Skip to main content

A graph flow engine for East Low Carbon

Project description

elcflow

Build Status PyPI version Coverage Status Documentation Status

介绍

ELC使用

指令

生成文档

cd docs && sphinx-apidoc -o source ../elcflow/ && make html

测试

pytest --cov=elcflow

使用[V2]

注册算子

使用register_elc_function_v2来注册是一个算子,算子包含以下属性:

  • name
    • 唯一的标识符,用来找到这个算子
  • parameters
    • 字典类型, 包含了默认的参数
    • 主要适用于例如指数函数,其指数当作参数
      • 给指数2 -> 平方

算子的输入和输出被严格规定 - 输入 - global_states - 全局的一个state是一个字典 - result - 前一个节点的结果 - parameters - 参数 - 输出 - 一定是一个字典

用例

from elcflow.base import register_elc_function_v2

@register_elc_function_v2(name='elc_select_data_v2')
def elc_select_data_v2(global_states, result, parameters):
    return {'return': global_states[parameters['key']]}


@register_elc_function_v2(name='elc_add_plus_plus_v2')
def elc_add_plus_plus_v2(global_states, result, parameters):
    return {'return': result['return'] + 1}


@register_elc_function_v2(name='elc_mul_v2')
def elc_mul_v2(global_states, result, parameters):
    _result = result['return'] * int(global_states['multiplier'])
    global_states['elc_mul_v2_result'] = _result
    return {'return': _result}


@register_elc_function_v2(name='elc_pow_for_mul_v2')
def elc_pow_for_mul_v2(global_states, result, parameters):
    _result = global_states['elc_mul_v2_result'] ** int(parameters['a'])
    global_states['elc_pow_for_mul_v2_result'] = _result
    return {'return': _result}

数据结构

图主要由以下两部分组成

  • nodes
    • 包含了所有的节点的信息
    • id
    • label
    • 类型
      • 数据类型
      • 算子类型
    • 其他
      • 例如算子的参数
  • edges
    • 有方向的边,从一个node的数据流向另一个node
    • id
    • source
      • 起始节点的id
    • target
      • 结束节点的id

例子

from elcflow.graph import *
from elcflow.graph import *
from elcflow.helpers import json_stringify, json_parse

_model_2_globals = {
    "global_variable_1": 5,
    "multiplier": 3
}

_model_dict_v2 = {
    "nodes": [
        {"label": "data-selector", "id": "0ea5a129", "_elc_node_type": 'operator', "_elc_function": 'elc_select_data_v2', "_elc_parameters": {"key": "global_variable_1"}},
        {"label": "add_plus_plus", "id": "8ac87236", "_elc_node_type": 'operator', "_elc_function": "elc_add_plus_plus_v2"},
        {"label": "multiplier x", "id": "0d1af6ff", "_elc_node_type": 'operator', "_elc_function": "elc_mul_v2"},
        {"label": "pow_for_mul", "id": "9d1af6ff", "_elc_node_type": 'operator', "_elc_function": "elc_pow_for_mul_v2", "_elc_parameters": {"a": 4}},
    ],
    "edges": [
        {"source": "0ea5a129", "target": "8ac87236", "id": "74bc97ca"},
        {"source": "8ac87236", "target": "0d1af6ff", "id": "d3645364"},
        {"source": "0d1af6ff", "target": "9d1af6ff", "id": "b0eb9a9b"},
    ]
}
# 从json结构parse成图
_graph = ELCGraph.create_from_elc_json(_model_dict_v2, elc_graph_version=ELC_GRAPH_VERSION_V2)
# 并且初始化global
_graph.set_state(_globals=_model_2_globals)
# 编译图: 包含了对于节点拓扑排序并且初始化的工作
_graph.compile()
# 执行图
_graph.execute()
# 画图
_graph.plot(show=True, with_state=True)

Example Graph 2

使用[V1]

注册算子

使用register_elc_function来注册是一个算子,算子包含以下属性:

  • name
    • 唯一的标识符,用来找到这个算子
  • outputs
    • list
    • 给输出取名字(在图上徐)
    • 对应到是节点输出的id
  • inputs
    • 不可选, 则使用函数中定义的名称
    • 对应到是节点输入的id
  • parameters
    • 字典类型, 包含了默认的参数
    • 主要适用于例如指数函数,其指数当作参数
      • 给指数2 -> 平方

用例

from elcflow.base import register_elc_function
@register_elc_function(outputs=['elc_output'], name='elc_exp_and_plus', parameters={'x': 5})
# 一个例子
# 表明注册了一个名称为elc_exp_and_plus算子
# 他的输入有[a,b,x]
# 其中x是参数给的默认值是5
# 它输出的名字叫做elc_output
def elc_exp_and_plus_(a, b, x=2):
    return a ** x + b ** x

数据结构

图主要由以下两部分组成

  • nodes
    • 包含了所有的节点的信息
    • id
    • label
    • 类型
      • 数据类型
      • 算子类型
    • 其他
      • 例如算子的参数
  • edges
    • 有方向的边,从一个node的某一个输出数据流向另一个node的某一个输入
    • id
    • source
      • 起始节点的id
    • target
      • 结束节点的id
    • _elc_source_output_id
      • 起始节点输出的id
        • 可能有多个输出, 取哪一个?
    • __elc_target_input_id
      • 结束节点输入的id
        • 可能有多个输入, 对应到哪一个?

例子

from elcflow.graph import *
from elcflow.graph import *
from elcflow.helpers import json_stringify, json_parse

json_model = {
    "nodes": [
        {"label": "Input-2", "id": "3362b879", "_elc_node_type": 'data'},
        {"label": "Input-1", "id": "0ea5a129", "_elc_node_type": 'data'},
        {"label": "Add", "id": "8ac87236", "_elc_node_type": 'operator', "_elc_function": "elc_add"},
        {"label": "MUL", "id": "0d1af6ff", "_elc_node_type": 'operator', "_elc_function": "elc_mul"},
        {"label": "POW", "id": "9d1af6ff", "_elc_node_type": 'operator', "_elc_function": "elc_pow", "_elc_parameters": {"a": 4}},
        {"label": "OUTPUT", "id": "1d1af6ff", "_elc_node_type": 'operator', "_elc_function": "elc_output", },
        {"label": "OUTPUT", "id": "2d1af6ff", "_elc_node_type": 'operator', "_elc_function": "elc_output"},
    ],
    "edges": [
        {"source": "0ea5a129", "target": "8ac87236", "id": "74bc97ca", "_elc_source_output_id": '', "_elc_target_input_id": "a"},
        {"source": "3362b879", "target": "8ac87236", "id": "d3645364", "_elc_source_output_id": '', "_elc_target_input_id": "b"},
        {"source": "8ac87236", "target": "0d1af6ff", "id": "b0eb9a9b", "_elc_source_output_id": 'sum_result', "_elc_target_input_id": "a"},
        {"source": "3362b879", "target": "0d1af6ff", "id": "0e6c0fde", "_elc_source_output_id": '', "_elc_target_input_id": "b"},
        {"source": "0d1af6ff", "target": "9d1af6ff", "id": "7e6c0fde", "_elc_source_output_id": 'mul_result', "_elc_target_input_id": "x"},
        {"source": "0d1af6ff", "target": "1d1af6ff", "id": "1e6c0fde", "_elc_target_input_id": "kwargs"},
        {"source": "9d1af6ff", "target": "2d1af6ff", "id": "2e6c0fde", "_elc_target_input_id": "kwargs"},
    ]
}
# 从json结构parse成图
_graph = ELCGraph.create_from_elc_json(json_model)
# 编译图: 包含了对于节点拓扑排序并且初始化的工作
_graph.compile()
# 把输入节点的数据喂进去
_graph.feed_data_dict({
    '3362b879': 5,
    '0ea5a129': 6
})
# 执行图
_graph.execute()
# 序列化整个图结构
graph_str = json_stringify(_graph.to_dict())
# 解析序列化结构到图
new_graph = ELCGraph.load(json_parse(graph_str))
# 把结果画出来
new_graph.plot(show=True, with_state=True)

Example Graph 1

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

elc-flow-0.2.0.tar.gz (12.6 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page