alaas 0.1.6

A Distributed MLOps System for Efficient Active Learning

ALaaS: Active Learning as a Service.

Active Learning as a Service (ALaaS) is a fast and scalable service framework for users to conduct the data selection before human labeling. It can be easily integrated with existing data processing and labeling platforms as a microservice.

ALaaS is featured for

• :rocket: Fast Use the stage-level parallel to achieve over 10x speedup than normal active learning process.
• :collision: Elastic Scale up and down multiple active workers on single or multiple GPU devices.
• :hatching_chick: Easy-to-use With <10 lines of code to start APIs that prototype an active learning workflow.

Try It Out

You may just want to use the pre-trained model as the active data selector to help you select the most informative data samples from the unlabeled data pool. We have a CPU-based server for data selection demonstration (least confidence strategy with ResNet-18), try it by yourself!

HTTP

gRPC

curl \ -X POST http://13.213.8.21:8081/post \ -H 'Content-Type: application/json' \ -d '{"data":[{"uri": "https://www.cs.toronto.edu/~kriz/cifar-10-sample/airplane1.png"}, {"uri": "https://www.cs.toronto.edu/~kriz/cifar-10-sample/airplane2.png"}, {"uri": "https://www.cs.toronto.edu/~kriz/cifar-10-sample/airplane3.png"}, {"uri": "https://www.cs.toronto.edu/~kriz/cifar-10-sample/airplane4.png"}, {"uri": "https://www.cs.toronto.edu/~kriz/cifar-10-sample/airplane5.png"}], "parameters": {"budget": 3}, "execEndpoint":"/query"}'

from alaas.client import Client url_list = [ 'https://www.cs.toronto.edu/~kriz/cifar-10-sample/airplane1.png', 'https://www.cs.toronto.edu/~kriz/cifar-10-sample/airplane2.png', 'https://www.cs.toronto.edu/~kriz/cifar-10-sample/airplane3.png', 'https://www.cs.toronto.edu/~kriz/cifar-10-sample/airplane4.png', 'https://www.cs.toronto.edu/~kriz/cifar-10-sample/airplane5.png' ] client = Client('grpc://13.213.8.21:60035') print(client.query_by_uri(url_list, budget=3))

Then you will see 3 data samples has been selected from all the 5 data points by active learner.

Installation

You can easily install the ALaaS by PyPI,

pip install alaas

The package of ALaaS contains both client and server parts. You can build an active data selection service on your own servers or just apply the client to perform data selection.

Start the active learning server

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

from alaas import Server

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

How to customize a configuration for your deployment scenarios can be found here.

Querying data from client

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

from alaas.client import Client

client = Client('http://0.0.0.0:60035')
queries = client.query_by_uri(<url_list>, budget=<budget number>)

The output data is a subset uris/data in your request, which means the selection results for further data labeling.

Support Strategy

Currently we supported several active learning strategies shown in the following table,

Type	Setting	Abbr	Strategy	Year	Reference
Random	Pool-base	RS	Random Sampling	-	-
Uncertainty	Pool-base	LC	Least Confidence Sampling	1994	DD Lew et al.
Uncertainty	Pool-base	MC	Margin Confidence Sampling	2001	T Scheffer et al.
Uncertainty	Pool-base	RC	Ratio Confidence Sampling	2009	B Settles et al.
Uncertainty	Pool-base	ES	Entropy Sampling	2009	B Settles et al.
Uncertainty	Pool-base	BALD	Bayesian Active Learning Disagreement	2017	Y Gal et al.
Clustering	Pool-base	KCG	K-Center Greedy Sampling	2017	Ozan Sener et al.
Clustering	Pool-base	KM	K-Means Sampling	2011	Z Bodó et al.
Clustering	Pool-base	CS	Core-Set Selection Approach	2018	Ozan Sener et al.
Diversity	Pool-base	DBAL	Diverse Mini-batch Sampling	2019	Fedor Zhdanov
Adversarial	Pool-base	DFAL	DeepFool Active Learning	2018	M Ducoffe et al.

License

The theme is available as open source under the terms of the Apache 2.0 License.