AI Winter Camp
Over winter break of 2021–2022, I took an AI Camp program that taught
• Machine Learning Basics
• How to become a great developer
• Strategies to breakdown and solve complex challenges
• Real world data analytics and visualization best practices
The ultimate goal was to give a presentation on our project, which was analyzing data from the real world and creating a website showing the trends.
The 1st day, we learned about AI, the different types of machine learning (supervised, unsupervised), reinforcement learning, and deep learning. We were also introduced to input layers, the convolutional neural network, and the generative adversarial network.. We also reviewed some python
Day 2, we dove right into exploratory data analysis. Using Kaggle, we took a couple of datasets and imported into our CoCalc, a python-based collaborative platform for computational analytics. I thought the Heart Failure Prediction Dataset would’ve produced the most interesting answers, considering how many factors were involved and the vast amount of conclusions we could draw. So we began familiarizing ourselves both with the content and how to use CoCalc and its functions. Using python functions (such as .head() and .tail()), we displayed our results and looked for trends. Some of the trends we found were that
- Age and heart disease are positively correlated
- Maximum heart rate and age are negatively correlated
- Men are at a slightly higher risk of developing heart disease
- People with ASY (Asymptomatic) symptoms have higher chance of heart disease (which we predict could be influenced by genetics)
The 3rd day, we learned how to use CoCalc to display different graphs. For this, we used plotly, which gave us different function templates to apply onto our own dataset to display graphs (basically to create graphs and data visualizations, helping us display correlations between variables in our data). For example, we used their bubble graph template, which we then displayed a bubble chart comparing maximum heart-rate and age and cholesterol for male and female patients.
We also created more began creating our website by transferring all of our Jupyter notebook into functions onto our python notebook.
On day 4, we learned about reinforcement learning and regression models. We also constructed our own prediction form using a regression model.
After that, we continued working on our presentation; we created two people who would use the regression models and give different results. Throughout the week, we also created a website displaying our graphs, and the regression model: https://share.streamlit.io/taranatarajan9/heart-disease/main/app-v2.py
On day 5, we presented our presentation and website to the rest of the camp.
https://share.streamlit.io/taranatarajan9/heart-disease/main/app-v2.py
Overall, I learned a lot from this program. I learned how to analyze data sets, discover trends, create websites, write regression models, and retouched on python basics (but learned more functions). On this team, I helped find the data sets to use (although there were other ones, I specifically pitched the heart disease one because of how many trends there were — the more trends, the more analysis we could use). I also wrote many of the slides. Additionally, on our co-calc, I created many of the functions (which we imported from other resources), and found a lot of trends. I also screen-shared for our final presentation.
Another challenge was collaboration — it was important that people had their cameras on because at times, it would only be me and some other people working on our project while others had their cameras off not contributing. Also, working on the same co-calc was challenging because everyone did something different, but at times, there were overlap and the overall journal/code would look messy. We had to go back in and make sure that when we created the website, it wouldn’t look messy.
Some challenges we faced were getting used to the AI functions, and understanding what methods we could use to display certain results. Along with this, we also needed to figure out which variables to look at, to give us the best results for what we were trying to find. There were also logic/syntax errors that prevented us from reaching our goals (which we eventually figured out).
