Power Bi Key influencers visualization | TEST OR TRUST?

Просмотров 10 тыс.

% 242

Power Bi Key influencers visualization | TEST OR TRUST?
Do you know how the key influencer visual works behind the scenes? What do you do with the results, you test or trust?
Here are my thoughts on the topic and dont miss the SCI analogy at the end!!
Back at you, do you trust or test?
Tutorial on key influencer: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-5huk5emC2hc.html
ML.net article: dotnet.microsoft.com/apps/machinelearning-ai/ml-dotnet/customers/power-bi
Article on verifying the key influencer visual: medium.com/@prickofdeath/exploring-key-influencers-algorithms-with-citizen-tools-850284d26dc6
The wald test: www.statisticshowto.datasciencecentral.com/wald-test/
01:30 How does the key influencer visual work?
04:30 How to check the key influencer visual as a data scientist
05:30 Suggestions for improvements on the visual
07:30 Should we check the results of the visual?
Here you can download all the pbix files: curbal.com/donwload-center
SUBSCRIBE to learn more about Power and Excel BI!
ru-vid.com/show-UCJ7UhloHSA4wAqPzyi6TOkw
Our PLAYLISTS:
- Join our DAX Fridays! Series: goo.gl/FtUWUX
- Power BI dashboards for beginners: goo.gl/9YzyDP
- Power BI Tips & Tricks: goo.gl/H6kUbP
- Power Bi and Google Analytics: goo.gl/ZNsY8l
☼☼☼☼☼☼☼☼☼☼
POWER BI COURSES:
Want to learn Power BI? How about you take one of our courses? Here you can find the available courses:
curbal.com/courses-overview
☼☼☼☼☼☼☼☼☼☼
ABOUT CURBAL:
Website: www.curbal.com
Contact us: www.curbal.com/contact
************
************
QUESTIONS? COMMENTS? SUGGESTIONS? You’ll find me here:
Linkedin ► goo.gl/3VW6Ky
Twitter ► @curbalen, @ruthpozuelo
Facebook ► goo.gl/bME2sB
#CURBAL #SUBSCRIBE

Опубликовано:

4 мар 2020

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 34

@colacan92 3 года назад

Thanks Ruth, always getting big help from your videos. Hope I can contribute a bit this time. The key influencer is nice tool for customer scoring and/or segmentation I believe. Even for the traditional modelling method in real business, an analyst doesn't need to pay attention too much on how the model has been derived statistically (To do that you might need to learn very high level statistics. And mostly it is not the interest of audience/customer anyway) but they pay attention on how accurate the model is. And here the "Design of experiment" comes. First you need to build a model with certain% of total population (70%~80% if you have enough size, reduce if not) then test the model (eg. frequency table) with the rest. In case the model is for some kind of prediction, select data from earlier period. for example, if you have 2015~2019 (5 years) data, make a model with 2015~18 data then test the model with 2019 data and if the test result is good, this is how you convince customer. I see many people below disappointed with the result of it, the main reason would be the dataset they are using. For scoring or modelling, we don't hardly use raw data as it is but have to design new & dedicated table(s) with lots of potentially meaningful (computed) factors in it. For a complex model, certain level of knowledge on statistics is strongly recommended to avoid misleading

@CurbalEN 3 года назад

How interesting, thanks for sharing! Would love to hear more, let me know if you have time here: curbal.com/contact

@colacan92 3 года назад

@@CurbalEN I am honored being fixed on the top. BTW, seeing the gentlemen's feed back below saying your video can be 4 times shorter, I have to say that I am big fan of you and style of your video which I can enjoy with a cup of coffee relaxed. It makes me feel like having a comfort conversation. the length of video is only 10 min anyway... please keep your style. thanks x

@CurbalEN 3 года назад

I cant do it any other way, so don't worry, the style is here to stay ;)

@jpvdw365 3 года назад

Well said agree 100%. I use it as a guide carefully

@sphnyhptrs 4 года назад

I totally agree with your position of not trusting blindly as I believe any report should not be trusted blindly. We definitely need to follow the lead as you suggested and see where it takes us!! Great video!!

@romnyyepez 3 года назад

So true! In stationary multivariate time series, we ought to try many tests. For instance: cross-correlations in order to check who leads who, impulse-response-functions to get the sign, intensity, and persistence of the effect, (not to mention the stability of the model but a crucial step), Granger Causality Test, Forecast Error Variance Decomposition, and finally a set of forecastings with exogenous variables, at least....

@RobertoStaltari 4 года назад

This video could be the first of a series over ML and Power BI. ML is not a magic wand and must be used with care and knowledge, otherwise results can lead to disaster. Thanks Ruth!

@CurbalEN 4 года назад

ML is tough for me because I dont have that interest and get bored trying to learn it, but they are getting more and more intrusive and very soon unavoidable. Using AI responsible will be a must or wrong decisions will be taken along the way. /Ruth

@feng3625 3 года назад

Good point! Thank you. I think common sense is the first!

@CurbalEN 3 года назад

Think so too!

@mikar636 7 месяцев назад

Hi Curbal nice video as always! One question I have: Key influencer visualization not working in embed URL view. Is there any walkaround for this issue?

@andresfelipehiguera785 4 года назад

Trust but verify. How would we go about checking the results? It would be nice if you make a video about it.

@CurbalEN 4 года назад

Absolutely, added to my list! /Ruth

@euvieeugenio 3 года назад

You are correct. Not that just with this key influencer thing, but in Data Science/Machine Learning generally. I use lots of “ML” algos (just a sexy term of computational statistics) but I am always suspicious and careful. As a statistician, I always follow the adequacy tests to verify your data is appropriate for a model or method. And often times, these ML/data scientist aren’t aware of these. Someone should have a very deep understanding of mathematics and statistics to make it correctly 🙂

@CurbalEN 3 года назад

Agree!

@rayng4336 3 года назад

I was struggling with this issue. My chart analyzes the SLA status (above or under) by Name. Some of the result does not make sense. Like supervisors and managers were appearing as the worst performers in the organization. If you think about it, the majority of their time is spent managing the employees and customers. They may occasionally help when there are fewer employees than expected. So what I did was to filter them out.

@CurbalEN 3 года назад

Test always your results, specially with the AI visuals and you should be ok.

@gurbachansethi9548 4 года назад

Hi, I've used Key Influencers multiple times but each time experience hasn't been satisfying. At times the nos. reflect as agreggated values, for which one may think, why would I need KI for it. Then at many occasions the top factors at times just don't sink in. Satisfactory explanation etc. is missing, values in histogram chart at right side don't make sense.. There are many such issues... whi h makes it difficult to blindly believe output of KI.

@CurbalEN 4 года назад

Thanks for sharing your experiences, they will help others! /Ruth

@Phoenixspin 4 года назад

Trust Ruth, test everything else.

@CurbalEN 4 года назад

😂 i should make a t-shirt out of it! /Ruth

@EricaDyson 4 года назад

Well when I used the key influencers on a couple of projects, I thought the results (from a business point of view), were quite frankly, rubbish. Simply couldn't be used .. well maybe there were a couple of outputs that were vaguely acceptable.. but I couldn't have said they were key influencers. I've no doubt that the results were accurate but they were of no business value!

@CurbalEN 4 года назад

Oh, I see , thanks for sharing your experiences! /Ruth

@Same7WZ 4 года назад

Well don't get me wrong.. I like your channel and I'm actually subscribed to it.. but the duration of this video vs the amount of info presented is far from proportional.. you could have made a video like 4 times shorter to present these ideas.. many stops, thinking, and pauses of uncertainty.. and though it feels genuine, I'd have still loved to come up with a lot more in my 10 min commute to work.. I'd say do better preparation to improve the length-to-info ratio.. Again don't get me wrong, I like the channel :)

@CurbalEN 4 года назад

Thanks Sameh really appreciate the feedback, especially like this being so detailed. Just to let you know, I am working on it :) Have a nice day and thanks again!

@Same7WZ 4 года назад

@@CurbalEN no worries.. keep going (Nice nail polish by the way 🙂)

@CurbalEN 4 года назад

I like it too ;)

@peterbogdanovich4043 3 года назад

Today on Feb 9th of 2021 I was to raise the same point, but then I thought I don't want to sound arrogant or mean. But I can't help it to agree with Sameh W. that materials on this channel seem like a random chat over a coffee or a drink. Don't give the feeling that perhaps the material was prepared before hand and then presented. It's not concise, to the point and seems lengthy. No hard feeling I hope :)

@CurbalEN 3 года назад

Not at all, one thing I have learned is that you can't please everyone, and that is fine. Don't despair though, there are a 1M power bi channels nowadays and I am sure you will find your true match ( if you havent already). Have a nice evening!