使用DataFlow API构建graph
- 构建FlowGraph。通过FlowNode和FunctionPp构建graph。示例如下：
1 2 3 4 5 6 7 8 9 10 11 12 13
import dataflow as df # 定义输入 data0 = df.FlowData() data1 = df.FlowData() # 定义FuncProcessPoint实现Add功能的FlowNode pp0 = df.FuncProcessPoint(compile_config_path=”./add_func.json”) flow_node0 = df.FlowNode(input_num=2, output_num=1) flow_node0.add_process_point(pp0) # 构建连边关系 flow_node0_out = flow_node0(data0, data1) # 构建FlowGraph dag = df.FlowGraph([flow_node0_out])
- 运行graph
- Dataflow方式，示例如下：
1 2 3 4 5 6 7 8 9 10 11 12 13 14
# dataflow 初始化参数按需设置 options = { "ge.exec.deviceId":"0", "ge.exec.logicalDeviceClusterDeployMode":"SINGLE", "ge.exec.logicalDeviceId":"[0:0]", "ge.socVersion": "Ascendxxx", # 必须设置的参数 } df.init(options) flow_graph.feed_data({data0:feed_data0, data1:feed_data1}, flow_info) # 异步喂数据 result = dag.fetch_data() # 异步取结果 # 释放dataflow资源 df.finalize()
- Dataflow方式，示例如下：
父主题： UDF开发