Summary of GSoC'21 (under Tensorflow) with pyprobml

My journey with pyprobml started in march'21 after going through the announced GSoC projects. I worked on making the legacy code of probml in python and also on new code for the freshly-brewed topics from the vol2 (in making!) of the book

Almost every figure in the books will be generated by a script/notebook that's optimized and thoroughly documented.

This has been amazing with all the new learnings-unlearnings, challenges and accomplishments.

In this blog, I would like to share all my work(that's been available publicly) for the pyprobml codebase and also a bunch of awesome! things that I have witnessed and learnt during this period.

My contributions to the probml org:

Below are the list of PRs that's been merged, each of them refer to a specific column in the books. 

        Pre-GSoC:

• Cluster yeast data using kmeans 🔗
• Different kernel binary classifiers demo on bishop data 🔗
• Kernel regression demo using NW Smoother 🔗
• Ridge Regression demo using rbf kernel 🔗
• SVM demo tweaking for different C(regularizer) and gamma(rbf scale) values 🔗
• Implemented Probit regression (using EM) and and compared with the bfgs version 🔗
• Implemented Bayes linreg with gaussian prior and variational bayes with demo 🔗
• custom VAE on celeba and mnist using pytorch 🔗

       During-GSoC:

• prior and posterior predictive for beta binomial. 🔗
• Bayes Multi-Linear regression with imputation using numpyro 🔗
• Sparse coding(Dictionary learning) demo using spams 🔗
• Implemented Fisher LDA fit and a demo on vowel data 🔗
• Feature Extractor for any tensorflow-dataset using CLIP on TPUs 🔗
• MLP, Logreg demos on clip-feature extracted Imagenette dataset using flax, pytorch-lightning 🔗
• Tutorial on Google Cloud Storage(GCS) using google-colab in different ways 🔗
• Tutorial on Git-LFS using googe-colab 🔗
• Feature Extractor for any tensorflow-dataset using jax version of CLIP on TPUs 🔗
• Improved Cifar-cnn(Resnet-18) using pytorch-lightning 🔗
• Very-deep-vae(vdvae)-pytorch recons and gens demo on google-colab 🔗
• Very-deep-vae-jax(vdvae-jax) recons and gens demo on google-colab 🔗
• Robust prior(Cauchy) demo 🔗
• Plotting Upper bounds for the sigmoid(logistic) function🔗
• Plotting optimal lower bounds for exp(-x)🔗
• Gaussian graphical model(ggm) fit demo given the adjacency graph 🔗 

Hey, if you want to know about some cool stuff I learnt in this period go here ➤