alaas 0.0.1

A Distributed MLOps System for Efficient Active Learning

ALaaS: Active learning as a service.

Installation

You can easily install the AlaaS by pip,

pip install alaas

Quick Start

Define the global configuration

We use a global configuration file to control the whole system, an example looks like:

name: "IMG_CLASSIFICATION"
version: 0.1
active_learning:
  budget: 10
  strategy:
    algorithm: "LeastConfidence"
    infer_model:
      model_hub: "huggingface"
      name: "resnet"
      batch_size: 4
      input_dtype: "float32"
  al_server:
    address: "54.251.31.100:8900"
    gpus: 'all'

The infer_model is designed for inference service inside some active learning strategies, by setting the al_server, we can help you automatically deploy the inference model from the model repository.

Start the active learning server

You need to start an active learning service before conducting the data selection.

from alaas import Server

al_server = Server(config_path='./you_config.yml')
al_server.start()

Define the active learning client and perform querying

You can easily start the data selection by the following code,

from alaas import Client 

al_client = Client(server_url="127.0.0.1:8888")
al_client.push(data_list, asynchronous=True)
al_client.query(budget=100)

Example output

the example output will be something like:

preparing data...
Files already downloaded and verified
start active learning, query number: 1000...
[uri_1, uri_2, uri_3, ...., uri_n]

which are the uri of input data selected by active learning.