A library for users to write (experiment in research) configurations in Python Dict or JSON format, read and write parameter value via dot . in code, while can read parameters from the command line to modify values. 一个供用户以Python Dict或JSON格式编写（科研中实验）配置的库，在代码中用点.读写属性，同时可以从命令行中读取参数配置并修改参数值。
A library for users to write (experiment in research) configurations in Python Dict or JSON format, read and write parameter value via dot . in code, while can read parameters from the command line to modify values.
标签 Labels： Python, Command Line, commandline, config, configuration, parameters, 命令行，配置，传参，参数值修改。
Github URL: https://github.com/NaiboWang/CommandlineConfig
The following fields are reserved and cannot be used as parameter names: config_name.
-h# Install via pip
pip3 install commandline_config
# import package
from commandline_config import Config
# Define configuration dictionary
config = {
"index":1,
"lr": 0.1,
"dbinfo":{
"username": "NUS"
}
}
# Generate configuration class based on configuration dict
c = Config(config)
# Print the configuration of the parameters
print(c)
# Read and write parameters directly via dot . and support multiple layers.
c.index = 2
c.dbinfo.username = "ZJU"
print(c.index, c.dbinfo.username, c["lr"])
# On the command line, modify the parameter values with --
python example.py --index 3 --dbinfo.username XDU
# Get the parameter descriptions via the help method in the code, or on the command line via -h or -help (customization required, see detailed documentation below for details)
c.help()
python example.py -h
If you encounter any problems during using with this tool, please raise an issue in the github page of this project, I will solve the bugs and problems encountered at the first time.
Meanwhile, welcome to submit issues to propose what functions you want to add to this tool and I will implement them when possible.
There are two ways to install this library:
pip3 install commandline_config
If already installed, you can upgrade it by the following command:
pip3 install commandline_config --upgrade
/commandline_config folder of the github project into your own project directory, you need to install the dependency package prettytable:pip3 install prettytable
Or install via requirements.txt:
pip3 install -r requirements.txt
from commandline_config import Config
# comment. Currently supports nesting a dict inside another dict, and can nest unlimited layers.preset_config = {
"index": 1, # Index of party
"dataset": "mnist",
'lr': 0.01, # learning rate
'normalization': True,
"pair": (1,2),
"multi_information": [1, 0.5, 'test', "TEST"], # list
"dbinfo": {
"username": "NUS",
"password": 123456,
"retry_interval_time": 5.5,
"save_password": False,
"pair": ("test",3),
"multi":{
"test":0.01,
},
"certificate_info": ["1", 2, [3.5]],
}
}
That is, the initial configuration of the program is generated. Each key defined in preset_config dict is the parameter name and each value is the initial value of the parameter, and at the same time, the initial value type of the parameter is automatically detected according to the type of the set value.
The above configuration contains seven parameters: index, dataset, batch, normalization, pair, multi_information and dbinfo, where the type of the parameter index is automatically detected as int, the default value is 1 and the description is “Index of party”.
Similarly, The type and default value of the second to fifth parameter are string: "mnist"; float:0.01; bool:True; tuple:(1,2); list:[1,0.5,'test', "TEST"].
The seventh parameter is a nested dictionary of type dict, which also contains 7 parameters, with the same type and default values as the first 7 parameters, and will not be repeated here.
preset_config dict to Config in any function you want.if __name__ == '__main__':
config = Config(preset_config)
# Or give the configuration a name:
config_with_name = Config(preset_config, name="Federated Learning Experiments")
# Or you can store the preset_config in local file configuration.json and pass the filename to the Config class.
config_from_file = Config("configuration.json")
This means that the configuration object is successfully generated.
print function:print(config_with_name)
The output results are:
Configurations of Federated Learning Experiments:
+-------------------+-------+--------------------------+
| Key | Type | Value |
+-------------------+-------+--------------------------+
| index | int | 1 |
| dataset | str | mnist |
| lr | float | 0.01 |
| normalization | bool | True |
| pair | tuple | (1, 2) |
| multi_information | list | [1, 0.5, 'test', 'TEST'] |
| dbinfo | dict | See sub table below |
+-------------------+-------+--------------------------+
Configurations of dict dbinfo:
+---------------------+-------+---------------------+
| Key | Type | Value |
+---------------------+-------+---------------------+
| username | str | NUS |
| password | int | 123456 |
| retry_interval_time | float | 5.5 |
| save_password | bool | False |
| pair | tuple | ('test', 3) |
| multi | dict | See sub table below |
| certificate_info | list | ['1', 2, [3.5]] |
+---------------------+-------+---------------------+
Configurations of dict multi:
+------+-------+-------+
| Key | Type | Value |
+------+-------+-------+
| test | float | 0.01 |
+------+-------+-------+
Here the information of all parameters will be printed in table format. If you want to change the printing style, you can modify it by config_with_name.set_print_style(style=''). The values that can be taken for style are: both, table, json which means print both table and json at the same time, print only table, and json dictionary only.
E.g.:
# Only print json
config_with_name.set_print_style('json')
print(config_with_name)
print("----------")
# Print table and json at the same time
config_with_name.set_print_style('table')
print(config_with_name)
The output results are:
Configurations of Federated Learning Experiments:
{'index': 1, 'dataset': 'mnist', 'lr': 0.01, 'normalization': True, 'pair': (1, 2), 'multi_information': [1, 0.5, 'test', 'TEST'], 'dbinfo': 'See below'}
Configurations of dict dbinfo:
{'username': 'NUS', 'password': 123456, 'retry_interval_time': 5.5, 'save_password': False, 'pair': ('test', 3), 'multi': 'See below', 'certificate_info': ['1', 2, [3.5]]}
Configurations of dict multi:
{'test': 0.01}
----------
Configurations of Federated Learning Experiments:
+-------------------+-------+--------------------------+
| Key | Type | Value |
+-------------------+-------+--------------------------+
| index | int | 1 |
| dataset | str | mnist |
| lr | float | 0.01 |
| normalization | bool | True |
| pair | tuple | (1, 2) |
| multi_information | list | [1, 0.5, 'test', 'TEST'] |
| dbinfo | dict | See sub table below |
+-------------------+-------+--------------------------+
{'index': 1, 'dataset': 'mnist', 'lr': 0.01, 'normalization': True, 'pair': (1, 2), 'multi_information': [1, 0.5, 'test', 'TEST'], 'dbinfo': 'See below'}
Configurations of dict dbinfo:
+---------------------+-------+---------------------+
| Key | Type | Value |
+---------------------+-------+---------------------+
| username | str | NUS |
| password | int | 123456 |
| retry_interval_time | float | 5.5 |
| save_password | bool | False |
| pair | tuple | ('test', 3) |
| multi | dict | See sub table below |
| certificate_info | list | ['1', 2, [3.5]] |
+---------------------+-------+---------------------+
{'username': 'NUS', 'password': 123456, 'retry_interval_time': 5.5, 'save_password': False, 'pair': ('test', 3), 'multi': 'See below', 'certificate_info': ['1', 2, [3.5]]}
Configurations of dict multi:
+------+-------+-------+
| Key | Type | Value |
+------+-------+-------+
| test | float | 0.01 |
+------+-------+-------+
{'test': 0.01}
Configuration parameter values can be written in three ways.
--index 1 on the command line to modify the value of index to 1. Also, the considerations for passing values to different types of arguments are：0 or False for False, 1 or True or no value after the parameter for True: --normalization 1 or --normalization True or --normalization all can set the value of parameter normalization in the configuration to True.--nested-parameter-name.sub-parameter-name.sub-parameter-name.….sub-parameter-name value to modify the value in the nested object, such as --dbinfo.password 987654 to change the value of the password parameter in the dbinfo subobject to 987654; --dbinfo.multi.test 1 to change the value of the test parameter in the multi dict which is in dbinfo subobject to ```. Currently this tool can supports unlimited layers/levels of nesting.preset_config object defined above:python test.py --dbinfo.password 987654 --dbinfo.multi.test 1 --index 0 --dataset emnist --normalization 0 --multi_information [\'sdf\',1,\"3.3\",,True,[1,[]]]
config.index = 2 directly in the code to change the value of the parameter index to 2. Again, list type parameters can be assigned as empty or multidimensional arrays. For nested objects, you can use config.dbinfo.save_password=True to modify the value of the save_password parameter in sub dict dbinfo to True.preset_config does not match, the program will report an error, therefore, if you do not want to force type checking, you can use config["index"] = "sdf" to force the value of the parameter index to the string sdf (not recommended, it will cause unexpected impact).Read the value of the parameter dataset directly by means of config.dataset or config["dataset"].
print(config.dataset, config["index"])
The value of an argument a will be read by this order: the last value modified by config.a = * > the value of --a 2 specified by the command line > the initial value specified by "a":1 defined by preset_config.
For the list type, if a multidimensional array is passed, the information can be read via standard slice of python:
config.dbinfo.certificate_info = [1,[],[[2]]]
print(config.dbinfo.certificate_info[2][0][0])
For parameters in a single nested object, there are four ways to read the values of the parameters, all of which can be read successfully:
print(config.dbinfo.username)
print(config["dbinfo"].password)
print(config.dbinfo["retry_interval_time"])
print(config["dbinfo"]["save_password"])
Simply pass the above config object as a parameter to the function and call it:
def print_dataset_name(c):
print(c.dataset, c["dataset"], c.dbinfo.certificate_info)
print_dataset_name(c=config)
A deep copy of the configuration object can be made by the deepcopy method:
from copy import deepcopy
copy_config = deepcopy(config)
# Modify new configuration's parameter value, will not affect the orignal configuration
copy_config.index=15
The entire parameter configuration can be stored to a local file or uploaded to a remote server such as mongodb, simply by config.save() storing the configuration as a config name (or config if there is no name).json file in the directory, or you can specify the file name and path as follows:
config.save("config/test_config.json")
Then we successfully save the configuration to the local configuration.json file inside the config folder. The file content is as follows:
{
"index": 1,
"dataset": "mnist",
"lr": 0.01,
"normalization": true,
"pair": [1, 2],
"multi_information": [1, 0.5, "test", "TEST"],
"dbinfo": {
"username": "NUS",
"password": 123456,
"retry_interval_time": 5.5,
"save_password": false,
"pair": ["test", 3],
"multi": { "test": 0.01 },
"certificate_info": ["1", 2, [3.5]]
}
}
To store it into the database such as mongodb, you need to get the json sequence first corresponding to the parameters with the info = config.get_config() command, and serialize it with the json library.
For example, to store the config_with_name configuration to mongodb:
import pymongo
myclient = pymongo.MongoClient('mongodb://username:example.com:27017/', connect=False)
mydb = myclient['exps']
table = mydb["table"]
# Get the configurations
configuration = config.get_config()
# Insert configuration dict into mongodb table
table.insert_one(configuration)
# Or make configuration as part of a bigger dict
all_info = {
"exp_time":"20220925",
"configuration":configuration
}
table.insert_one(all_info)
Note that tuples are not supported by JSON, so whether stored locally or in a database, tuple arguments will be converted to lists.
Set advanced options, such as enumerating Enum types, by passing the options parameter of the Config argument to the Config class.
option={}
config = Config(preset_config, options=option)
If you want to limit the value of a parameter to a certain range, you can do so by configuring:
advanced_options = {
'lr': {
"enum": [0.001, 15.5, 0.01, 0.1] # restrict the lr value to one of 0.001, 15.5, 0.01, 0.1
},
'index': {
"enum": [1, 2, 3] # Restrict the index value to 1, 2 and 3
},
"dbinfo": {
"username": {
"enum": ["XDU", "ZJU", "NUS"] # restrict the dbinfo.username field to XDU, ZJU and NUS
},
"multi":{
"test":{
"enum": [1,0.1, 0.01, 15] # 3 layers nested
}
}
},
}
config = Config(preset_config, options=advanced_options)
If enum is set, the following three ways to set a parameter to a value other than the qualified/speficied value will all report an error.
index is set to a value other than 1,2,3 in preset_config:preset_config = {
"index":4,
}
lr argumentpython example.py --lr 0.02
dbinfo.username to a value other than XDU, ZJU and NUS.config.dbinfo.username = "UEST"
The output are:
AttributeError: Can not set value 4 because the key 'index' has set enum list and you the value 4 is not in the enum list [1, 2, 3]!
AttributeError: Can not set value 0.02 because the key 'lr' has set enum list and you the value 0.02 is not in the enum list [0.001, 15.5, 0.01, 0.1]!
AttributeError: Can not set value nus because the key 'username' has set enum list and you the value nus is not in the enum list ['XDU', 'ZJU', 'NUS']!
Set the parameter description helpers by specifying the helpers parameter in the Config class.
helpers = {
"index": "index of information",
"dbinfo_help": "information dict for database",
"dbinfo": {
"username": "username for database",
"multi":{
"test":"test information"
}
}
}
config = Config(preset_config, helpers=helpers)
Note that since the dbinfo parameter is a dict, if you want to set the parameter description for dbinfo, you need to set a dbinfo_help parameter to write the description in the helpers dictionary, i.e. add _help after the dict parameter name to set the parameter description for the dict field.
Two ways to print parameter descriptions, by passing -h or -help on the command line, or by calling the help() function in code.
config_with_name.help()
or
python example.py -h
# OR
python example.py -help
Note that it is only one short slash - and no other command line arguments be added to get help instructions, and the output of both methods is:
Parameter helps for Federated Learning Experiments:
+-------------------+-------+-------------------------------+
| Key | Type | Comments |
+-------------------+-------+-------------------------------+
| index | int | index of information |
| dataset | str | - |
| lr | float | - |
| normalization | bool | - |
| pair | tuple | - |
| multi_information | list | - |
| dbinfo | dict | information dict for database |
+-------------------+-------+-------------------------------+
Parameter helps for dict dbinfo:
+---------------------+-------+-----------------------+
| Key | Type | Comments |
+---------------------+-------+-----------------------+
| username | str | username for database |
| password | int | - |
| retry_interval_time | float | - |
| save_password | bool | - |
| pair | tuple | - |
| multi | dict | Multiple Parameters |
| certificate_info | list | - |
+---------------------+-------+-----------------------+
Parameter helps for dict multi:
+------+-------+------------------+
| Key | Type | Comments |
+------+-------+------------------+
| test | float | test information |
+------+-------+------------------+
This library cannot read command line arguments at the same time with the argparse library, so please do not use args = parser.parse_args() to read command line arguments while using this library.
The type of the parameter will be automatically detected as the same type of the initial value set in preset_config, and the value of the command line parameter will be forced converted to the corresponding type value, such as the default value of index in the above preset_config dict is 1, then the type of the parameter index is int with the initial value of 1. If you specify --index 15.5 on the command line, the parameter index will be automatically assigned to value 15, that is, 15.5 will be automatically forced converted to int type.
If the parameter value specified on the command line parameters can not be forcedly converted to specific type, it will report an error, such as if the command line specified --index sdf, as sdf with orignal format of string can not be converted to int type, so it will report an error.
When the command line argument is set to the input list type, if the element in the list is a string, you must use add a backslash \ before each single/double quote to parse it correctly, otherwise the argument value will be treated as an int or float type. If there are spaces in the command line they will be merged automatically (but the command line environment can not be zsh, if it is zsh environment then must remove all the spaces inside the list, bash and sh does not have this problem, that is, in the zsh environment, you cannot add any space(s) between 15 and \'12\' in --a [15,\'12\']).
If the parameters can be set as follows:
python test.py --array [1,2.3,\'sdf\']
That can correctly resolve the array parameter whose value is a list, and the content of [1,2.3,'sdf', "qwe"], that is, a list containing int, float, string type of data simultaneously.
When the command line parameter is set to the input tuple type, the specified tuple type value must be enclosed in quotes; and if the element in the tuple is a string, a backslash must be added before each single/double quote \ for proper parsing, otherwise the parameter value will be treated as an int or float type. Similarly, if there are spaces in the command line they will be merged automatically (but the command line environment cannot be zsh, if it is zsh environment then all internal spaces must be removed, bash and sh do not have this problem).
For example, the parameter can be set to
python test.py --pair "(1,2,\'msg\')"
The value of the pair parameter is a tuple of type (1,2, "msg"), i.e. a tuple of type int, float, string.
If the parameter name contains special characters such as -+. or space or other python reserved characters, you must use the middle bracket [] to read and write the parameter value instead of . E.g., if the parameter name is multi-information, it can only be accessed by config["multi-information"], cannot do config.multi-information, because the minus sign - is a python language’s reserved symbol.
Now the tool can support unlimited layers of nesting, other supported parameter types are: int, float, string, bool, tuple and list.
The name of the parameter passed on the command line must be defined in preset_config in advance, otherwise an error will be reported, e.g.
python test.py --arg1 1
Since the parameter name arg1 is not defined in preset_config dict, an error is reported indicating that the arg1 parameter is not defined. This function is set to perform parameter integrity checking to avoid entering incorrect parameter names through the command line.
If zsh: no matches found occurs when passing list arguments in the zsh Shell environment, please add a line setopt no_nomatch at the end of the ~/.zshrc file, after save it then run source ~/.zshrc on the command line to refresh zsh, then the problem will be solved.
An example will be given below to demonstrate the convenience of this tool compared to the argparse tool.
The code that needs to be written using the argparse tool:
parser = argparse.ArgumentParser(description='PyTorch local error training')
parser.add_argument('--model', default='vgg8b',
help='model, mlp, vgg13, vgg16, vgg19, vgg8b, vgg11b, resnet18, resnet34, wresnet28-10 and more (default: vgg8b)')
parser.add_argument('--dataset', default='CIFAR10',
help='dataset, MNIST, KuzushijiMNIST, FashionMNIST, CIFAR10, CIFAR100, SVHN, STL10 or ImageNet (default: CIFAR10)')
parser.add_argument('--batch-size', type=int, default=128,
help='input batch size for training (default: 128)')
parser.add_argument('--num-layers', type=int, default=1,
help='number of hidden fully-connected layers for mlp and vgg models (default: 1')
parser.add_argument('--lr', type=float, default=5e-4,
help='initial learning rate (default: 5e-4)')
parser.add_argument('--lr-decay-milestones', nargs='+', type=int, default=[200,300,350,375],
help='decay learning rate at these milestone epochs (default: [200,300,350,375])')
parser.add_argument('--optim', default='adam',
help='optimizer, adam, amsgrad or sgd (default: adam)')
parser.add_argument('--beta', type=float, default=0.99,
help='fraction of similarity matching loss in predsim loss (default: 0.99)')
args = parser.parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
if args.cuda:
cudnn.enabled = True
cudnn.benchmark = True
Code to be written after conversion with this tool:
'''
:param model: model, mlp, vgg13, vgg16, vgg19, vgg8b, vgg11b, resnet18, resnet34, wresnet28-10 and more (default: vgg8b)
:param dataset: dataset, MNIST, KuzushijiMNIST, FashionMNIST, CIFAR10, CIFAR100, SVHN, STL10 or ImageNet (default: CIFAR10)
:param batch-size: input batch size for training (default: 128)
:param num-layers: number of hidden fully-connected layers for mlp and vgg models (default: 1)
:param lr: initial learning rate (default: 5e-4)
:param lr-decay-milestones: decay learning rate at these milestone epochs (default: [200,300,350,375])
:param optim: optimizer, adam, amsgrad or sgd (default: adam)
:param beta: fraction of similarity matching loss in predsim loss (default: 0.99)
'''
config = {
'model':'vgg8b',
'dataset':'CIFAR10',
'batch-size':128,
'num-layers':1,
'lr':5e-4,
'lr-decay-milestones':[200,300,350,375],
'optim':'adam',
'beta':0.99,
}
args = Config(config, name='PyTorch local error training')
args.cuda = not args.no_cuda and torch.cuda.is_available()
if args.cuda:
cudnn.enabled = True
cudnn.benchmark = True
As we can see, the amount of code has been reduced and is more structured and neat.
Another case with library click:
import click
@click.command()
@click.option('--count', default=1, help='Number of greetings.')
@click.option('--name', prompt='Your name',
help='The person to greet.')
def hello(count, name):
"""Simple program that greets NAME for a total of COUNT times."""
for x in range(count):
click.echo(f"Hello {name}!")
if __name__ == '__main__':
hello()
Can be converted to the following code:
from commandline_config import Config
def hello(o):
"""Simple program that greets NAME for a total of COUNT times."""
for x in range(o.count):
print(f"Hello {o.name}!")
if __name__ == '__main__':
args = {
"count":1, # Number of greetings.
"name":"", # The person to greet.
}
options = Config(args)
hello(options)
You can run example.py in the Github project to test the whole tool, most functions’s codes had been provided inside the file:
# Get help for all parameters of example.py
python example.py -h
# Specify parameter values
python example.py --dbinfo.multi.test 0.01 --dbinfo.username NUS
The following describes the author’s personal reasons for developing and the benefits/conveniences of this package.
For us who often run research experiments, do you often need to set a lot of command line arguments at the beginning of a python file and call them in the following code as args.*：
For example, the following example paragraph:
parser = argparse.ArgumentParser(description='index')
parser.add_argument('--index', default=0, type=int, help='party index')
parser.add_argument('--party_num', default=100, type=int)
args = parser.parse_args()
print(args.index)
With one more additional parameter, we need to write one line of parser.add_argument(...), when handwriting configuration of each parameter, it will be very tedious such as the name needs to add --, and modify the default value, type and description of the time is very troublesome, finally will lead to very long code and inconvenient to maintain.
Even if you use a more advanced library click, you still need to write option constantly, and you need to write the same amount of parameters at the input parameter field of the function to match all the option, writing code is really tedious, such as the following Click code:
import click
from caesar_encryption import encrypt
@click.command()
@click.argument('text', nargs=-1)
@click.option('--decrypt/--encrypt', '-d/-e')
@click.option('--key', '-k', default=1)
def caesar(text, decrypt, key):
text_string = ' '.join(text)
if decrypt:
key = -key
cyphertext = encrypt(text_string, key)
click.echo(cyphertext)
if __name__ == '__main__':
caesar()
I don’t want to specify option over and over again when writing code, and write a lot of corresponding parameters in the function name parameter field, which is very troublesome.
So would it make the code look more structured and clearer if the parameter configuration could be written in the format of Python objects/JSON?
Also, is it possible to directly read and write command line arguments as if they were Python objects, or deep copy them, as if no command line arguments had been configured?
Most importantly, can we make the process of writing code faster, instead of adding a little bit of configuration line by line?
In order to solve the above pain points, this tool has been developed.
My goal is to make it easier for us to write more concise codes that works as the same as verbose command line parsing packages like Argparse or Click, while also making the configuration look structured and easy to maintain. Json was chosen because the Python dictionary type Dict is native in JSON format, so can make the same format as Python code. In this way, the configuration information can be directly embedded in the Python code file (anywhere), without the need to establish a separate configuration file such as TOML. At the same time, you can use the library to write the configuration at any position within the code file to read the command line parameters and modify the value of each field at any time.
Similar tools including:
| Name | Advantage | Disadvantage |
|---|---|---|
| Fire | Can direct do conversion of function parameters to command line parameters | Cannot pass parameters to other functions. |
| hydra | Easy to read and write to yaml | Need additional yaml file with specifed path, also cannot perform integrity check and type conversion/check, not friendly for printing |
| ml_collections | Similar functionality to this tool with more configurable items | A little tideous to pass commandline parameters, when nesting need to artificially set specifed Class, and also without integrity check, not friendly for printing |
Welcome to submit issues to propose what functions you want to add to this tool and I will implement them when possible.