hei4/pytorch-active-learning

Active Learning for PyTorch

pytorch-active-learning

Active Learning with PyTorch

This implementation is based on "Human-in-the-Loop Machine Learning: Active learning and annotation for human-centered AI". official repository: URL

Requirement

jupyter==1.0.0
matplotlib==3.5.1
numpy==1.22.3
pandas==1.4.2
Pillow==9.1.0
scikit-learn==1.0.2
scipy==1.8.0
torch==1.10.1
torchmetrics==0.8.1
torchvision==0.11.2

---

Uncertainty Sampling

Usage

python uncertainty_sampling.py --algorithm {least,margin,ratio,entropy,montecarlo} --data {moons,circles,gaussian,blobs}

• algorithm
  • least: Least condidence sampling
  • margin: Margin of confidence sampling
  • ratio: Ratio of confidence sampling
  • entropy: Entropy-based sampling
  • montecarlo: Monte Carlo dropout sampling
• data
  • moons: sklearn.datasets.make_moons
  • circles: sklearn.datasets.make_circles
  • gaussian: sklearn.datasets.make_gaussian_quantiles
  • blobs: sklearn.datasets.make_blobs

Examples

Entropy-based sampling & make_gaussian_quantiles dataset

Monte Carlo dropout sampling & make_moons dataset

---

Diversity Sampling

Usage

python diversity_sampling.py --algorithm {outlier,cluster,random} --data {moons,circles,gaussian,blobs}

• algorithm
  • outlier: Model-based outlier sampling
  • cluster: Cluster-based sampling
  • random: Random sampling
• data
  • moons: sklearn.datasets.make_moons
  • circles: sklearn.datasets.make_circles
  • gaussian: sklearn.datasets.make_gaussian_quantiles
  • blobs: sklearn.datasets.make_blobs

Examples

Model-based outlier sampling & make_gaussian_quantiles dataset

Cluster-based sampling & make_moons dataset

---

Advanced active learning

Least confidence with cluster-based sampling

python least_cluster_sampling.py --data {moons,circles,gaussian,blobs}

Margin of confidence with model-based outlier sampling

python margin_outlier_sampling.py --data {moons,circles,gaussian,blobs}

Sampling from the highest entropy cluster

python cluster_entropy_sampling.py --data {moons,circles,gaussian,blobs}

---

Active transfer learning

Active transfer learning for uncertainty sampling

python least_cluster_sampling.py --data {moons,circles,gaussian,blobs}

Adaptive active transfer learning for uncertainty sampling

python least_cluster_sampling.py --data {moons,circles,gaussian,blobs} --adaptive