README.md

Instructions

Build using:

swift build

Example run:

swift run -c release NoisyLabelsExperiments run \
  --dataset weather-sentiment \
  --train-data-portion 0.9 \
  --synthetic-predictors-count 16 \
  --use-synthetic-predictor-features \
  --results-dir temp/results-synthetic

Configurations

For medical-causes and medical-treats we are using LIA configured with the following options:

• Predictor Embedding Size: 32
• Instance Hidden Unit Counts: [32, 32, 32, 32]
• Predictor Hidden Unit Counts: []
• Confusion Latent Size: 1
• Gamma: 0.0
• Entropy Weight: 0.0
• Use Soft Predictions: true
• Learning Rate: 1e-4
• Learning Rate Decay Factor: 1.0
• Batch Size: 512
• M Step Count: 1000
• EM Step Count: 2
• Marginal Step Count: 1000