๐Ÿ‘ฉโ€๐Ÿซ Implementation Guidelines

๐Ÿ“š Common Terms

Term

Description

๐Ÿ‘ฉโ€๐ŸŽ“ Bayes Error

The lowest possible error rate for any classifier (The optimal error ๐Ÿค”)

๐Ÿ‘ฉโ€๐Ÿซ Human Level Error

The error rate that can be obtained by a human

๐Ÿ‘ฎโ€โ™€๏ธ Avoidable Bias โ€

The difference between Bayes error and human level error

I did my best, my project is still doing bad, what shall I do? ๐Ÿ˜ฅ

Well, in this stage we have a criteria, is your model doing worse than humans (Because humans are quite good at a lot of tasks ๐Ÿ‘ฉโ€๐ŸŽ“)? If yes, you can:

  • ๐Ÿ‘ฉโ€๐Ÿซ Get labeled data from humans

  • ๐Ÿ‘€ Gain insight from manual error analysis; (Why did a person get this right? ๐Ÿ™„)

  • ๐Ÿ”Ž Better analysis of bias / variance ๐Ÿ”

๐Ÿค” Note: knowing how well humans can do on a task can help us to understand better how much we should try to reduce bias and variance

๐Ÿง Is your model doing better than humans?

  • Processes are less clear ๐Ÿ˜ฅ

Suitable techniques will be added here

๐Ÿค“ Study case

Let's assume that we have these two situations:

โ€‹

Case1

Case2

Human Error

1%

7.5%

Training Error

8%

8%

Dev Error

10%

10%

Even though training and dev errors are same we will apply different tactics for better performance

  • In Case1, We have High Bias so we have to focus on bias reduction techniques ๐Ÿค”, in other words we have to reduce the difference between training and human errors the avoidable error

    • Better algorithm, better NN structure, ......

  • In Case2, We have High Variance so we have to focus on variance reduction techniques ๐Ÿ™„, in other words we have to reduce the difference between training and dev errors

    • Adding regularization, getting more data, ......

We call this procedure of analysis Error analysis ๐Ÿ•ต๏ธโ€

๐Ÿ‘€ Error Types Visualization

In computer vision issues, human-level-error โ‰ˆ bayes-error because humans are good in vision tasks

๐Ÿค— Problems that ML surpasses human level performance

  • Online advertising

  • Product recommendations

  • Logistics

  • Loan approvals

  • .....

โœจ My Detailed Notes on Bias / Variance and Related Procedures

๐Ÿคธโ€โ™€๏ธ It is recommended to

When we have a new project it is recommended to produce an initial model and then iterate over it until you get the best model, this is more practical than spending time building model theoretical and thinking about the best hyperparameter -which is almost impossible ๐Ÿ™„-

So, just don't overthink! (In both ML problems and life problems ๐Ÿค—๐Ÿ™†โ€)