Using APIs to Construct and Execute a Graph

import dataflow as df
import numpy as np

# Initialize the system.
options = {
    "ge.exec.deviceId":"0",
    "ge.experiment.data_flow_deploy_info_path":"./data_flow_deploy_info.json",
    "ge.socVersion": "AscendXXX"  # Modify the version based on the environment.
}
df.init(options)

# Define FlowData.
data0 = df.FlowData()
data1 = df.FlowData()

# Define FuncProcessPoint to implement the Add function and add it to FlowNode.
pp0 = df.FuncProcessPoint(compile_config_path='config/add_func.json')
add_node = df.FlowNode(input_num=2, output_num=1)
add_node.add_process_point(pp0)

# Construct edge connections between FlowData and FlowNode.
add_out = add_node(data0, data1)

# Build a FlowGraph using FlowOut.
dag = df.FlowGraph([add_out])

# Call FlowGraph.feed_data to feed inputs.
flow_info = df.FlowInfo()
dag.feed_data({data0:np.array([[1, 2]], dtype=np.int32), data1:np.array([[2, 3]], dtype=np.int32)}, flow_info)

# Call FlowGraph.fetch_data to obtain outputs.
print("dataflow fetch result:", dag.fetch_data())

# Release system resources.
df.finalize()