload all files that are not ignored by git.
if not using git, then upload all '*.py' file
"""
self.code_dir = code_dir
self.use_git = use_git
@rank_zero_only
def on_train_start(self, trainer, pl_module):
logger = get_wandb_logger(trainer=trainer)
experiment = logger.experiment
code = wandb.Artifact("project-source", type="code")
if self.use_git:
# get .git folder
# https://alexwlchan.net/2020/11/a-python-function-to-ignore-a-path-with-git-info-exclude/
git_dir_path = Path(
subprocess.check_output(["git", "rev-parse", "--git-dir"]).strip().decode("utf8")
).resolve()
for path in Path(self.code_dir).resolve().rglob("*"):
if (
path.is_file()
# ignore files in .git
and not str(path).startswith(str(git_dir_path)) # noqa: W503
# ignore files ignored by git
and ( # noqa: W503
subprocess.run(["git", "check-ignore", "-q", str(path)]).returncode == 1
)
):
code.add_file(str(path), name=str(path.relative_to(self.code_dir)))
else:
for path in Path(self.code_dir).resolve().rglob("*.py"):
code.add_file(str(path), name=str(path.relative_to(self.code_dir)))
experiment.log_artifact(code)
class UploadCheckpointsAsArtifact(Callback):
"""Upload checkpoints to wandb as an artifact, at the end of run."""
def __init__(self, ckpt_dir: str = "checkpoints/", upload_best_only: bool = False):
self.ckpt_dir = ckpt_dir
self.upload_best_only = upload_best_only
@rank_zero_only
def on_keyboard_interrupt(self, trainer, pl_module):
self.on_train_end(trainer, pl_module)
@rank_zero_only
def on_train_end(self, trainer, pl_module):
logger = get_wandb_logger(trainer=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 path in Path(self.ckpt_dir).rglob("*.ckpt"):
ckpts.add_file(str(path))
experiment.log_artifact(ckpts)
class LogConfusionMatrix(Callback):
"""Generate confusion matrix every epoch and send it to wandb.
Expects validation step to return predictions and targets.
"""
def __init__(self):
self.preds = []
self.targets = []
self.ready = True
def on_sanity_check_start(self, trainer, pl_module) -> None:
self.ready = False
def on_sanity_check_end(self, trainer, pl_module):
"""Start executing this callback only after all validation sanity checks end."""
self.ready = True
def on_validation_batch_end(
self, trainer, pl_module, outputs, batch, batch_idx, dataloader_idx
):
"""Gather data from single batch."""
if self.ready:
self.preds.append(outputs["preds"])
self.targets.append(outputs["targets"])
def on_validation_epoch_end(self, trainer, pl_module):
"""Generate confusion matrix."""
if self.ready:
logger = get_wandb_logger(trainer)
experiment = logger.experiment
preds = torch.cat(self.preds).cpu().numpy()
targets = torch.cat(self.targets).cpu().numpy()
confusion_matrix = metrics.confusion_matrix(y_true=targets, y_pred=preds)
# set figure size
plt.figure(figsize=(14, 8))
# set labels size
sn.set(font_scale=1.4)
# set font size
sn.heatmap(confusion_matrix, annot=True, annot_kws={"size": 8}, fmt="g")
# names should be uniqe or else charts from different experiments in wandb will overlap
experiment.log({f"confusion_matrix/{experiment.name}": wandb.Image(plt)}, commit=False)
# according to wandb docs this should also work but it crashes
# experiment.log(f{"confusion_matrix/{experiment.name}": plt})
# reset plot
plt.clf()
self.preds.clear()
self.targets.clear()
class LogF1PrecRecHeatmap(Callback):
"""Generate f1, precision, recall heatmap every epoch and send it to wandb.
Expects validation step to return predictions and targets.
"""
def __init__(self, class_names: List[str] = None):
self.preds = []
self.targets = []
self.ready = True
def on_sanity_check_start(self, trainer, pl_module):
self.ready = False
def on_sanity_check_end(self, trainer, pl_module):
"""Start executing this callback only after all validation sanity checks end."""
self.ready = True
def on_validation_batch_end(
self, trainer, pl_module, outputs, batch, batch_idx, dataloader_idx
):
"""Gather data from single batch."""
if self.ready:
self.preds.append(outputs["preds"])
self.targets.append(outputs["targets"])
def on_validation_epoch_end(self, trainer, pl_module):
"""Generate f1, precision and recall heatmap."""
if self.ready:
logger = get_wandb_logger(trainer=trainer)
experiment = logger.experiment
preds = torch.cat(self.preds).cpu().numpy()
targets = torch.cat(self.targets).cpu().numpy()
f1 = f1_score(targets, preds, average=None)
r = recall_score(targets, preds, average=None)
p = precision_score(targets, preds, average=None)
data = [f1, p, r]
# set figure size
plt.figure(figsize=(14, 3))
# set labels size
sn.set(font_scale=1.2)
# set font size
sn.heatmap(
data,
annot=True,
annot_kws={"size": 10},
fmt=".3f",
yticklabels=["F1", "Precision", "Recall"],
)
# names should be uniqe or else charts from different experiments in wandb will overlap
experiment.log({f"f1_p_r_heatmap/{experiment.name}": wandb.Image(plt)}, commit=False)
# reset plot
plt.clf()
self.preds.clear()
self.targets.clear()
class LogImagePredictions(Callback):
"""Logs a validation batch and their predictions to wandb.
Example adapted from:
https://wandb.ai/wandb/wandb-lightning/reports/Image-Classification-using-PyTorch-Lightning--VmlldzoyODk1NzY
"""
def __init__(self, num_samples: int = 8):
super().__init__()
self.num_samples = num_samples
self.ready = True
def on_sanity_check_start(self, trainer, pl_module):
self.ready = False
def on_sanity_check_end(self, trainer, pl_module):
"""Start executing this callback only after all validation sanity checks end."""
self.ready = True
def on_validation_epoch_end(self, trainer, pl_module):
if self.ready:
logger = get_wandb_logger(trainer=trainer)
experiment = logger.experiment
# get a validation batch from the validation dat loader
val_samples = next(iter(trainer.datamodule.val_dataloader()))
val_imgs, val_labels = val_samples
# run the batch through the network
val_imgs = val_imgs.to(device=pl_module.device)
logits = pl_module(val_imgs)
preds = torch.argmax(logits, dim=-1)
# log the images as wandb Image
experiment.log(
{
f"Images/{experiment.name}": [
wandb.Image(x, caption=f"Pred:{pred}, Label:{y}")
for x, pred, y in zip(
val_imgs[: self.num_samples],
preds[: self.num_samples],
val_labels[: self.num_samples],
)
]
}
)
|