safeu 0.1.0

Safe learning for unlabeled data

SafeU: A python toolkit of Safe Learning for Unlabeled Data

Authors: De-Ming Liang, Feng Shi, Hai-Yu Chen, Xiao-Shuang Lv, Yong-Nan Zhu, Yu-Feng Li

Introduction

SafeU (Safe learning for Unlabeled data), is a python toolkit of safe machine learning algorithms utilizing unlebeled data (A brief introduction of safe semi-supervised learning can be found here). It builds in multiple safe semi-supervised learning algorithms, and provide a weakly-supervised learning experiment framework including some well-defined protocols for learning algorithms, experiments and evaluation metrics. With this toolkit, you build up your comparing experiments between learning algorithms with different learning settings like supervised, semi/weakly-supervised, as well as different tasks such as single/multi-label learning. We hope this toolkit could help you explore the classic semi-supervised learning algorithms and go further to test your ones.

Submit bugs or suggestions in the Issues section or feel free to submit your contributions as a pull request.

Getting Start

• For the latest news, blog posts, tutorials, papers, etc. related to SafeU, check out (Need a official release website)
• Get set up quickly
• Try the (tutorial).
• Read the (documents).

Setup

You can get safeu simply by:

pip install SafeU

Or clone safeu source code to your local directory and build from source:

cd SafeU
python setup.py safeu
pip install dist/*.whl

The dependencies of SafeU are:

1. Python dependency

python == 3.6 | 3.7

2. Basic Dependencies

numpy >= 1.15.1
scipy >= 1.1.0
scikit-learn >= 0.19.2
cvxopt >= 1.2.0

Examples

We can use safeu for algorithm experiments. The following example shows a possible way to experiment based on built-in algorithms and data sets:

import sys, os
from safeu.Experiments import SslExperimentsWithoutGraph
from safeu.model_uncertainty.S4VM import S4VM

# algorithm configs
configs = [
        ('S4VM', S4VM(), {
            'kernel': 'RBF',
            'gamma':[0],
            'C1': [50,100],
            'C2': [0.05,0.1],
            'sample_time':[100]
        })
    ]

# datasets
# name,feature_file,label_file,split_path,graph_file
datasets = [
	('house', None, None, None, None),
	('isolet', None, None, None, None)
	]

# experiments
experiments = SslExperimentsWithoutGraph(transductive=True, n_jobs=4)
experiments.append_configs(configs)
experiments.append_datasets(datasets)
experiments.set_metric(performance_metric='accuracy_score')

results = experiments.experiments_on_datasets(unlabel_ratio=0.75,test_ratio=0.2,
	number_init=2)

Reference

[1] Yu-Feng Li, Lan-Zhe Guo, Zhi-Hua Zhou. Towards Safe Weakly Supervised Learning. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), In press.
[2] Yu-Feng Li and Zhi-Hua Zhou. Towards making unlabeled data never hurt. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 37(1):175-188, 2015.
[3] Yu-Feng Li, De-Ming Liang. Lightweight Label Propagation for Large-Scale Network Data. IEEE Transactions on Knowledge and Data Engineering (TKDE), in press.
[4] Tong Wei, Lan-Zhe Guo, Yu-Feng Li, Wei Gao. Learning safe multi-label prediction for weakly labeled data. Machine Learning (MLJ). 107(4): 703-725, 2018.
[5] Yu-Feng Li, Shao-Bo Wang, Zhi-Hua Zhou. Graph quality judgement: A large margin expedition. In: Proceedings of the 25th International Joint Conference on Artificial Intelligence (IJCAI'16), New York, NY, 2016, pp.1725-1731.