File size: 11,632 Bytes
9c6594c |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 |
"""W&B Integration for Metaflow.
This integration lets users apply decorators to Metaflow flows and steps to automatically log parameters and artifacts to W&B by type dispatch.
- Decorating a step will enable or disable logging for certain types within that step
- Decorating the flow is equivalent to decorating all steps with a default
- Decorating a step after decorating the flow will overwrite the flow decoration
Examples can be found at wandb/wandb/functional_tests/metaflow
"""
import inspect
import pickle
from functools import wraps
from pathlib import Path
from typing import Union
import wandb
from wandb.sdk.lib import telemetry as wb_telemetry
try:
from metaflow import current
except ImportError as e:
raise Exception(
"Error: `metaflow` not installed >> This integration requires metaflow! To fix, please `pip install -Uqq metaflow`"
) from e
try:
from plum import dispatch
except ImportError as e:
raise Exception(
"Error: `plum-dispatch` not installed >> "
"This integration requires plum-dispatch! To fix, please `pip install -Uqq plum-dispatch`"
) from e
try:
import pandas as pd
@dispatch
def _wandb_use(
name: str,
data: pd.DataFrame,
datasets=False,
run=None,
testing=False,
*args,
**kwargs,
): # type: ignore
if testing:
return "datasets" if datasets else None
if datasets:
run.use_artifact(f"{name}:latest")
wandb.termlog(f"Using artifact: {name} ({type(data)})")
@dispatch
def wandb_track(
name: str,
data: pd.DataFrame,
datasets=False,
run=None,
testing=False,
*args,
**kwargs,
):
if testing:
return "pd.DataFrame" if datasets else None
if datasets:
artifact = wandb.Artifact(name, type="dataset")
with artifact.new_file(f"{name}.parquet", "wb") as f:
data.to_parquet(f, engine="pyarrow")
run.log_artifact(artifact)
wandb.termlog(f"Logging artifact: {name} ({type(data)})")
except ImportError:
wandb.termwarn(
"`pandas` not installed >> @wandb_log(datasets=True) may not auto log your dataset!"
)
try:
import torch
import torch.nn as nn
@dispatch
def _wandb_use(
name: str,
data: nn.Module,
models=False,
run=None,
testing=False,
*args,
**kwargs,
): # type: ignore
if testing:
return "models" if models else None
if models:
run.use_artifact(f"{name}:latest")
wandb.termlog(f"Using artifact: {name} ({type(data)})")
@dispatch
def wandb_track(
name: str,
data: nn.Module,
models=False,
run=None,
testing=False,
*args,
**kwargs,
):
if testing:
return "nn.Module" if models else None
if models:
artifact = wandb.Artifact(name, type="model")
with artifact.new_file(f"{name}.pkl", "wb") as f:
torch.save(data, f)
run.log_artifact(artifact)
wandb.termlog(f"Logging artifact: {name} ({type(data)})")
except ImportError:
wandb.termwarn(
"`pytorch` not installed >> @wandb_log(models=True) may not auto log your model!"
)
try:
from sklearn.base import BaseEstimator
@dispatch
def _wandb_use(
name: str,
data: BaseEstimator,
models=False,
run=None,
testing=False,
*args,
**kwargs,
): # type: ignore
if testing:
return "models" if models else None
if models:
run.use_artifact(f"{name}:latest")
wandb.termlog(f"Using artifact: {name} ({type(data)})")
@dispatch
def wandb_track(
name: str,
data: BaseEstimator,
models=False,
run=None,
testing=False,
*args,
**kwargs,
):
if testing:
return "BaseEstimator" if models else None
if models:
artifact = wandb.Artifact(name, type="model")
with artifact.new_file(f"{name}.pkl", "wb") as f:
pickle.dump(data, f)
run.log_artifact(artifact)
wandb.termlog(f"Logging artifact: {name} ({type(data)})")
except ImportError:
wandb.termwarn(
"`sklearn` not installed >> @wandb_log(models=True) may not auto log your model!"
)
class ArtifactProxy:
def __init__(self, flow):
# do this to avoid recursion problem with __setattr__
self.__dict__.update(
{
"flow": flow,
"inputs": {},
"outputs": {},
"base": set(dir(flow)),
"params": {p: getattr(flow, p) for p in current.parameter_names},
}
)
def __setattr__(self, key, val):
self.outputs[key] = val
return setattr(self.flow, key, val)
def __getattr__(self, key):
if key not in self.base and key not in self.outputs:
self.inputs[key] = getattr(self.flow, key)
return getattr(self.flow, key)
@dispatch
def wandb_track(
name: str,
data: Union[dict, list, set, str, int, float, bool],
run=None,
testing=False,
*args,
**kwargs,
): # type: ignore
if testing:
return "scalar"
run.log({name: data})
@dispatch
def wandb_track(
name: str, data: Path, datasets=False, run=None, testing=False, *args, **kwargs
):
if testing:
return "Path" if datasets else None
if datasets:
artifact = wandb.Artifact(name, type="dataset")
if data.is_dir():
artifact.add_dir(data)
elif data.is_file():
artifact.add_file(data)
run.log_artifact(artifact)
wandb.termlog(f"Logging artifact: {name} ({type(data)})")
# this is the base case
@dispatch
def wandb_track(
name: str, data, others=False, run=None, testing=False, *args, **kwargs
):
if testing:
return "generic" if others else None
if others:
artifact = wandb.Artifact(name, type="other")
with artifact.new_file(f"{name}.pkl", "wb") as f:
pickle.dump(data, f)
run.log_artifact(artifact)
wandb.termlog(f"Logging artifact: {name} ({type(data)})")
@dispatch
def wandb_use(name: str, data, *args, **kwargs):
try:
return _wandb_use(name, data, *args, **kwargs)
except wandb.CommError:
wandb.termwarn(
f"This artifact ({name}, {type(data)}) does not exist in the wandb datastore!"
f"If you created an instance inline (e.g. sklearn.ensemble.RandomForestClassifier), then you can safely ignore this"
f"Otherwise you may want to check your internet connection!"
)
@dispatch
def wandb_use(
name: str, data: Union[dict, list, set, str, int, float, bool], *args, **kwargs
): # type: ignore
pass # do nothing for these types
@dispatch
def _wandb_use(
name: str, data: Path, datasets=False, run=None, testing=False, *args, **kwargs
): # type: ignore
if testing:
return "datasets" if datasets else None
if datasets:
run.use_artifact(f"{name}:latest")
wandb.termlog(f"Using artifact: {name} ({type(data)})")
@dispatch
def _wandb_use(name: str, data, others=False, run=None, testing=False, *args, **kwargs): # type: ignore
if testing:
return "others" if others else None
if others:
run.use_artifact(f"{name}:latest")
wandb.termlog(f"Using artifact: {name} ({type(data)})")
def coalesce(*arg):
return next((a for a in arg if a is not None), None)
def wandb_log(
func=None,
# /, # py38 only
datasets=False,
models=False,
others=False,
settings=None,
):
"""Automatically log parameters and artifacts to W&B by type dispatch.
This decorator can be applied to a flow, step, or both.
- Decorating a step will enable or disable logging for certain types within that step
- Decorating the flow is equivalent to decorating all steps with a default
- Decorating a step after decorating the flow will overwrite the flow decoration
Args:
func: (`Callable`). The method or class being decorated (if decorating a step or flow respectively).
datasets: (`bool`). If `True`, log datasets. Datasets can be a `pd.DataFrame` or `pathlib.Path`. The default value is `False`, so datasets are not logged.
models: (`bool`). If `True`, log models. Models can be a `nn.Module` or `sklearn.base.BaseEstimator`. The default value is `False`, so models are not logged.
others: (`bool`). If `True`, log anything pickle-able. The default value is `False`, so files are not logged.
settings: (`wandb.sdk.wandb_settings.Settings`). Custom settings passed to `wandb.init`. The default value is `None`, and is the same as passing `wandb.Settings()`. If `settings.run_group` is `None`, it will be set to `{flow_name}/{run_id}. If `settings.run_job_type` is `None`, it will be set to `{run_job_type}/{step_name}`
"""
@wraps(func)
def decorator(func):
# If you decorate a class, apply the decoration to all methods in that class
if inspect.isclass(func):
cls = func
for attr in cls.__dict__:
if callable(getattr(cls, attr)):
if not hasattr(attr, "_base_func"):
setattr(cls, attr, decorator(getattr(cls, attr)))
return cls
# prefer the earliest decoration (i.e. method decoration overrides class decoration)
if hasattr(func, "_base_func"):
return func
@wraps(func)
def wrapper(self, *args, settings=settings, **kwargs):
if not isinstance(settings, wandb.sdk.wandb_settings.Settings):
settings = wandb.Settings()
settings.update_from_dict(
{
"run_group": coalesce(
settings.run_group, f"{current.flow_name}/{current.run_id}"
),
"run_job_type": coalesce(settings.run_job_type, current.step_name),
}
)
with wandb.init(settings=settings) as run:
with wb_telemetry.context(run=run) as tel:
tel.feature.metaflow = True
proxy = ArtifactProxy(self)
run.config.update(proxy.params)
func(proxy, *args, **kwargs)
for name, data in proxy.inputs.items():
wandb_use(
name,
data,
datasets=datasets,
models=models,
others=others,
run=run,
)
for name, data in proxy.outputs.items():
wandb_track(
name,
data,
datasets=datasets,
models=models,
others=others,
run=run,
)
wrapper._base_func = func
# Add for testing visibility
wrapper._kwargs = {
"datasets": datasets,
"models": models,
"others": others,
"settings": settings,
}
return wrapper
if func is None:
return decorator
else:
return decorator(func)
|