Talks S2E7 (Konrad Banachewicz): Time Series Analysis - Vintage Toolkit For Modern Times

Abstract : An overview of time series methods - from classics to modern ones - and how you can use them in practice; from power consumption to sales data, multiple seasonalities to almost random, there is something in the time series toolkit that will come in handy.

Speaker Bio: After finishing his PhD in statistics, Konrad has been crunching numbers for a living for years and have dabbled in just about everything along the way (credit risk analysis, trading commodities, predictive maintenance). These days he leads the central data science team at eClassifiedsGroup (part of Adevinta), where they optimize the e-commerce experience so people actually get what they need, and not merely what they clicked upon

---
Please subscribe and like the video to help me keep motivated to make awesome videos like this one. :)

Follow me on:

[shekharpandey]

Thanks Konrad and Abhishek for having this session and sharing knowledge on TS.

[astaconteazamaiputin]

This is a great learning resource. I've just finished an introductory book to time series, with hands on examples, and still I find great additional value in following this presentation. Thank you, Konrad, for sharing your knowledge and Abhishek for the effort to organize this, edit, upload and all the work involved ❤
And yes, the humour is so enjoyable, especially given the topics which are not the most relaxing thing to wrap your head around ^^ 1:10:00 is 💯

[sounds]

One of the most informative video about time series.. Just wishing to get an opportunity to work with Konrad someday and excited to change my career from predictive modelling to time series.. One day

[srijithsasidharanpillai]

Lot of stuff learned from this.. Thanks Konrad and Abhishek.. 👏👏👍

[hemanthharikrishnan]

Really an awesome talk.... Thank you @Abishek for organizing... really loved it, can you please share the names of the authors?

[mmkumars]

From the talk, "Intermittent demand, one of my favorite horror stories" :)

[abhinavraj]

thanks Abhishek for inviting him, can we have a session with him on quant finance models

[CodeToMuch]

Sir, I am very happy you join Google, Can you please Make a video on your journey from beginner to Hero.

[vigneshpadmanabhan]

Thanks for the interview... Good one... Shall we have the notebook link. Thanks!

[princepatil]

Please make a video on Chaii Google competition you have participated 10th rank Gold medal

[NikhilRaj-vzcq]

Hlo sir are u taking part in the Santa 2021 kaggle competition from Jan?

[urnakundu]

Could you share the notebook links, couldn't find any.bwill be helpful in my current org project

[XX-vujo]

I remember this guy being so motivated to make YouTube videos. Now, he realize that its not going the way he wants, he just quits. LOL!!!

[thsstphok]

Let say my model need 1 hour to train, I have 10 features ideas, and 10 hours limit to solve the problem. From a time efficiency perspective, what is better: change one thing at time or several?

1. Add features one at time and evaluate, e.g. If I will experiment 10 features, I need 10 hours ( 1 hour per feature ).

2. I can add several features and then eliminate the poor features after model evaluates their importance. This way I can train in one hour for 10 features and make more experiments trying diverse combinations or new features ideas in the next 9 hours.

In Design of Experiments is a proven fact that changing several factors at time (factorial experiment) is better than One Factor at Time ( OFAT). This holds true for DS experiment?