Examples of my projects:
recsys.ipynb: Hacking problem of building a DL recommendation system for product usage data. Tried approaches - Nearest Neighbors, Alternating Least Squares, Logistic Matrix Factorization, Factorization Machines. Frameworks - Implicit, Gluon, Mxnet. MAP@5 = 0.12.dlexample.ipynb: A problem in a datathon to build a Computer Vision classifier. I created a CNN and pretrained models VGG16, Mobile Net 2. Accuracy of 91%+ on validation set. It is written to be executed in Google Colab (some packages are specifically for that environment)salary_it_linear.ipynb: A linear regressor for employee salary prediction based on their performance in certain areas. I performed data exploration, attempted stratified sampling and new feature design, built a linear regression model and applied XGBoosting. The final RMSE is 0.0392, R2 score on validation set is 0.969.parallelization_taskdirectory: contains a jupyter notebook and a function for parallelization on CPU. I performed feature engineering based on email traffic. The key aim was performance optimization, as a result of my parallelization function I was able to cut processing time 3.5 times (from 45 seconds to build 9 features just down to 13 seconds)NUD1p1.ipynb: Given mobile traffic input with unnamed features, it was needed to predict the age of a user. Mean Absolute Error for age prediction: down to 9.4 on validation set.