imarnabsaha95 / dynamic_weighted_clustering

It contained the solution of Shell.ai Hackathon for determining the depot and refinery location. Was selected in the top 20 finalist among more than 5700 competitors.

Dynamic_Weighted_Clustering

It contained the solution of Shell.ai Hackathon for determining the depot and refinery location. Was selected in the top 20 finalist among more than 5700 competitors. The Shell Hackathon file is the original file where all the experiments including some EDA and then the development of the Dynamic Weighted Clustering was done. It also contains some forecasting as well as result generation part. In short it is all is all notebook for the actual Hackathon result generation and hence it is supposed to be a bit messy. The Dynamic_Weighted_Clustering notebook is the notebook that you should refer to if you are only interested in the clustering part. The clustering is developed on top of K-Means clustering but if has a threshold of werights which means that it can only accept weights upto a certain limit in each cluster. Also the distacne of the weight from the center is an important factor. For example : If the weight is too huge and the distance from the center is quite less then it will definitely be selected. If the weight is huge but the distance is high or the weight is too small but the distance from the center is also small then those may or may not be selected based on the scenario. But if the weight is too low and the distance from the center is quite low as well then the change of that point being selected reduces. Another important factor is how the cluster centers are selected, becuase here as well importance has been given to both distance as well as the weight. If you want to have an indepth understanding then you may consider looking into the library where the code has been build from scratch. There is also a YouTube Video explaining the concept which was originally made for Shell Hackathon. You can have a look into it as well. URL : https://youtu.be/gB7cmxTGu5U It may not be possible to directly take the function from the library and use it, as it has been developed based on the criteria and the dataset provided. But the idea can be adopted with minor changes for further use in different applications.