Week 22, 2021
Those who do not know how to fight worry die young - Dale Carnegie
The staggering amount of workload that I have signed up for is starting to weigh on me. Thanks to my biking commutes to keep my body and mind in shape and my young age, I have been able to manage it all. I wonder for how long I can keep this up.
Work ★ ★ ★ ☆ ☆
I was focused on two things at work this week. First, problem-solving – triaging bugs, recreating them in demo environments, and passing onto software developers to fix those bugs in production. Second, I am supporting sales team of the company. We recently signed one of our biggest projects in terms of number of battery systems installed. The plan is to now use this approved project as a fulcrum to land more projects from utilities in the US and beyond. It’s a new area for me. It’s interesting.
Projects ★ ★ ★ ★ ☆
Machine Learning Bootcamp
- Tesla Car Sales Prediction: The task was to use sales data for Quarter 1 and 2 from Tesla, and predict the sales of the given models for the next 2 quarters. We had data for 2019. We trained the model on Q1 and Q2 sales, and tested the model on Q3 and Q4. This was a test to compare different regression models. We tried Gradient Descent, Normal Equations, Generalized Linear Models (non-linear regression), and Random Forest Regression. We compared the results based on RMSE and R2 scores. Random Forest Regressor with 100 estimators gave us the best results out of all.
- Walmart Labs Funnel Analysis – The task was to classify purchasing customers vs non-purchasing customers on their e-commerce store. Given were 15 features which included number of sessions, number of carts, max price of cart, time of day, day of week etc. After some pre-processing, like one-hot encoding all the categorical features, we dropped features with high correlation. We tried feature selection with p-value as well. We then used Random Forest to get feature rails/importances for each feature. Afterwards we ran the processed data through multiple classification algorithms – logistic regression, SVM, Neural networks etc. The comparison of each of these models using the confusion matrix was really cool. Using metrics like Accuracy, Precision, Recall, F1-score gave us a complete picture of the performance of the models.
CEEW – Sustainable Mobility Research Assistant
This week at CEEW, we reached a milestone in trying to build a Rapid Bus Shelter Assessment model. The code we have built so far – segments objects from an image – people, bus, car, animals, motorcycles etc. This is actually done by the pre-trained YOLO model. Much thanks to Tensorflow and whoever trained the model. Our code now takes this segmented image and applied certain “rules” to classify the image as Accessible/Inaccessible or Safe/Unsafe. The results are pretty cool given that we have less than 2 weeks to finish this. Dashboard : The next task for this project is building a web app for this service. I am using Dash/Plotly packages from Python to achieve this. Let’s see how this implementation goes.Books ☆ ☆ ☆ ☆ ☆
No reading at all this week 🙁
Fitness ★ ★ ★ ★ ☆
- Biking: Decent miles covered this week. I approached close to 30 miles. Thanks to commuting.
- Gym: I went to the gym only once this week. I feel horrible. Same as last week.