A wrapper to run and monitor absl app.
Project description
ABSL-Extra
A collection of utils I commonly use for running my experiments. It will:
- Notify on execution start, finish or failed.
- By default, Notifier will just log those out to
stdout
. - I prefer receiving those in Slack, though (see example below).
- By default, Notifier will just log those out to
- Log parsed CLI flags from
absl.flags.FLAGS
and config values fromconfig_file:get_config()
- Inject
ml_collections.ConfigDict
fromconfig_file
, if kwarg provided. - Inject
pymongo.collection.Collection
ifmongo_config
kwarg provided. - Select registered task to run based on --task= CLI argument.
Minimal example
import os
from pymongo.collection import Collection
from ml_collections import ConfigDict
from absl import logging
import tensorflow as tf
from absl_extra import tf_utils, tasks, notifier
@tasks.register_task(
config_file="config.py",
mongo_config=dict(uri=os.environ["MONGO_URI"], db_name="my_project", collection="experiment_1"),
notifier=notifier.SlackNotifier(slack_token=os.environ["SLACK_BOT_TOKEN"], channel_id=os.environ["CHANNEL_ID"])
)
@tf_utils.requires_gpu
def main(config: ConfigDict, db: Collection) -> None:
if tf_utils.supports_mixed_precision():
tf.keras.mixed_precision.set_global_policy("mixed_float16")
with tf_utils.make_gpu_strategy().scope():
logging.info("Doing some heavy lifting...")
if __name__ == "__main__":
tasks.run()
Planned for:
- global app state for different tasks
- list of pre/post hooks
- keras callback with notifier
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
absl_extra-0.0.3.dev3.tar.gz
(13.2 kB
view hashes)
Built Distribution
Close
Hashes for absl_extra-0.0.3.dev3-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | a06326ef2381e882be798003c38e25e3cae6869bf89e18a6a572c0262e790ae7 |
|
MD5 | 23f294e456a7ec5598f7b384220a8260 |
|
BLAKE2b-256 | 53b3a9dd0f34cb8ca0629b10e4e41416eec3336ca7a47167f133b581a126e53b |