Dhaval Patel

I implemented different variants of model-free deep reinforcement learning algorithms in Python and analyzed the data efficiency of each of the variants.

The overall objective of the entire project can be summarized in the below points:
• Configure Mujoco simulator to work on Windows.
• Implement deep deterministic policy gradient algorithm with generalized advantage estimation and asynchronous deep deterministic policy gradient with multiple updates.
• Analyze the data efficiency of both the algorithms along with a number of update steps.

The results and analysis can be found at the GitHub link.

Generate the headline of a news story (a group of news stories).

I architected and developed the entire pipeline which takes the set of news articles, performs the LexRank algorithm to select the candidate sentences, and passes it to the natural language generation algorithm to generate the headline of the news story.

I deployed this pipeline in Kubernetes to generate the headlines in real-time.

This project aims at applying the K-means clustering algorithm on the Stack Overflow dataset to group similar users and posts. Appropriate features are selected to extract the skill sets of users and the relevance of the posts.

The algorithm is completely implemented on PySpark to make use of parallel computation of spark and HDFS. The code is implemented without using the MLlib library of Spark, results are discussed and finally, it is compared with the results obtained after using Spark's Machine Learning library (MLlib).

The elbow method was applied to obtain the optimal number of clusters for both user and posts dataset. Additionally, two other functions are written to normalize the data and to implement one-hot notations for string type data (e.g., badges, tags).