Hi Lim, thank you a lot for your great feedback! We definitely will in the future. We have tutorials on how to use Facebook Robyn with R if you want to start diving in: th-cam.com/video/l1Q42qx8z9I/w-d-xo.html
@@cassandra4533 Even I would like to request for same. Please create a tutorial in R as well other than fb Robyn. I am curious to know how the Excel solver like hyperparameter tuning & optimization can be in done in R . Thanks !!
Fantastic series! Exactly what I was hoping to see! One question: your model, I think, assumes, that the marketing spend in FB/TV/Radio will have an immediate effect on sales. How do you model in cases where marketing spend may take longer before it impacts sales?
Hello Luna, we're really happy you liked our course! As this is an introductionary series we used a simply Adstock transformation which doesn't handle Lagged Effect. However in Cassandra as well as other production-level MMMs other functions such as Weibull PDF are used to handle such behaviour. You can find more information on this other playlist in our channel: th-cam.com/video/PU9zuR1axUA/w-d-xo.html&ab_channel=Cassandra
I think if we can learn how to visualize the result of this model from different angles will be great! It will be quite interesting if the chart can enable decision makers to instinctively learn how much they can save/extra earn/other benefits from this model. 😍
Hello. Thanks for this tutorial. Quick question on the transformed FB, TV and Radio data. In your example you found the single values of alpha and beta that maximized the r-squared. My question is can I run the regression with multiple values of alpha and beta for each media, for example can I have an additional 3 transformed features for media with alpha and beta of 0.2 and 0.3 and have 6 transformed features instead of 3, OR do I have to find the single values of alpha and beta that maximize the r-squared? I hope my question makes sense. Thanks.
Hey! You should always stick with 1 value per each media, otherwise it would just be like counting Facebook 3 times in your model which doesn't really make sense in the real world
You are right, this is due to excel limitation, as I said in the previous comment. There are workarounds on solving this problem on excel, but the best solution long term is to switch to python or R programming language. They give you more elasticity while modeling MMM
Ehi, thank you for commenting everytime you have questions, I really appreciate it. Negative values for the intercept are usually cause by the OLS linear regression function. It basically approximates statistically how many sales we would get if all our inputs were equal to 0 This is technically a limitation of the function linear regression on excel, to solve this problem you should use an evolutionary methodology with excel solver (I can make a tutorial if you need me to) Or switching to R and Python programming language you are gonna have better elasticity when modeling your MMM
Hey, i im intrested in you to show how to use bayesan regression within MMM in python. Bayesan, cause ad market changes too fast and it's model should update coefs with new portion of data. And in python cause it could be automated for my job. 🙂
I have taken a look at r tutorials. In those, not only paid but also organic activity like newspapers and external factors like event from Robyn's sample data are considered. How are those columns included in the excel file? Is it easy or difficult?
In my Excel, formulas in cells below L6 are greyed out. So, when I run the solver to optimise alpha and beta, Excel says that the operation cannot be performed because L8 (objective) does not contain a formula. Do you know how to get around this please?
![NONT TANONT - คลั่งเธอ (Dopamine) [Official MV]](http://i.ytimg.com/vi/imr2dwueUxs/mqdefault.jpg)
this series was great, super easy to follow and concise enough, thank you! can you share more about MMM using R?
Hi Lim, thank you a lot for your great feedback!
We definitely will in the future.
We have tutorials on how to use Facebook Robyn with R if you want to start diving in: th-cam.com/video/l1Q42qx8z9I/w-d-xo.html
@@cassandra4533 Even I would like to request for same. Please create a tutorial in R as well other than fb Robyn. I am curious to know how the Excel solver like hyperparameter tuning & optimization can be in done in R . Thanks !!
Well explained !! Thank you
Fantastic series! Exactly what I was hoping to see! One question: your model, I think, assumes, that the marketing spend in FB/TV/Radio will have an immediate effect on sales. How do you model in cases where marketing spend may take longer before it impacts sales?
Hello Luna, we're really happy you liked our course!
As this is an introductionary series we used a simply Adstock transformation which doesn't handle Lagged Effect.
However in Cassandra as well as other production-level MMMs other functions such as Weibull PDF are used to handle such behaviour.
You can find more information on this other playlist in our channel: th-cam.com/video/PU9zuR1axUA/w-d-xo.html&ab_channel=Cassandra
The content rlly helps a lot!
Happy to hear that!
Let me know if there is something else I can do/create for you :)
I think if we can learn how to visualize the result of this model from different angles will be great!
It will be quite interesting if the chart can enable decision makers to instinctively learn how much they can save/extra earn/other benefits from this model. 😍
That was a really nice video. Can you also create playlist of marketing mix modeling using python?
Hey Shubham, glad you liked it! We are planning to but it's a quite busy time for us so it might take a while :)
Hello. Thanks for this tutorial. Quick question on the transformed FB, TV and Radio data. In your example you found the single values of alpha and beta that maximized the r-squared. My question is can I run the regression with multiple values of alpha and beta for each media, for example can I have an additional 3 transformed features for media with alpha and beta of 0.2 and 0.3 and have 6 transformed features instead of 3, OR do I have to find the single values of alpha and beta that maximize the r-squared? I hope my question makes sense. Thanks.
Hey! You should always stick with 1 value per each media, otherwise it would just be like counting Facebook 3 times in your model which doesn't really make sense in the real world
Appreciate your great work! Possible to have one episode going through how MMM to be derived by python?
Karena, we will soon 👌
When the slope is
You are right, this is due to excel limitation, as I said in the previous comment.
There are workarounds on solving this problem on excel, but the best solution long term is to switch to python or R programming language.
They give you more elasticity while modeling MMM
Hi it’s me again🤣
Would love to know why is that when there are many variables, the intercept (baseline) will become minus.
Ehi, thank you for commenting everytime you have questions, I really appreciate it.
Negative values for the intercept are usually cause by the OLS linear regression function.
It basically approximates statistically how many sales we would get if all our inputs were equal to 0
This is technically a limitation of the function linear regression on excel, to solve this problem you should use an evolutionary methodology with excel solver (I can make a tutorial if you need me to)
Or switching to R and Python programming language you are gonna have better elasticity when modeling your MMM
Hey, i im intrested in you to show how to use bayesan regression within MMM in python. Bayesan, cause ad market changes too fast and it's model should update coefs with new portion of data. And in python cause it could be automated for my job. 🙂
I suggest you to use Orbit, a time varying coefficient mmm library.
This is a really good article that describes it : getrecast.com/uber-orbit/
I have taken a look at r tutorials. In those, not only paid but also organic activity like newspapers and external factors like event from Robyn's sample data are considered. How are those columns included in the excel file? Is it easy or difficult?
You model them in the exact way of the media columns.
Just use sessions or impressions, as KPIs and do a linear regression
Keep in mind tho that organic variables can't be transformed into adstock and diminishing returns
Got it. Thanks!
In my Excel, formulas in cells below L6 are greyed out. So, when I run the solver to optimise alpha and beta, Excel says that the operation cannot be performed because L8 (objective) does not contain a formula. Do you know how to get around this please?
Facing same issue
I have the same problem, have you already solved it?
I failed to do the auto calculation,
It says the column that shows r^2 isn’t a formula, so it can’t calculate the optimized alpha and beta
Please kindly help😭 Thank you so much!
Can you send the Google sheet doc?
Problem solved! Thanks a lot!
@@user-og2qp6in8i I am getting the same problem, can you tell me how did you solve it?