pipez 0.0.108

No project description provided

Pipez - lightweight library for fast deploy stream handling

Install

For installing default version of library use

pip install pipez

If you want install specific version pipez - use

pip install pipez[<your choice>]

Now available cv, fastapi and onnxruntime versions. If you want install pypez with all depencies, you can use

pip install pipez[all]

If you want to install a few version - see nex example:

pip install pipez[cv, onnxruntime]

Quick start

Developing custom node

If you want use your node - you can use Registry.add as class decorator from pipez.registry. You should also import base Node class from pipez.node. For example:

from pipez.core.legacy_node import Node
from pipez.core.legacy_registry import Registry

Registry.add


class MyNode(Node):
    ...

Once required method which you should override: work_func(...) which handle Batch from pipez.batch. However, methods post_init(...) and close(...) also available. See next example:

from typing import Optional

from pipez.core.legacy_batch import Batch, BatchStatus
from pipez.core.legacy_node import Node
from pipez.core.legacy_registry import Registry

Registry.add


class MyNode(Node):
    def __init__(
            self,
            a: int = 1,
            **kwargs
    ):
        super().__init__(**kwargs)
        self._a = a

    def post_init(self):
        self._a *= 10

    def close(self):
        self._a = 0

    def work_func(
            self,
            data: Optional[Batch] = None
    ) -> Batch:
        self._a *= 2
        if self._a > 1000:
            return Batch(status=BatchStatus.END)
        return Batch(data=[dict(a=self._a)])

Build pipelines

When you defined all nodes what you need, we build pipeline from them. You can use json describe or class for node. See next examples:

For using json describing you must add Registry.add as class decorator for you node, else you will get error.

{
    "cls": "MyNode",
    "a": 5,
    "type": "Process",
    "output": "some_trash"
}

For using class you must import your node class.

from pipez.core.legacy_node import NodeType

from ... import MyNode

MyNode(
    a=5,
    type=NodeType.PROCESS,
    output='some_trash'
)

As we can see, we used NodeType, which define type of node.

For building pipeline, we must use build_pipeline from pipez.build. For example:

from pipez.core.legacy_build import build_pipeline
from pipez.nodes import DummyNode
from pipez.core.legacy_node import NodeType
from ... import MyNode

watchdog = build_pipeline(
    pipeline=[
        MyNode(
            a=10,
            type=NodeType.THREAD,
            output='q1'
        ),
        DummyNode(
            type=NodeType.PROCESS,
            input='q1',
            output='q2'
        ),
        DummyNode(
            type=NodeType.THREAD,
            input=['q1, q2'],
            output='q3'
        ),
        {
            "cls": "DummyNode",
            "type": "thread",
            "input": "q3"
        }
    ]
)

As we can see, build_pipeline return watchdog. You can read about it in next section.

WatchDog

TODO

РЎontributors

• Alexander, https://github.com/tam2511
• Vitaly, https://github.com/purple63