Skip to content

Semi-Supervised Regression Comparison

Script: examples/semi_supervised_regression_comparison.py

→ API: PseudoLabelConsistencyLoss, PseudoLabelNLL. Guide: Uncertain ground truth.

Compares the pseudo-label + teacher-consistency branch of semi-supervised regression on a real-data proxy (sklearn Diabetes) using:

  • SupervisedMSE
  • PseudoLabelConsistencyLoss
  • PseudoLabelNLL

Workflow:

  • label only part of the train split
  • fit a Gaussian teacher on labeled data
  • generate confidence-weighted pseudo labels for unlabeled data
  • train SSL students with shared architecture and budget

Reported metrics:

  • MSE
  • MAE
  • R2
  • PseudoAcceptRate
  • PseudoMeanConf
  • train_s, eval_s

Run

pixi run python examples/semi_supervised_regression_comparison.py

Summary Artifact

pixi run python examples/semi_supervised_regression_comparison.py \
  --summary-json-path reports/example_summaries/semi_supervised_regression_comparison_full.json