Week 31, 2021
I demo’ed the Bus Shelter Assessment app to the team. Ask me how? - I exposed my personal laptop over the internet. Read more below. I went against the warnings of one of cybersecurity-expert friend - Aditya Shinde and deployed it anyway. Things I do for low-budget proof of concepts!
Work ★ ★ ★ ★ ☆
The last two weeks have been productive at work. Just usual stuff. Putting out fires, making progress with ongoing ## projects and sprints, asset monitoring and management, and closely working with customers to answer questions with data and solve their issues if any. I have learned to be flexible at this company and I am not overly concerned about what I do day to day.
Projects ★ ★ ★ ★ ☆
Machine Learning Bootcamp
Decent week at the Bootcamp. In the last 2 weeks, we focussed on GAN models – Generative Adversarial Networks and Time series and Sequence modeling. One application of cycleGAN that we explored in our live session assignment is Uber’s use of the model to generate nighttime and extreme weather event images to train their autonomous driving models. In Sequence modeling, we explored RNN modes to predict outcomes of online browsing sessions for Cosmetics company. Pretty cool stuff – https://github.com/unmeshmali25/Sequence-Modelling-Cosmetics- Dataset/blob/main/Sequence_Learning_on_Cosmetics_Data_4B_v1_Unmesh.ipynb
CEEW – Sustainable Mobility Research Assistantp
I managed to solve the deployment problem with the web app to rank bus shelters in India using Computer Vision.
- The problem of deployment failure with PythonAnywhere, DigitalOcean, and AWS ElasticBeanstalk was memory limit. The virtual environment containing the web app, required dependencies , and the trained ML model sat at 1.4GB of space. Each of the above mentioned FREE third party hosting services allow for not more than 512MB . And, Heroku no longer supports Python-3.7.6 which is required (strictly) to run the ImageAI python package. That sucked.
- The solution I came up with is more of a hack than a scalable solution. I deployed the model on my local machine as a WSGI server using the gunicorn python package and exposed my IP address over the internet using secure HTTPs URL. I used a third party tunneling service called gw.run .
The last remaining task in this project is to transfer the app, model, and dependencies to a windows computer of my project supervisor. He has a good GPU on his laptop and we might be able to speed up the processing. Right now, our model takes about 5 seconds to process 1 image. I assume the city of Delhi has around 2000 bus shelters or more. That’s 3 hours to rank all the bus shelters in Delhi. Now Imagine that the state or central government install cameras on each of these bus shelters, those cameras will be feeding video and images of bus shelters all day long. With that kind of infrastructure, our model would be very inefficient. After completion of the project last week, I expressed my enthusiasm to work on other ingoing projects at CEEW and they sent me a list of projects to choose from. Next week, I will spend some time on preliminary research to choose a project. Enabling multi-modal freight logistics in India sounds like an awesome project. Another interesting project is Addressing fuel tax revenue loss with high EV penetration . Exciting times ahead with CEEW.
Books ★ ★ ★ ★ ★
Genentech: The Beginnings of Biotech | Sally Smith Hughes
I finished this book over the weekend. What an amazing experience it was to live the story of Genetech and the rest of the Biotech industry. I vaguely remembered studying biology in school but I never understood what Biotech meant. Hell, I didn’t even know that it is a scientific field that one could actually pursue outside academics. Thanks to the book, that perception has changed. I still have plethora of questions about the science and its production scaling but at least I have a rough mental model to put the right questions in the right place. Genentech was the first company (or a R&D facility) sponsored by large private pharmaceutical corporations to synthesize genes (think DNA) that if inserted into the right bacteria enables the bacteria to produce proteins (think Hormones) that can be used by humans. Insulin and Human Growth Hormones were the breakthrough products of Genetech that made the founders – Ron Swanson and Herbert Boyer et.al. rich beyond their wildest expectations. One amazing personal connection – During the research and development of Human Growth Hormone – Genetech recruited a scientist from UCSF. That scientist had to steal (or take what was his- long story) a DNA sample from UCSF by walking into the UCSF lab at mid night. A decade later, lawsuits against UCSF (along with this little maneuver) cost Genetech $150 million in fines to UCSF and $50 million dollars towards a R&D building at UCSF’s mission bay campus in San Francisco. This building is right across the street from where I have been going for Physiotherapy for close to a year now !!!
Fitness ★ ★ ☆ ☆ ☆
Really bad weeks for fitness 🙁 Been to the gym maybe only 3 times. I have been biking, playing Frisbee and Volleyball though. Need to concentrate more on fitness going forward.