#GeneralizedAdditiveModels - Twitter Hashtag

almost 3 years ago

A Simple Guide for Generalized Additive Models (GAMs) 1/ 🧵Let's dive into the world of #GeneralizedAdditiveModels or #GAMs! These are flexible regression models that can capture non-linear relationships. Perfect for when life (and data) isn’t just a straight line! 📈➰ 2/ At its core, GAM is a generalization of the linear model. Instead of fitting a straight line (or plane), GAMs fit smooth curves to the data. Think of it as letting the data guide the shape of the relationship, rather than forcing it into a straight jacket!🕺 3/ Why use GAMs? 🤔 • Your scatter plot suggests a wavy pattern • Residual plots from linear models show patterns (they shouldn't!) • You have complex temporal or spatial data • You want flexibility without manually creating polynomial terms 4/ How do GAMs achieve this? 🧐 Through smoothing functions. These are mathematical constructs that allow for bends and twists in the relationship between predictors and the outcome. Splines are a common choice for these functions! 5/ One beauty of #GAMs is that they can handle multiple types of data distributions. Whether you're predicting a continuous variable, binary outcomes, or counts, there’s a GAM for that! It's like GLMs but with added flexibility. 🎯 6/ Now, while GAMs sound dreamy (and often they are!), there's a balance. More flexibility can sometimes lead to overfitting. You know, when your model is TOO tailored to your training data & performs poorly on new data. It's like memorizing answers for a test but failing the real exam. 📚✖️ 7/ Fortunately, GAMs have built-in penalties to control for overfitting. It's like having an internal check, making sure the model isn't getting too carried away with wiggles and bends! 🙌 8/ Another plus? Interpreting GAMs can be pretty intuitive. You get visual plots showing the effect of each predictor. Instead of squinting at coefficients, you can see the shape of the relationship directly! 📊 9/ In practice, tools like R's mgcv package make it pretty straightforward to fit GAMs. But, as with all models, understanding the underlying mechanics & assumptions can really up your GAM game! 🛠️ 10/ In summary: • Use GAMs for non-linear relationships • They're flexible but have checks to prevent overfitting • Visual interpretations are a boon! 11/ So next time your data doesn’t play nice with straight lines, consider giving #GeneralizedAdditiveModels a spin. They're a powerful tool in a statistician’s toolbox, ready to tackle those wavy, bendy data challenges! 🌀🔍 12/ Liked this thread? Found it useful? Feel free to like, share, and comment with your experiences or questions about GAMs! Let’s keep the #StatsChat going! 🗣️👥🎉 #Statistics #DataScience

selcukorkmaz's tweet photo. A Simple Guide for Generalized Additive Models (GAMs)

1/ 🧵Let's dive into the world of #GeneralizedAdditiveModels or #GAMs! These are flexible regression models that can capture non-linear relationships. Perfect for when life (and data) isn’t just a straight line! 📈➰

2/ At its core, GAM is a generalization of the linear model. Instead of fitting a straight line (or plane), GAMs fit smooth curves to the data. Think of it as letting the data guide the shape of the relationship, rather than forcing it into a straight jacket!🕺

3/ Why use GAMs? 🤔
• Your scatter plot suggests a wavy pattern
• Residual plots from linear models show patterns (they shouldn't!)
• You have complex temporal or spatial data
• You want flexibility without manually creating polynomial terms

4/ How do GAMs achieve this? 🧐 Through smoothing functions. These are mathematical constructs that allow for bends and twists in the relationship between predictors and the outcome. Splines are a common choice for these functions!

5/ One beauty of #GAMs is that they can handle multiple types of data distributions. Whether you're predicting a continuous variable, binary outcomes, or counts, there’s a GAM for that! It's like GLMs but with added flexibility. 🎯

6/ Now, while GAMs sound dreamy (and often they are!), there's a balance. More flexibility can sometimes lead to overfitting. You know, when your model is TOO tailored to your training data & performs poorly on new data. It's like memorizing answers for a test but failing the real exam. 📚✖️

7/ Fortunately, GAMs have built-in penalties to control for overfitting. It's like having an internal check, making sure the model isn't getting too carried away with wiggles and bends! 🙌

8/ Another plus? Interpreting GAMs can be pretty intuitive. You get visual plots showing the effect of each predictor. Instead of squinting at coefficients, you can see the shape of the relationship directly! 📊

9/ In practice, tools like R's mgcv package make it pretty straightforward to fit GAMs. But, as with all models, understanding the underlying mechanics & assumptions can really up your GAM game! 🛠️

10/ In summary:
• Use GAMs for non-linear relationships
• They're flexible but have checks to prevent overfitting
• Visual interpretations are a boon!

11/ So next time your data doesn’t play nice with straight lines, consider giving #GeneralizedAdditiveModels a spin. They're a powerful tool in a statistician’s toolbox, ready to tackle those wavy, bendy data challenges! 🌀🔍

12/ Liked this thread? Found it useful? Feel free to like, share, and comment with your experiences or questions about GAMs! Let’s keep the #StatsChat going! 🗣️👥🎉

#Statistics #DataScience

16

1K

244

1K

297K

Christel de Bruijn, PhD @christeldebruijn.bsky.soc @ChristelDBruijn

over 3 years ago

Came across a really nice, conceptual explanation of linear and non-linear regression splines by #MikkoRönkkö. It's only 8.5 minutes long, and well worth the time. #GeneralizedAdditiveModels #epitwitter https://t.co/3voPjeXSd3

1

12

3

14

0

Royal Statistical Society @RoyalStatSoc

almost 5 years ago

GAM can strike a balance between flexibility and interpretability, while being able to handle large data sets! Book a place on our course to find out more. #GeneralizedAdditiveModels #regressionmodels https://t.co/t6xgXFY6bt

RoyalStatSoc's tweet photo. GAM can strike a balance between flexibility and interpretability, while being able to handle large data sets! Book a place on our course to find out more. #GeneralizedAdditiveModels #regressionmodels
https://t.co/t6xgXFY6bt https://t.co/Ll3S8DB3k2

0

5

2

0

Royal Statistical Society @RoyalStatSoc

almost 5 years ago

GAM can strike a balance between flexibility and interpretability, while being able to handle large data sets! Book a place on our course to find out more. #GeneralizedAdditiveModels #regressionmodels https://t.co/t6xgXFY6bt

0

5

0

Rob Calver @RobCalver5

almost 5 years ago

Wow, Simon Wood's '#GeneralizedAdditiveModels: An Introduction with #RStats, Second Edition' is the 9th best regression book of all time! https://t.co/BQsrAnqEAh @CRC_MathStats #statistics #DataScience #GAMs

RobCalver5's tweet photo. Wow, Simon Wood's '#GeneralizedAdditiveModels: An Introduction with #RStats, Second Edition' is the 9th best regression book of all time! https://t.co/BQsrAnqEAh @CRC_MathStats #statistics #DataScience #GAMs https://t.co/Hms2E9c8SP

1

50

8

20

0

CrowdShakti @crowdshakti

almost 7 years ago

Generalized Additive Models in R. This short course will teach you how to use these flexible, powerful tools to model data and solve data science problems. https://t.co/FHVrembrqV #GeneralizedAdditiveModels #R #GAM #MLmodel #DataScience