|
| 1 | +Quick Start |
| 2 | +=========== |
| 3 | + |
| 4 | +HyperParameter is a configuration and parameter management library for Python. HyperParameter provides the following features: |
| 5 | + |
| 6 | +1. `HyperParameter` class, providing object-style api for accessing dict data and handling undefine parameters with default values; |
| 7 | +2. `param_scope` context manager, maintains a thread-safe `HyperParameter` and manage parameter modifications with nested `param_scope`; |
| 8 | +3. `auto_param` decorator, allowing default parameters of a function or class to be supplied from `param_scope`; |
| 9 | + |
| 10 | +HyperParameter is particularly well suited for machine learning experiments and related systems, which have many parameters and nested codes. |
| 11 | + |
| 12 | +`HyperParameter` |
| 13 | +---------------- |
| 14 | + |
| 15 | +### create hyperparameter |
| 16 | + |
| 17 | +```python |
| 18 | +>>> from hyperparameter import HyperParameter |
| 19 | +>>> hp = HyperParameter(param1=1, param2="A") |
| 20 | +>>> hp |
| 21 | +{'param1': 1, 'param2': 'A'} |
| 22 | + |
| 23 | +``` |
| 24 | + |
| 25 | +### access parameters in hyperparameter |
| 26 | + |
| 27 | +we can access the parameter stored in HyperParameter with object-style API: |
| 28 | + |
| 29 | +```python |
| 30 | +>>> hp.param1 |
| 31 | +1 |
| 32 | + |
| 33 | +>>> hp.param2 |
| 34 | +'A' |
| 35 | + |
| 36 | +``` |
| 37 | + |
| 38 | +nested parameters are also supported: |
| 39 | + |
| 40 | +```python |
| 41 | +>>> hp = HyperParameter(param1=1, obj2={"prop3": 3}) |
| 42 | +>>> hp.obj2.prop3 |
| 43 | +3 |
| 44 | + |
| 45 | +``` |
| 46 | + |
| 47 | +### access undefined parameter |
| 48 | + |
| 49 | +Undefined parameter is treated as false in `if` statement: |
| 50 | + |
| 51 | +```python |
| 52 | +>>> if not hp.undefined_param: print("param not defined") |
| 53 | +param not defined |
| 54 | + |
| 55 | +``` |
| 56 | + |
| 57 | +The hyperparameter object is a `Callable`, and will be turned into `safe-mode` when called. In safe-mode the users must provide a default value when accessing parameters. |
| 58 | + |
| 59 | +```python |
| 60 | +>>> hp().undefined_param("default_value") |
| 61 | +'default_value' |
| 62 | + |
| 63 | +``` |
| 64 | + |
| 65 | +`param_scope` |
| 66 | +------------- |
| 67 | + |
| 68 | +### manage hyper-parameters with `param_scope` |
| 69 | + |
| 70 | +`param_scope` create a `HyperParameter` object for the context it manages: |
| 71 | + |
| 72 | +```python |
| 73 | +>>> from hyperparameter import param_scope |
| 74 | +>>> with param_scope(param1=1, param2="A") as ps: |
| 75 | +... ps.param1 |
| 76 | +1 |
| 77 | + |
| 78 | +``` |
| 79 | + |
| 80 | +### access hyper-parameter in functions |
| 81 | + |
| 82 | +`param_scope` is often used to configure the behavior of a function. |
| 83 | + |
| 84 | +```python |
| 85 | +>>> def foo(): # access parameter |
| 86 | +... ps = param_scope.current() |
| 87 | +... return ps.param1 |
| 88 | + |
| 89 | +>>> def foo_with_default(): # access parameter with default value |
| 90 | +... ps = param_scope.current() |
| 91 | +... return ps().param1(2) # default value is `2` |
| 92 | + |
| 93 | +>>> with param_scope(param1=1, param2="A") as ps: |
| 94 | +... foo() |
| 95 | +1 |
| 96 | + |
| 97 | +>>> with param_scope(param1=1, param2="A") as ps: |
| 98 | +... foo_with_default() |
| 99 | +1 |
| 100 | + |
| 101 | +>>> with param_scope(param2="A") as ps: |
| 102 | +... foo_with_default() |
| 103 | +2 |
| 104 | + |
| 105 | +``` |
| 106 | + |
| 107 | +`auto_param` |
| 108 | +------------ |
| 109 | + |
| 110 | +### Define a function with `auto_param` decorator |
| 111 | + |
| 112 | +`auto_param` provides an easier way to define a configurable function. The configurable parameters and their default values are defined with kwargs if `auto_param` is used. |
| 113 | + |
| 114 | +```python |
| 115 | +>>> from hyperparameter import auto_param |
| 116 | +>>> @auto_param |
| 117 | +... def foo(a, b=1, c=2): |
| 118 | +... return (a, b, c) |
| 119 | + |
| 120 | +>>> foo(0) |
| 121 | +(0, 1, 2) |
| 122 | + |
| 123 | +``` |
| 124 | + |
| 125 | +### config function default parameter with `param_scope` |
| 126 | + |
| 127 | +We can config the default behavior of a function with `param_scope`. Since the `param_scope` API is scoped and thread-safe, we can use it without too much warry. |
| 128 | + |
| 129 | +```python |
| 130 | +>>> with param_scope("foo.b=0"): |
| 131 | +... foo(0) |
| 132 | +(0, 0, 2) |
| 133 | + |
| 134 | +``` |
0 commit comments