qlinsey / basal

AL

PROJECT: B-ASAL active learning framework

1. Install pip install -r requirements.txt

2. Framework

   • Multi-Class Discriminator/Classifier
   • Bi-Discriminator
   • Feature Encoder

3. Workflow - Cold-start data points generation -> Human Label -> Retrain Model -> Need more label (Y/N) ->Y, Retrain Model->...-> converge

4. Components

   • Dataset
   • Cold-start Samples
   • Training
   • Performance report
   • Inference/Score
   • Query for labelling
   • Add labelled data
   • Generate pesudo labelled data

5. Execution

   • Tree sh run_main.sh tree

   • Data info: sh run_main.sh info

   • Run Model Training: sh run_main.sh train

   • Collecting performance report: sh run_main.sh perf

   • Query data to be labelled: sh run_main.sh query

   • Predict/Score data points: sh run_main.sh score

   • Retrain or not: sh run_main.sh retrain

   • Cold Start: sh run_main.sh coldstart

6. References