Popular articles by boryana

Popular comments by boryana


I really like that the whole process of modelling is backed up by sound business logics. I find that your approach of incorporating exogenous features such as data on relevant stocks and Google trends adds a significant extent of originality in the adopted research methodology.
Data prep is conducted in compliance with the core theoretical requirements.
The implementation of rolling sample is correct.
What I would like to advise on the model is to consider statistical significance, especially for the estimates associated with the exogenous explanatory variables. Also, looking at the plot of actual vs predicted 1-step-ahead data, I might state that the model captures really tightly the series volatility for the first 12 000 observations. In order to tune better the model you might consider the reason behind the deteriorated performance aftermath. Once again, my suggestion is to inspect how statistical significance of delivered estimates changes over time.
Congrets on reporting the figure of directional symmetry!
Also, I really like the way your workflow is organized taking advantage of both R and Python utilizing the one that is best suited to the research task at hand.
Great job, guys!

Datathon – HackNews – Solution – Leopards

What I like most about your article is the feature engineering process. As it seems you’ve invested a lot of time in it. As a consequence the feature space is well-grounded.

Datathon – HackNews – Solution – LAMAs

It is admirable that you’ve put efforts on solving the three tasks rather than focusing on a single one. Even though the report is concise it presents clearly major results.

Datathon Sofia Air Solution – Team Duckelicious Case Telelink

Working with data for 2018 only is a good solution so as to deliver more quickly consistent representation. Also, focusing on one main station is a good choice with respect to the timing of the task. I like the presented maps. I would like to see at least part of your code.