Chapter 2 · Machine learning basics
Loss: measuring mistakes
Loss is one number that says how wrong the model is; training tries to make it small.
Preparing loss: measuring mistakes
Chapter 2 · Machine learning basics
Loss is one number that says how wrong the model is; training tries to make it small.
The red bar from each dot to the line is that dot's mistake. Loss squeezes all the mistakes into one number.
For every example, subtract the model’s prediction from the true answer. That gap is the error.
Change slope and intercept. The red bars are the mistakes and the meter is the loss. Squared loss punishes one big miss much more than several small ones.
The line under the meter is your loss after each move. Training does exactly this search, but automatically.
What does a lower loss mean?
Why do we square the errors?
A model has loss 0 on its training data. Its predictions there are…