Define logit

[rdong8]

I don't think logits are ever formally defined anywhere in the book, they're just suddenly mentioned in 4.1 Softmax Regression.

[muzammil5539]

I don't think logits are ever formally defined anywhere in the book, they're just suddenly mentioned in 4.1 Softmax Regression.

Logits are like the "raw scores" a neuron calculates before any activation function is applied.

Here's a simple example:

Logits = (weights × inputs) + bias

These logits are then passed through an activation function to get the final output.

Example logits calculation

logits = (weights * inputs) + bias

Apply activation function (e.g., sigmoid for binary classification)

output = activation(logits)