A short thread that will help you understand how a useful machine learning model differs from a crappy one.
(In English, without any jargon or math.)
This is also the way I first understood what "overfitting" means.
↓ 1/7
(In English, without any jargon or math.)
This is also the way I first understood what "overfitting" means.
↓ 1/7
When building a machine learning model, our ultimate goal is to "generalize" the lessons it learns to unseen data.
When we train it, we expect it to extract what's relevant, discard what's not, and understand the data's underlying patterns.
↓ 2/7
When we train it, we expect it to extract what's relevant, discard what's not, and understand the data's underlying patterns.
↓ 2/7
Imagine we have an image, and we want to determine whether it's a picture of a person or not.
These are some features that could be relevant to the model:
• There're two eyes
• There's a nose
• There's a mouth
• There are ears
↓ 3/7
These are some features that could be relevant to the model:
• There're two eyes
• There's a nose
• There's a mouth
• There are ears
↓ 3/7
And here are some features that aren't relevant to determine whether the image shows a person or not:
• The color of the eyes
• Whether the mouth is open
• The color of the skin
• The length of the hair
↓ 4/7
• The color of the eyes
• Whether the mouth is open
• The color of the skin
• The length of the hair
↓ 4/7
A model that generalizes well will learn to recognize relevant features and discard anything that's not relevant.
Such a model will be very good at predicting data it hasn't seen before.
↓ 5/7
Such a model will be very good at predicting data it hasn't seen before.
↓ 5/7
A bad model memorizes patterns from the data it has seen. It focuses on irrelevant features.
For example, imagine a model that uses the color of the eyes as an important feature.
That would be a bad model. Only people with specific eyes will be considered "people."
↓ 6/7
For example, imagine a model that uses the color of the eyes as an important feature.
That would be a bad model. Only people with specific eyes will be considered "people."
↓ 6/7
When a model memorizes patterns instead of generalizing them, we say it is "overfitting."
This problem happens when we don't train the model with enough data or its architecture is not appropriate.
It's one of the main issues we face when building models.
7/7
This problem happens when we don't train the model with enough data or its architecture is not appropriate.
It's one of the main issues we face when building models.
7/7
We can do this shit together!
Follow me @svpino, and every week I’ll help you navigate this machine learning thing from doing it, failing at it, learning, and fixing it.
Follow me @svpino, and every week I’ll help you navigate this machine learning thing from doing it, failing at it, learning, and fixing it.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
