Choosing the right machine learning algorithm for the problem can be hard.
This thread is about 3 questions that can help in the model selection process.
A thread🧵👇
This thread is about 3 questions that can help in the model selection process.
A thread🧵👇
1. Will the predictions have to be fully explainable?
Most models are not explainable. If you're looking for explainability, neural nets may not help. Models like decision trees can.
More interesting facts about decision trees beyond explainability👇
Most models are not explainable. If you're looking for explainability, neural nets may not help. Models like decision trees can.
More interesting facts about decision trees beyond explainability👇
https://twitter.com/Jeande_d/status/1413790290763264003?s=20
2. How big is the dataset?
For big datasets, neural nets and other complex models like ensemble methods can be a good choice.
For big datasets, neural nets and other complex models like ensemble methods can be a good choice.
3. How long do you have to train the model?
Simple linear models usually train faster, whereas neural networks and ensemble methods can be slow.
Another similar consideration: A model that takes too long to train will likely take too long to make predictions.
Simple linear models usually train faster, whereas neural networks and ensemble methods can be slow.
Another similar consideration: A model that takes too long to train will likely take too long to make predictions.
Although the above questions can help, choosing the right ML algorithm is a no-free lunch scenario.
There is no algorithm that is guaranteed to solve a problem before trying & evaluate different algorithms.
There is no algorithm that is guaranteed to solve a problem before trying & evaluate different algorithms.
This is the end of the thread.
To summarize, choosing the right ML algorithm is a no-free lunch scenario.
A number of factors have to be considered.
Some of the factors we discussed are:
1. Explainability
2. The size of training data
3. Training/prediction speed
To summarize, choosing the right ML algorithm is a no-free lunch scenario.
A number of factors have to be considered.
Some of the factors we discussed are:
1. Explainability
2. The size of training data
3. Training/prediction speed
Thank you for reading.
I am regularly writing about machine learning. My goal is to keep providing you with actionable ideas and best practices.
You can support me by following @Jeande_d,
leaving a ♡, or sharing this post with your friends.
I am regularly writing about machine learning. My goal is to keep providing you with actionable ideas and best practices.
You can support me by following @Jeande_d,
leaving a ♡, or sharing this post with your friends.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
