Probabilistic-Programming

Assignment 1

• Link to WandB project : https://wandb.ai/aliseyfi/LDA/runs/1ykjf0cn/overview?workspace=user-aliseyfi

Project

• bbvi/ : here is bbvi code for lr=0.05 and L=5
  • You have access to wandb where you can have that beautiful plot of ELBO
  • bbvi.out you have q after each 25 steps here
  • save/ you have elbo_plot, posterior_plot, and q_plot plots
• IS/:
  • you have thetas.pt and weights.pt which is for 20000 samples
  • IS.out results of 20000 samples
  • Experiment_without_BBVI: This is the results of starting with the original prior we had
    • IS_#samples.out results of #samples
    • results/ :
      • MSE, thetas, weights, within_3 results are here, they are all pytorch objects that you can load them into code using pytorch.load().
  • Experiment_with_BBVI: This is the results of starting with final proposal of BBVI:
    • IS_#samples.out results of #samples
    • results/ :
      • MSE, thetas, weights, within_3 results are here, they are all pytorch objects that you can load them into code using pytorch.load().
• MCMC/: results of 1000 samples
  • MCMC.out : final results
  • figures/ containing histograms and traces for each theta
• smc_IS : This is our implementation of SMC using IS
  • Experiment_without_BBVI: This is the results of starting with the original prior we had
    • smc_#particles.out results of #particles
    • results:
      • MSE, thetas, weights, within_3 , and logZ results are here, they are all pytorch objects that you can load them into code using pytorch.load().
  • Experiment_with_BBVI: This is the results of starting with the original prior we had
    • smc_#particles.out results of #particles
    • results:
      • MSE, thetas, weights, within_3 , and logZ results are here, they are all pytorch objects that you can load them into code using pytorch.load().