An argparse extension for hpman
Project description
hpargparse
An argparse extension for hpman
Installation
python3 -m pip install hpargparse
Brief Introduction
The following example lies in examples/02-brief.
main.py
#!/usr/bin/env python3
from hpman.m import _
import hpargparse
import argparse
def func():
weight_decay = _("weight_decay", 1e-5)
print("weight decay is {}".format(weight_decay))
def main():
parser = argparse.ArgumentParser()
_.parse_file(__file__)
hpargparse.bind(parser, _)
parser.parse_args()
func()
if __name__ == "__main__":
main()
results in:
$ ./main.py
weight decay is 1e-05
$ ./main.py --weight-decay 1e-4
weight decay is 0.0001
$ ./main.py --weight-decay 1e-4 --hp-list
weight_decay: 0.0001
$ ./main.py --weight-decay 1e-4 --hp-list detail
All hyperparameters:
['weight_decay']
Details
╔══════════════╦═══════╦════════╦═════════════════════════════════════════════════════════════════╗
║ name ║ type ║ value ║ details ║
╠══════════════╬═══════╬════════╬═════════════════════════════════════════════════════════════════╣
║ weight_decay ║ float ║ 0.0001 ║ occurrence[0]: ║
║ ║ ║ ║ ./main.py:10 ║
║ ║ ║ ║ 5: ║
║ ║ ║ ║ 6: import argparse ║
║ ║ ║ ║ 7: ║
║ ║ ║ ║ 8: ║
║ ║ ║ ║ 9: def func(): ║
║ ║ ║ ║ ==> 10: weight_decay = _("weight_decay", 1e-5) ║
║ ║ ║ ║ 11: print("weight decay is {}".format(weight_decay)) ║
║ ║ ║ ║ 12: ║
║ ║ ║ ║ 13: ║
║ ║ ║ ║ 14: def main(): ║
║ ║ ║ ║ 15: parser = argparse.ArgumentParser() ║
╚══════════════╩═══════╩════════╩═════════════════════════════════════════════════════════════════╝
$ ./main.py -h
usage: main.py [-h] [--weight-decay WEIGHT_DECAY] [--hp-save HP_SAVE]
[--hp-load HP_LOAD] [--hp-list [{detail,yaml,json}]]
[--hp-detail] [--hp-serial-format {auto,yaml,pickle}]
[--hp-exit]
optional arguments:
-h, --help show this help message and exit
--weight-decay WEIGHT_DECAY
(default: 1e-05)
--hp-save HP_SAVE Save hyperparameters to a file. The hyperparameters
are saved after processing of all other options
(default: None)
--hp-load HP_LOAD Load hyperparameters from a file. The hyperparameters
are loaded before any other options are processed
(default: None)
--hp-list [{detail,yaml,json}]
List all available hyperparameters. If `--hp-list
detail` is specified, a verbose table will be print
(default: None)
--hp-detail Shorthand for --hp-list detail (default: False)
--hp-serial-format {auto,yaml,pickle}
Format of the saved config file. Defaults to auto. It
can be set to override auto file type deduction.
(default: auto)
--hp-exit process all hpargparse actions and quit (default:
False)
hpcli: The Command Line Tool
Besides using hpargparse.bind in you code, we also come with a command line
tool hpcli to provide similar functions to any existing file using hpman.
src.py
from hpman.m import _
_('num_channels', 128)
_('num_layers', 50)
In shell:
$ hpcli src.py
num_channels: 128
num_layers: 50
$ hpcli src.py --num-layers 101
num_channels: 128
num_layers: 101
$ hpcli src.py --num-layers 101 --hp-save config.yaml
num_channels: 128
num_layers: 101
$ hpcli src.py --num-layers 101 --hp-save config.yaml --hp-list detail
All hyperparameters:
['num_channels', 'num_layers']
Details
╔══════════════╦══════╦═══════╦═══════════════════════════════════════╗
║ name ║ type ║ value ║ details ║
╠══════════════╬══════╬═══════╬═══════════════════════════════════════╣
║ num_channels ║ int ║ 128 ║ occurrence[0]: ║
║ ║ ║ ║ src.py:3 ║
║ ║ ║ ║ 1: from hpman.m import _ ║
║ ║ ║ ║ 2: ║
║ ║ ║ ║ ==> 3: _("num_channels", 128) ║
║ ║ ║ ║ 4: _("num_layers", 50) ║
║ ║ ║ ║ 5: ║
╠══════════════╬══════╬═══════╬═══════════════════════════════════════╣
║ num_layers ║ int ║ 101 ║ occurrence[0]: ║
║ ║ ║ ║ src.py:4 ║
║ ║ ║ ║ 1: from hpman.m import _ ║
║ ║ ║ ║ 2: ║
║ ║ ║ ║ 3: _("num_channels", 128) ║
║ ║ ║ ║ ==> 4: _("num_layers", 50) ║
║ ║ ║ ║ 5: ║
╚══════════════╩══════╩═══════╩═══════════════════════════════════════╝
$ hpcli src.py -h
usage: hpcli [-h] [--num-channels NUM_CHANNELS] [--num-layers NUM_LAYERS]
[--hp-save HP_SAVE] [--hp-load HP_LOAD]
[--hp-list [{detail,yaml,json}]] [--hp-detail]
[--hp-serial-format {auto,yaml,pickle}] [--hp-exit]
optional arguments:
-h, --help show this help message and exit
--num-channels NUM_CHANNELS
(default: 128)
--num-layers NUM_LAYERS
(default: 50)
--hp-save HP_SAVE Save hyperparameters to a file. The hyperparameters
are saved after processing of all other options
(default: None)
--hp-load HP_LOAD Load hyperparameters from a file. The hyperparameters
are loaded before any other options are processed
(default: None)
--hp-list [{detail,yaml,json}]
List all available hyperparameters. If `--hp-list
detail` is specified, a verbose table will be print
(default: yaml)
--hp-detail Shorthand for --hp-list detail (default: False)
--hp-serial-format {auto,yaml,pickle}
Format of the saved config file. Defaults to auto. It
can be set to override auto file type deduction.
(default: auto)
--hp-exit process all hpargparse actions and quit (default:
False)
This could be a handy tool to inspect the hyperparameters in your code.
Example: Deep Learning Experiment
This example lies in examples/01-nn-training.
It is a fully functional example of training a LeNet on MNIST using
hpargparse and hpman collaboratively to manage hyperparameters.
We highly suggest you playing around this example.
Example: Basics Walkthrough
Now we break down the functions one-by-one.
The following example lies in examples/00-basic.
lib.py:
from hpman.m import _
def add():
return _("a", 1) + _("b", 2)
def mult():
return _("a") * _("b")
main.py
#!/usr/bin/env python3
import argparse
import hpargparse
from hpman.m import _
import os
BASE_DIR = os.path.dirname(os.path.realpath(__file__))
def main():
parser = argparse.ArgumentParser()
# ... do whatever you want
parser.add_argument(dest="predefined_arg")
# analyze everything in this directory
_.parse_file(BASE_DIR) # <-- IMPORTANT
# bind will monkey_patch parser.parse_args to do its job
hpargparse.bind(parser, _) # <-- IMPORTANT
# parse args and set the values
args = parser.parse_args()
# ... do whatever you want next
import lib
print("a = {}".format(_.get_value("a")))
print("b = {}".format(_.get_value("b")))
print("lib.add() = {}".format(lib.add()))
print("lib.mult() = {}".format(lib.mult()))
if __name__ == "__main__":
main()
Help
$ ./main.py -h
usage: main.py [-h] [--a A] [--b B] [--hp-save HP_SAVE] [--hp-load HP_LOAD]
[--hp-list [{detail,yaml,json}]] [--hp-detail]
[--hp-serial-format {auto,yaml,pickle}] [--hp-exit]
predefined_arg
positional arguments:
predefined_arg
optional arguments:
-h, --help show this help message and exit
--a A (default: 1)
--b B (default: 2)
--hp-save HP_SAVE Save hyperparameters to a file. The hyperparameters
are saved after processing of all other options
(default: None)
--hp-load HP_LOAD Load hyperparameters from a file. The hyperparameters
are loaded before any other options are processed
(default: None)
--hp-list [{detail,yaml,json}]
List all available hyperparameters. If `--hp-list
detail` is specified, a verbose table will be print
(default: None)
--hp-detail Shorthand for --hp-list detail (default: False)
--hp-serial-format {auto,yaml,pickle}
Format of the saved config file. Defaults to auto. It
can be set to override auto file type deduction.
(default: auto)
--hp-exit process all hpargparse actions and quit (default:
False)
Set Hyperparameters from Command Line Arguments
$ ./main.py some_thing --a 3 --b 5
a = 3
b = 5
lib.add() = 8
lib.mult() = 15
List All Hyperparameters
$ ./main.py some_arg --hp-list
a: 1
b: 2
... and details:
$ ./main.py some_arg --hp-list detail
All hyperparameters:
['a', 'b']
Details
╔══════╦══════╦═══════╦═══════════════════════════════════════════════════════════════════════════╗
║ name ║ type ║ value ║ details ║
╠══════╬══════╬═══════╬═══════════════════════════════════════════════════════════════════════════╣
║ a ║ int ║ 1 ║ occurrence[0]: ║
║ ║ ║ ║ /data/project/hpargparse/examples/00-basic/lib.py:8 ║
║ ║ ║ ║ 3: # for more usecases, please refer to hpman's document ║
║ ║ ║ ║ 4: ║
║ ║ ║ ║ 5: ║
║ ║ ║ ║ 6: def add(): ║
║ ║ ║ ║ 7: # define a hyperparameter on-the-fly with defaults ║
║ ║ ║ ║ ==> 8: return _("a", 1) + _("b", 2) ║
║ ║ ║ ║ 9: ║
║ ║ ║ ║ 10: ║
║ ║ ║ ║ 11: def mult(): ║
║ ║ ║ ║ 12: # reuse a pre-defined hyperparameters ║
║ ║ ║ ║ 13: return _("a") * _("b") ║
║ ║ ║ ║ occurrence[1]: ║
║ ║ ║ ║ /data/project/hpargparse/examples/00-basic/lib.py:13 ║
║ ║ ║ ║ 8: return _("a", 1) + _("b", 2) ║
║ ║ ║ ║ 9: ║
║ ║ ║ ║ 10: ║
║ ║ ║ ║ 11: def mult(): ║
║ ║ ║ ║ 12: # reuse a pre-defined hyperparameters ║
║ ║ ║ ║ ==> 13: return _("a") * _("b") ║
║ ║ ║ ║ 14: ║
╠══════╬══════╬═══════╬═══════════════════════════════════════════════════════════════════════════╣
║ b ║ int ║ 2 ║ occurrence[0]: ║
║ ║ ║ ║ /data/project/hpargparse/examples/00-basic/lib.py:8 ║
║ ║ ║ ║ 3: # for more usecases, please refer to hpman's document ║
║ ║ ║ ║ 4: ║
║ ║ ║ ║ 5: ║
║ ║ ║ ║ 6: def add(): ║
║ ║ ║ ║ 7: # define a hyperparameter on-the-fly with defaults ║
║ ║ ║ ║ ==> 8: return _("a", 1) + _("b", 2) ║
║ ║ ║ ║ 9: ║
║ ║ ║ ║ 10: ║
║ ║ ║ ║ 11: def mult(): ║
║ ║ ║ ║ 12: # reuse a pre-defined hyperparameters ║
║ ║ ║ ║ 13: return _("a") * _("b") ║
║ ║ ║ ║ occurrence[1]: ║
║ ║ ║ ║ /data/project/hpargparse/examples/00-basic/lib.py:13 ║
║ ║ ║ ║ 8: return _("a", 1) + _("b", 2) ║
║ ║ ║ ║ 9: ║
║ ║ ║ ║ 10: ║
║ ║ ║ ║ 11: def mult(): ║
║ ║ ║ ║ 12: # reuse a pre-defined hyperparameters ║
║ ║ ║ ║ ==> 13: return _("a") * _("b") ║
║ ║ ║ ║ 14: ║
╚══════╩══════╩═══════╩═══════════════════════════════════════════════════════════════════════════╝
Save/Load from/to YAML file
# save to yaml file
$ ./main.py some_arg --hp-save /tmp/config.yaml --hp-exit
$ cat /tmp/config.yaml
a: 1
b: 2
# load from yaml file
$ cat config_modified.yaml
a: 123
b: 456
$ ./main.py some_arg --hp-load config_modified.yaml --hp-list
a: 123
b: 456
Development
- Install requirements:
pip install -r requirements.dev.txt -r requirements.txt
- Activate git commit template
git config commit.template .git-commit-template.txt
- Install pre-commit hook
pre-commit install
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
hpargparse-0.14.0.tar.gz
(13.4 kB
view details)
Built Distribution
File details
Details for the file hpargparse-0.14.0.tar.gz.
File metadata
- Download URL: hpargparse-0.14.0.tar.gz
- Upload date:
- Size: 13.4 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.1 CPython/3.10.8
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 | d9cc575215d167783e7355be7a0aaccb6eb08e3d4e8bfa2d8bb27624eaba02c3 |
|
| MD5 | 81bb966fd0aea90380dae7bcd08f631e |
|
| BLAKE2b-256 | 9856a45aa7a0b6ce68033ae8e6b32f2ec21820833de31344845e541f4432795f |
File details
Details for the file hpargparse-0.14.0-py3-none-any.whl.
File metadata
- Download URL: hpargparse-0.14.0-py3-none-any.whl
- Upload date:
- Size: 11.3 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.1 CPython/3.10.8
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 | 7f58d5bd7fa9ba67955305bb0b5c831f0d664b73b3ca708ad4b9460993203be7 |
|
| MD5 | 80f89bebda40781a44c245df33ece53c |
|
| BLAKE2b-256 | 783464f0137d71edeb282a859227028bc672cc97ea04db005f831eb177da83ad |
