Predicting the NBA MVP: Machine Learning Project [part 3 of 3]

[@louiepillari7796]

This series was great and extremely informative. Enjoyed getting a detailed explanation and seeing the whole process of scraping the data to actually using it for ML!

[@gerocastano8]

Hello Dataquest! I've been watching your videos for a while, and I have to say, the clean way you code things and how you explain them is so entertaining to watch! I don't know if you already solved this problem, but I noticed that when you defined the backtest function and you pass in the model as an argument, you forgot to change the "reg.predict" line to "model.predict". This would mean you would always predict the MVP's with the same Ridge model. Thank you so much for all of your videos!

[@tomkmb4120]

Man, I'm so glad that I've found this channel. I am only just starting my journey with Python but this is exactly the sort of thing I aspire to be doing once I am competent enough. Great videos! The ones on the EPL are next on my watch list!

[@Dataquestio]

Thanks, glad you like them! -Vik

[@tomkmb4120]

@@Dataquestio Do you have any recommendations for entry level reading into machine learning or data science for this sort of thing?

[@solomonbalogun7651]

Thanks so much for your content, never seen machine leaning projects like yours.

[@wittttttt]

6:29 Shaq be like: "Dude.. really?" :)

[@israelschuman6631]

I had that exact thought lol

[@nitingowda1921]

Shouldn't adding the PTS_r, AST_r, BLK_r, STL_r, 3P_r result in high correlation between the PTS, AST, BLK, STL and 3P_r column, making the model less accurate?

[@edselmobiondi2625]

Hello! How to find/finalize the winner for the next year's MVP?

[@quoit99training83]

When can we expect NLP and computer vision added to curriculum? Original it was mentioned by DQ that it will be end of 2021… thank you

[@Dataquestio]

Thanks for asking this question. We've needed to work on internal tooling to get to the point where we can consistently release new courses at our quality standards. We've done this work over the past few months. Starting this month, we'll be launching new courses monthly. We have 15 courses slated to launch over the next 4 months, and more later this year. Now that we're consistently creating new courses, we'll be adding more machine learning and machine learning engineering content later this year. I can't make any specific promises about NLP and CV content, but I'll check with our content team about releasing a public course roadmap so you can see our progress.

[@Dataquestio]

Our course roadmap is now here - portal.productboard.com/74zqsppdoyzwfijdislv6qn3/ . NLP and CV are currently under consideration for Q3 and Q4 this year.

[@charlieguthmann2531]

First, Thank you for this series. So so helpful to me. Also though, at 6:30 you talking about filling the 3P% with 0s. I agree that if you don't attempt a 2 or FT there is no way you are in the MVP race, but there might have been big men who were in the race who did not attempt a 3 (in reality they probably took a few). If this was the case would it be an issue?

[@Dataquestio]

Hi Charlie - this is a good point. Filling it with a 0 shouldn't make a huge difference to the ranking, but you could also try filling in the missing data a different way (using the league average value, filling in a player's historical average, etc). -Vik

[@siddarthpuliyanda4039]

when doing the stat ratios near the end the average precision score ended up lower than the original value. why did you decide to keep those variables in the model if it did not improve the strength of the model

[@MasterKwonDont]

Did it work? How successful has it been at predicting outcomes?

[@kojorichardson4283]

What part of the backtest function is using the parameter years[5:] that is being passed in? It's already in the function body.

[@Dataquestio]

Good point, kojo! If you only want to remove the first 5 rows, you only need `years[5:]` once. You can remove the `[5:]` either when you call the function, or remove it from the function body. If you call it twice (like in the video), it will remove the first 10 rows instead. There is a small mistake in the video, and it is removing 10 rows instead of 5. It doesn't invalidate any results if you do remove 10 rows instead of 5, but it does change how many years you skip before starting to backtest.

[@rahprabhu]

Is there a reason why the predictors weren't scaled before running the ridge regression? Considering we have predictors that are percentages and other predictors that have different scales (i.e. blocks vs points scored)

[@Dataquestio]

Hi there - that's a good point. If ridge regression is the final algorithm you're going to use, you should definitely scale the predictors. In this case, I wanted to get to predictions quickly, and ended up switching to a different model (random forest), so I didn't scale. - Vik

[@mauricegibson599]

Now that the ‘22 season is over how can I test for the next year?

[@Dataquestio]

Hi Maurice - this algorithm is designed to work after the season starts, since it pulls seasons stats. You could modify it to predict the MVP for next year given stats from this year. -Vik

[@marthanyarkoa9007]

how do I put all these codes in my head?

[@Dataquestio]

Hi Martha - it's all about practice! You don't need to memorize everything, but practicing with a couple of projects will help you remember the important commands.

[@anthonymalary7616]

I'm self teaching how long did it take you to get your XP @@Dataquestio

[@oliesting4921]

I have been following Dataquest for a while now and I have to say the updates to the DS and DE career path have been very slow. I am interested in DE, and there is nothing about Airflow, DE on cloud (such as aws), Big query or Snowflake to name a few. Your DE path is lacking concepts of crucial tools used every day by engineers. You cannot claim it is teaching fundamental concepts as stuff related to DE is only covered in last section (pipelines, mapreduce). You have to update your DE career path with relevant info/lectures. Till then, I would suggest ppl to stay away from your DE path.

[@Dataquestio]

Hi Olie - thanks for following Dataquest, and for sharing your thoughts on how we can improve. I appreciate it - we're always trying to get better. Data engineering is a broad field, and there's a lot to learn. Our focus at Dataquest is always to teach the key concepts that help you apply your skills in the real world and get hired. In the DE path, we teach the core concepts you need to master ANY new technology that arises. Concepts like time complexity, memory management, multithreading, pipelines, and others that we teach are key to this. Many learners have been hired as data engineers after taking our DE courses, including Gino (www.dataquest.io/blog/learner-spotlight-gino-parages/) and Veena (www.dataquest.io/blog/learner-spotlight-veena-sanjeeve/). That said, you're right that there is more we can add to improve this path. We've needed to work on internal tooling to get to the point where we can consistently release new courses at our quality standards. We've done this work over the past few months. Starting this month, we'll be launching new courses monthly. We have 15 courses slated to launch over the next 4 months. Now that we're consistently creating new courses, we'll be adding cloud and other content to the DE path this year. I'll check with our content team about releasing a public course roadmap so you can see our progress.

[@Dataquestio]

Hi Olie - you can see our roadmap here - portal.productboard.com/74zqsppdoyzwfijdislv6qn3/ .

[@vasoochigava5213]

thanks a lot for these series, but what was the summary who did you predict to be the next nba mvp ?

[@Dataquestio]

I did this project before the 2021-2022 season finished - so you'd need to re-run now to find the prediction for this season!

[@vasoochigava5213]

Outcomes were never pointed out even in this video. It was giannis mvp in real results and as i see you ended with nikola jokic being mvp

[@lancelot801]

I'd like to do the same but using R, is there an equivalent code for that?

[@aradbeneliezer7129]

Hey, have u managed to do that?

[@lancelot801]

@Arad Ben Eliezer Not exactly, I ended up using the same data but instead of MVP predictions I used it for classification (position based on the stats).

[@aradbeneliezer7129]

@@lancelot801 u mean predicting what position a player plays ?

[@lancelot801]

@Arad Ben Eliezer Exactly, selected a couple of stats like PPG, 3PA, did the data cleaning then used three different methods to classify the players into a position then compared it to their actual position.