In September, Kaggle released its annual survey for the state of data science and machine learning.
Here are my thoughts on these:
- India has been steadily rising in the data science rankings – especially in the number of Kaggle grandmasters coming from India.
- The popularity of Python and SQL is not surprising
- VSCode continues to gain traction reflecting Microsoft’s ability to create great tools
- Colab also continues to gain traction reflecting a general move to web-based architectures
- Pytorch grows year on end, but I still do not see a compelling reason for TensorFlow developers to switch to Pytorch
- Transformers for images are interesting (as opposed to text). I think it’s an early-stage application that will gain a lot more traction next year
- All cloud platforms are showing steady growth but may not in the future, considering the general downtrend
In one sense, the results are not surprising. Data science is a glorified profession, but the actual work is often very mundane. A recent Reddit thread shows that
XGBoost and Linear/Logistic regression are all you need… in practice in most cases
source: kaggle
