Contrastive Learning

Experiment 3.2

Experiment 3.2 from Hermans et al., 2020 :

1. Install the hypothesis package first
2. Move to exp_2
3. Run hermans_tractable_mcmc_groundtruth.ipynb
4. Run hermans_tractable_mcmc_ratioestimator.ipynb

Experiment 3.3

More figures and details for Experiment 3.3 (Learning a Gaussian mixture using neural network classifier): Single layer NN and Multi-layer NN (to run, either install relevant Python packages or build conda environment from requirements.txt)

Experiment 3.4

Reproduce the experiment 3.4 with Bayes factors :

1. Install the hypothesis package first
2. Move to exp_4
3. Scenario 1 : run metropolis_hastings_1a.ipynb, then metropolis_hastings_1c.ipynb, finally bayesfactor_1ac.ipynb.
4. Scenario 2 : if metropolis_hastings_1a.ipynb has been run (see Scenario 1 above), then run metropolis_hastings_1b.ipynb, finally bayesfactor_1ab.ipynb.
5. Scenario 3 : run metropolis_hastings_2c.ipynb, then metropolis_hastings_2a.ipynb, finally bayesfactor_2ac.ipynb.