This was my first Posit Conference, and I was eager to learn about new technologies and tools. Interestingly, what I enjoyed most were the keynote speakers and the spontaneous conversations during breaks.
The keynote speakers were inspirational, offering a big-picture perspective and motivations for pursuing a meaningful career. The impromptu discussions with conference attendees—who came from diverse backgrounds gave me valuable insights, not just in data science, but in life.
Having had this wonderful opportunity to attend, I’d like to share a few key takeways. This blog isn’t about the specific topics covered in the conference session, but rather about what I took away from the overall experience.
Keynote take-away
At times work can settle into comfortable routines or become a pursuit of personal intellectual satisfaction.
While goals and objectives guide our work and shape our workflows, I often reflect on what the work truly means to me.
Some of the keynote messages resonated with me as questions i can ask myself.
How can we be innovative to make things better?
Whatever I do, aim to contribute–even in small ways–to creating positive change.
Create Virtuous cycles to benefit everyone involved. Think of ways to reinvest my benefits back into the community
Open Science: Use data to understand the world better. Then we know how to make it better
Negative bias is a serious threat to our well-being and ability to address the problems we face
Working with people from different background can be challenging.
Agree on a set of core values such as radical transparency, trust, respect, courage, commitment
Sessions & Conversation take-away
Working with people from different backgrounds can be challenging. It’s important to agree on a set of core values, such as radical transparency, trust, respect, courage, and commitment.
ChatGPT is becoming widely used, especially among the new generation of professionals. Traditional software engineering interviews often involve requesting code samples, but this practice is becoming less relevant as many candidates now use ChatGPT. What does the future hold? Perhaps it lies with those who can construct the “right” questions to solve difficult problems.
Key Statiscal and Machine Learning tools
{marginaleffects}package for model comparisons. you write a comparison function and the package help to generate ggplot and other output to compare models.- Prediction Intervals: github.com/brshallo/posit-2024, talked about different ways to construct and evaluate prediction intervals
- Interval width is widely used and good to evaluate if a model is improving as the width becomes narrow. (Constant variance desired)
- in Conformed prediction, the assumption is relaxed, and you compute upper and lower bands. (manokhin molar). Use train, calibration, and test datasets for estimating intervals.
- Deep learning:
{keras3}allows the use of keras and tensorflow in python throoughreticulate.- create neural network architecture just like python keras.
- offers the interface to GPU.
- save the models in *.keras to change the backend for different cycles. (save and read model)
- distributed training
- keras3.posit.co
*Avoid garbage in/garbage out: Use interactive tool like shiny for data cleaning
Tools to check out
- Web/Report publishing : Quarto Dashboard for web dashboard and quarto PDF using typst (faster than latex)
- Typst : Does Design Matter? Studies show aesthetically pleasant design make things more useful and builds trust. Make reports beautiful, reproducible, parameterized, automatically generated
- Education/Training : WebR/Quarto live for interactive web interface with programming language, ideal for training. It can be either a standalone web application or a code chunk in a quarto document.
- Cloud Computing: Posit Workbench is a cloud computing and it handles infrastrcuture manageement and security requirements. It can be deployed in Posit:Connect.
- Cloud and Collaboration: Github codespace for computing and sharing simultaneously and is free for # of hours and you can use docker image to set a virtual environment.
{pins}: customizing and automating report generation and delivery