Vuong Van Pham @ver2king

Today in my #MachineLearning course, we're diving into Convolutional Neural Networks (CNNs) 🧠📊 — a powerful technique in image-based #DeepLearning 🤖. To keep it practical, I’ve built a simple example that classifies random vs. smooth images 🎨📷. We'll train the model, evaluate its performance 📈, and even visualize the learned kernels and feature maps 🔍🧩 to see what the network is learning! I share it on #GitHub @ github.com/GeostatsGuy/Py… ∀.

keyboard_arrow_left Previous keyboard_arrow_right Next

My course lecture on building Transformers from first principles: dropbox.com/scl/fi/jhfgy8d… The PyTorch notebooks also demonstrate how to implement Transformers from scratch: github.com/xbresson/CS524…

keyboard_arrow_left Previous keyboard_arrow_right Next

I have removed all copy and paste images and hand drawn schematics from my courses! Every example includes links to my shared workflows and codes accessing my #OpenData to reproduce the exact images in my course notes! This is my commitment to reproducible, accessible and actionable #education for my students and all those that access my shared educational content on #YouTube and #GitHub. #AcademicChatter #AcademicTwitter

If your #DataScience does not impact the decision, you do not add value! This is what I tell my students. So after an entire course of building #spatial uncertainty models, today we will learn how to make #decisions in the presence of #uncertainty. To help my students, I built out this #Python @matplotlib dashboard for optimum decision making and shared it on #GitHub @ github.com/GeostatsGuy/Da… ∀.

When I teach & demonstrate #DataScience algorithms I use the @MythBusters approach! My kids and I loved watching Jamie Hyneman and Adam Savage (@donttrythis) run their experiments to failure! The best way to learn a system, model or theorem is to break it! For example, increase the sampling rate & vertical spatial continuity until kriging-based spatial estimation demonstrate irrational data weights - known as the 'string effect'. I share my #Python @matplotlib dashboard on #GitHub @ github.com/GeostatsGuy/Da… ∀.

I built out this interactive #Python @matplotlib dashboard to teach #MachineLearning hyperparameter tuning. Build a regression decision tree, change the number of leaf nodes while watching the training & testing data cross validation scatter plot and the position on the MSE vs. complexity curve. Ok, good, better, best → overfit! I share the code at github.com/GeostatsGuy/Da… ∀. #DataScience

Are you working in #DataScience, #DataAnalytics, & #MachineLearning? Want to deepen your understanding of feature engineering with distribution transformations? I'm happy to help! Join me for my brand-new #YouTube walk-through @ youtube.com/watch?v=kcbUjY… linked code! Part of my new "Data Science Interactive Python Demonstrations" series! Accessible and actionable data science education for all!

Are you confused by ordinary least squares (OLS) vs. maximum likelihood estimation (LME) for model parameter fitting? #MachineLearning To help my students, I just built out an interactive #Python dashboard with @matplotlib to visualize & compare OLS & MLE fitting of a Gaussian distribution to a dataset. I share it on #GitHub @ github.com/GeostatsGuy/Py… ∀. #DataScience

For improved rigor in my #MachineLearning course, I dived in and coded #Bayesian linear regression with McMC Metropolis-Hastings from scratch to build a super accessible demonstration. What could be better than that? I just added an interactive #Python dashboard with @matplotlib on #GitHub @ github.com/GeostatsGuy/Py… ∀. #DataScience

You do #MachineLearning? Do you calculate #correlation coefficients? How about covariance or PCA?, etc. OK, but beware! How do #outliers impact your results? ... totally! Check out my interactive #Python @matplotlib #Dashboard on #GitHub @ github.com/GeostatsGuy/Py… ∀. #DataScience

To help my students learn #deeplearning #convolutional #NeuralNets, @HonggeunJ & I coded a very simple demo #CNN to classify by the length of randomly positioned lines. Demonstrates scale and location invariance! To dive deeper, we made functions to visualize the kernels and feature maps over the entire model! Super cool! On #GitHub @ github.com/GeostatsGuy/Py… ∀. #DataScience

keyboard_arrow_left Previous keyboard_arrow_right Next

To help my students learn #deeplearning #convolutional #NeuralNets, @HonggeunJ & I coded a very simple demo #CNN to classify original & smoothed images. To dive deeper, we made functions to visualize the kernels and feature maps over the entire model! Super cool! On #GitHub @ github.com/GeostatsGuy/Py… ∀. #DataScience

keyboard_arrow_left Previous keyboard_arrow_right Next

I built an #interactive #Python @matplotlib #dashboard for #density-based #cluster analysis with #BDSCAN to support my students & anyone interested to learn about #MachineLearning! Aside, fun practice with code recursion! Stoked! On #github @ github.com/GeostatsGuy/Py… ∀. #DataScience

Doing #MachineLearning for spatial problems? #geostatistics-based #simulation is a powerful tool to calculate #spatial / #temporal models that honor univariate, bivariate & spatial distributions with #realizations to represent #uncertainty. I built out this #interactive #Python @matplotlib dashboard on #github @ github.com/GeostatsGuy/Py… ∀. #DataScience

What about #predictive #MachineLearning with #radial #basis functions (RBFs)? A predictive model from the summation of many RBFs, flexible with intuitive #hyperparameters! To teach RBFs to my students I built out this #interactive #Python #dashboard with @matplotlib on #GitHub @ github.com/GeostatsGuy/Py… ∀. #DataScience

I teach k-nearest neighbours prediction in my #MachineLearning course for the connection to #spatial #mapping & #convolution. To help explain the concepts I made an #interactive #Python @matplotlib #dashboard comparing convolution, kNN, for exhaustive and sparse data! See it on #GitHub @ github.com/GeostatsGuy/Py… ∀. #DataScience

I built out this interactive #Python @matplotlib #dashboard to teach #MachineLearning #hyperparameter tuning with a decision tree prediction #model. Grow the tree, increase model complexity and track the training and testing error from underfit to overfit! On GtiHub @ github.com/GeostatsGuy/Py… ∀. #DataScience

#Ridge & #LASSO #MachineLearning models use L2 & L1 #regularization respectively for model hyperparameter tuning. What does that look like? This! I just built out a #Python @matplotlib dashboard visualizing the data fit & regularization losses as the #hyperparameter is adjusted! Stoked! On #GitHub @ github.com/GeostatsGuy/Py… ∀.

I'm teaching ridge regression in my #MachineLearning course. Why? It demonstrates a lot of important concepts! 1. hyperparameter tuning, sometimes linear regression is too complicated! 2. regularization, to avoid overfit! 3. L2 vs. L1 norms, in comparison with LASSO! See my interactive #Python dashboard with @matplotlib on #GitHub @ github.com/GeostatsGuy/Py… ∀.

The concept of #statistical #significance is essential to protect ourselves from the fallacy of the belief in the law of small numbers. Differences may be due to #random #effect, especially if you have a small sample set! To support my students, I made a #Python dashboard with @matplotlib on #GitHub @ github.com/GeostatsGuy/Py… ∀.