hao 0.2

misc tools for configs, logs

hao

configurations, logs and others.

install

pip install hao

precondition

The folder contained any of the following files (searched in this very order) will be treated as project root path.

• requirements.txt
• VERSION
• conf
• setup.py
• .idea
• .git

If your project structure does NOT conform to this, it will not work as expected.

features

config

It will try to load YAML config file from conf folder

.                               # project root
├── conf
│   ├── config-{env}.yml        # if `export env=abc`, will raise error if not found
│   ├── config-{hhostname}.yml  # try to load this file, then the default `config.yml`
│   └── config.yml              # the default config file that should always exist
├── requirements.txt            # every project should have this file
├── VERSION                     # spanner.versions.get_version() will try to read this file
├── .git

In following order:

if os.environ.get("env") is not None:
    try_to_load(f'config-{env}.yml', fallback='config.yml')                   # echo $env
else:
    try_to_load(f'config-{socket.gethostname()}.yml', fallback='config.yml')  # echo hostname

Say you have the following content in your config file:

es:
  default:
    host: 172.23.3.3
    port: 9200
    indices:
      - news
      - papers

The get the configured values in your code:

import hao
es_host = hao.config.get('es.default.host')          # str
es_port = hao.config.get('es.default.port')          # int
indices = hao.config.get('es.default.indices')       # list
...

logs

Set the logger level in config.yml

e.g.

logging:
  __main__: DEBUG
  transformers: WARNING
  lightning: INFO
  pytorch_lightning: INFO
  elasticsearch: WARNING
  tests: DEBUG
  root: INFO                        # root level

If you want to change the log format:

logger:
  format: "%(asctime)s %(levelname)-7s %(name)s:%(lineno)-4d - %(message)s"

Declear your logger

import hao
LOGGER = hao.logs.get_logger(__name__)

namespaces

from hao.namespaces import from_args, attr

@from_args
#@from_args(adds=Trainer.add_argparse_args)
class TrainConf(Namespace):
    root_path_checkpoints = attr(str, default=hao.paths.get_path('data/checkpoints/'))
    dataset_train = attr(str, default='train.txt')
    dataset_val = attr(str, default='val.txt')
    dataset_test = attr(str, default='test.txt')
    batch_size = attr(int, default=128, key='train.batch_size')                          # key means try to load from config.yml by the key
    task = attr(str, choices=('ner', 'nmt'), default='ner')
    seed = attr(int)
    epochs = attr(int, default=5)

Where attr is a wrapper for argpars.add_argument()

Usage 1: overwrite the default value from command line

python -m your_module --task=nmt

Usage 2: overwrite the default value from constructor

train_conf = TrainConf(task='nmt')

Value lookup order:

• command line
• constructor
• config yml if key specified in attr
• default if specified in attr